Merge pull request #903 from mitchellh/macos-update

Custom Shaders (Metal and OpenGL)

.prettierignore

@@ -10,3 +10,6 @@ macos/
 
 # website dev run
 website/.next
+
+# shaders
+*.frag

build.zig

@@ -643,6 +643,14 @@ fn addDeps(
         .optimize = step.optimize,
         .@"enable-libpng" = true,
     });
+    const glslang_dep = b.dependency("glslang", .{
+        .target = step.target,
+        .optimize = step.optimize,
+    });
+    const spirv_cross_dep = b.dependency("spirv_cross", .{
+        .target = step.target,
+        .optimize = step.optimize,
+    });
     const mach_glfw_dep = b.dependency("mach_glfw", .{
         .target = step.target,
         .optimize = step.optimize,
@@ -655,6 +663,7 @@ fn addDeps(
         .target = step.target,
         .optimize = step.optimize,
     });
+    const opengl_dep = b.dependency("opengl", .{});
     const pixman_dep = b.dependency("pixman", .{
         .target = step.target,
         .optimize = step.optimize,
@@ -718,8 +727,11 @@ fn addDeps(
         fontconfig_dep.module("fontconfig"),
     );
     step.addModule("freetype", freetype_dep.module("freetype"));
+    step.addModule("glslang", glslang_dep.module("glslang"));
+    step.addModule("spirv_cross", spirv_cross_dep.module("spirv_cross"));
     step.addModule("harfbuzz", harfbuzz_dep.module("harfbuzz"));
     step.addModule("xev", libxev_dep.module("xev"));
+    step.addModule("opengl", opengl_dep.module("opengl"));
     step.addModule("pixman", pixman_dep.module("pixman"));
     step.addModule("ziglyph", ziglyph_dep.module("ziglyph"));
 
@@ -743,6 +755,14 @@ fn addDeps(
         try static_libs.append(tracy_dep.artifact("tracy").getEmittedBin());
     }
 
+    // Glslang
+    step.linkLibrary(glslang_dep.artifact("glslang"));
+    try static_libs.append(glslang_dep.artifact("glslang").getEmittedBin());
+
+    // Spirv-Cross
+    step.linkLibrary(spirv_cross_dep.artifact("spirv_cross"));
+    try static_libs.append(spirv_cross_dep.artifact("spirv_cross").getEmittedBin());
+
     // Dynamic link
     if (!static) {
         step.addIncludePath(freetype_dep.path(""));

build.zig.zon

@@ -13,8 +13,8 @@
             .hash = "12202da6b8e9024c653f5d67f55a8065b401c42b3c08b69333d95400fe85d6019a59",
         },
         .zig_objc = .{
-            .url = "https://github.com/mitchellh/zig-objc/archive/146a50bb018d8e1ac5b9a1454d9db9a5eba5361f.tar.gz",
-            .hash = "12209f62dae4fccae478f5bd5670725c55308d8d985506110ba122ee2fb5e73122e0",
+            .url = "https://github.com/mitchellh/zig-objc/archive/a38331cb6ee366b3f22d0068297810ef14c0c400.tar.gz",
+            .hash = "1220dcb34ec79a9b02c46372a41a446212f2366e7c69c8eba68e88f0f25b5ddf475d",
         },
         .zig_js = .{
             .url = "https://github.com/mitchellh/zig-js/archive/60ac42ab137461cdba2b38cc6c5e16376470aae6.tar.gz",
@@ -32,10 +32,15 @@
         .harfbuzz = .{ .path = "./pkg/harfbuzz" },
         .libpng = .{ .path = "./pkg/libpng" },
         .macos = .{ .path = "./pkg/macos" },
+        .opengl = .{ .path = "./pkg/opengl" },
         .pixman = .{ .path = "./pkg/pixman" },
         .tracy = .{ .path = "./pkg/tracy" },
         .zlib = .{ .path = "./pkg/zlib" },
 
+        // Shader translation
+        .glslang = .{ .path = "./pkg/glslang" },
+        .spirv_cross = .{ .path = "./pkg/spirv-cross" },
+
         // System headers
         .apple_sdk = .{ .path = "./pkg/apple-sdk" },
     },

pkg/glslang/build.zig

@@ -0,0 +1,137 @@
+const std = @import("std");
+
+pub fn build(b: *std.Build) !void {
+    const target = b.standardTargetOptions(.{});
+    const optimize = b.standardOptimizeOption(.{});
+
+    _ = b.addModule("glslang", .{ .source_file = .{ .path = "main.zig" } });
+
+    const upstream = b.dependency("glslang", .{});
+    const lib = try buildGlslang(b, upstream, target, optimize);
+    b.installArtifact(lib);
+
+    {
+        const test_exe = b.addTest(.{
+            .name = "test",
+            .root_source_file = .{ .path = "main.zig" },
+            .target = target,
+            .optimize = optimize,
+        });
+        test_exe.linkLibrary(lib);
+        const tests_run = b.addRunArtifact(test_exe);
+        const test_step = b.step("test", "Run tests");
+        test_step.dependOn(&tests_run.step);
+
+        // Uncomment this if we're debugging tests
+        // b.installArtifact(test_exe);
+    }
+}
+
+fn buildGlslang(
+    b: *std.Build,
+    upstream: *std.Build.Dependency,
+    target: std.zig.CrossTarget,
+    optimize: std.builtin.OptimizeMode,
+) !*std.Build.Step.Compile {
+    const lib = b.addStaticLibrary(.{
+        .name = "glslang",
+        .target = target,
+        .optimize = optimize,
+    });
+    lib.linkLibC();
+    lib.linkLibCpp();
+    lib.addIncludePath(upstream.path(""));
+    lib.addIncludePath(.{ .path = "override" });
+
+    var flags = std.ArrayList([]const u8).init(b.allocator);
+    defer flags.deinit();
+    try flags.appendSlice(&.{
+        "-fno-sanitize=undefined",
+        "-fno-sanitize-trap=undefined",
+    });
+
+    lib.addCSourceFiles(.{
+        .dependency = upstream,
+        .flags = flags.items,
+        .files = &.{
+            // GenericCodeGen
+            "glslang/GenericCodeGen/CodeGen.cpp",
+            "glslang/GenericCodeGen/Link.cpp",
+
+            // MachineIndependent
+            //"MachineIndependent/glslang.y",
+            "glslang/MachineIndependent/glslang_tab.cpp",
+            "glslang/MachineIndependent/attribute.cpp",
+            "glslang/MachineIndependent/Constant.cpp",
+            "glslang/MachineIndependent/iomapper.cpp",
+            "glslang/MachineIndependent/InfoSink.cpp",
+            "glslang/MachineIndependent/Initialize.cpp",
+            "glslang/MachineIndependent/IntermTraverse.cpp",
+            "glslang/MachineIndependent/Intermediate.cpp",
+            "glslang/MachineIndependent/ParseContextBase.cpp",
+            "glslang/MachineIndependent/ParseHelper.cpp",
+            "glslang/MachineIndependent/PoolAlloc.cpp",
+            "glslang/MachineIndependent/RemoveTree.cpp",
+            "glslang/MachineIndependent/Scan.cpp",
+            "glslang/MachineIndependent/ShaderLang.cpp",
+            "glslang/MachineIndependent/SpirvIntrinsics.cpp",
+            "glslang/MachineIndependent/SymbolTable.cpp",
+            "glslang/MachineIndependent/Versions.cpp",
+            "glslang/MachineIndependent/intermOut.cpp",
+            "glslang/MachineIndependent/limits.cpp",
+            "glslang/MachineIndependent/linkValidate.cpp",
+            "glslang/MachineIndependent/parseConst.cpp",
+            "glslang/MachineIndependent/reflection.cpp",
+            "glslang/MachineIndependent/preprocessor/Pp.cpp",
+            "glslang/MachineIndependent/preprocessor/PpAtom.cpp",
+            "glslang/MachineIndependent/preprocessor/PpContext.cpp",
+            "glslang/MachineIndependent/preprocessor/PpScanner.cpp",
+            "glslang/MachineIndependent/preprocessor/PpTokens.cpp",
+            "glslang/MachineIndependent/propagateNoContraction.cpp",
+
+            // C Interface
+            "glslang/CInterface/glslang_c_interface.cpp",
+
+            // ResourceLimits
+            "glslang/ResourceLimits/ResourceLimits.cpp",
+            "glslang/ResourceLimits/resource_limits_c.cpp",
+
+            // SPIRV
+            "SPIRV/GlslangToSpv.cpp",
+            "SPIRV/InReadableOrder.cpp",
+            "SPIRV/Logger.cpp",
+            "SPIRV/SpvBuilder.cpp",
+            "SPIRV/SpvPostProcess.cpp",
+            "SPIRV/doc.cpp",
+            "SPIRV/disassemble.cpp",
+            "SPIRV/CInterface/spirv_c_interface.cpp",
+        },
+    });
+
+    if (!target.isWindows()) {
+        lib.addCSourceFiles(.{
+            .dependency = upstream,
+            .flags = flags.items,
+            .files = &.{
+                "glslang/OSDependent/Unix/ossource.cpp",
+            },
+        });
+    } else {
+        lib.addCSourceFiles(.{
+            .dependency = upstream,
+            .flags = flags.items,
+            .files = &.{
+                "glslang/OSDependent/Windows/ossource.cpp",
+            },
+        });
+    }
+
+    lib.installHeadersDirectoryOptions(.{
+        .source_dir = upstream.path(""),
+        .install_dir = .header,
+        .install_subdir = "",
+        .include_extensions = &.{".h"},
+    });
+
+    return lib;
+}

pkg/glslang/build.zig.zon

@@ -0,0 +1,11 @@
+.{
+    .name = "glslang",
+    .version = "13.1.1",
+    .paths = .{""},
+    .dependencies = .{
+        .glslang = .{
+            .url = "https://github.com/KhronosGroup/glslang/archive/refs/tags/13.1.1.tar.gz",
+            .hash = "1220481fe19def1172cd0728743019c0f440181a6342b62d03e24d05c70141516799",
+        },
+    },
+}

pkg/glslang/c.zig

@@ -0,0 +1,4 @@
+pub usingnamespace @cImport({
+    @cInclude("glslang/Include/glslang_c_interface.h");
+    @cInclude("glslang/Public/resource_limits_c.h");
+});

pkg/glslang/init.zig

@@ -0,0 +1,9 @@
+const c = @import("c.zig");
+
+pub fn init() !void {
+    if (c.glslang_initialize_process() == 0) return error.GlslangInitFailed;
+}
+
+pub fn finalize() void {
+    c.glslang_finalize_process();
+}

pkg/glslang/main.zig

@@ -0,0 +1,9 @@
+pub const c = @import("c.zig");
+pub const testing = @import("test.zig");
+pub usingnamespace @import("init.zig");
+pub usingnamespace @import("program.zig");
+pub usingnamespace @import("shader.zig");
+
+test {
+    @import("std").testing.refAllDecls(@This());
+}

pkg/glslang/override/glslang/build_info.h

@@ -0,0 +1,62 @@
+// Copyright (C) 2020 The Khronos Group Inc.
+//
+// All rights reserved.
+//
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions
+// are met:
+//
+//    Redistributions of source code must retain the above copyright
+//    notice, this list of conditions and the following disclaimer.
+//
+//    Redistributions in binary form must reproduce the above
+//    copyright notice, this list of conditions and the following
+//    disclaimer in the documentation and/or other materials provided
+//    with the distribution.
+//
+//    Neither the name of The Khronos Group Inc. nor the names of its
+//    contributors may be used to endorse or promote products derived
+//    from this software without specific prior written permission.
+//
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+// FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+// COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+// INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+// BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+// LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+// CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+// LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
+// ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+// POSSIBILITY OF SUCH DAMAGE.
+
+#ifndef GLSLANG_BUILD_INFO
+#define GLSLANG_BUILD_INFO
+
+#define GLSLANG_VERSION_MAJOR 13
+#define GLSLANG_VERSION_MINOR 1
+#define GLSLANG_VERSION_PATCH 1
+#define GLSLANG_VERSION_FLAVOR ""
+
+#define GLSLANG_VERSION_GREATER_THAN(major, minor, patch) \
+    ((GLSLANG_VERSION_MAJOR) > (major) || ((major) == GLSLANG_VERSION_MAJOR && \
+    ((GLSLANG_VERSION_MINOR) > (minor) || ((minor) == GLSLANG_VERSION_MINOR && \
+     (GLSLANG_VERSION_PATCH) > (patch)))))
+
+#define GLSLANG_VERSION_GREATER_OR_EQUAL_TO(major, minor, patch) \
+    ((GLSLANG_VERSION_MAJOR) > (major) || ((major) == GLSLANG_VERSION_MAJOR && \
+    ((GLSLANG_VERSION_MINOR) > (minor) || ((minor) == GLSLANG_VERSION_MINOR && \
+     (GLSLANG_VERSION_PATCH >= (patch))))))
+
+#define GLSLANG_VERSION_LESS_THAN(major, minor, patch) \
+    ((GLSLANG_VERSION_MAJOR) < (major) || ((major) == GLSLANG_VERSION_MAJOR && \
+    ((GLSLANG_VERSION_MINOR) < (minor) || ((minor) == GLSLANG_VERSION_MINOR && \
+     (GLSLANG_VERSION_PATCH) < (patch)))))
+
+#define GLSLANG_VERSION_LESS_OR_EQUAL_TO(major, minor, patch) \
+    ((GLSLANG_VERSION_MAJOR) < (major) || ((major) == GLSLANG_VERSION_MAJOR && \
+    ((GLSLANG_VERSION_MINOR) < (minor) || ((minor) == GLSLANG_VERSION_MINOR && \
+     (GLSLANG_VERSION_PATCH <= (patch))))))
+
+#endif // GLSLANG_BUILD_INFO

pkg/glslang/program.zig

@@ -0,0 +1,60 @@
+const std = @import("std");
+const c = @import("c.zig");
+const testlib = @import("test.zig");
+const Shader = @import("shader.zig").Shader;
+
+pub const Program = opaque {
+    pub fn create() !*Program {
+        if (c.glslang_program_create()) |ptr| return @ptrCast(ptr);
+        return error.OutOfMemory;
+    }
+
+    pub fn delete(self: *Program) void {
+        c.glslang_program_delete(@ptrCast(self));
+    }
+
+    pub fn addShader(self: *Program, shader: *Shader) void {
+        c.glslang_program_add_shader(@ptrCast(self), @ptrCast(shader));
+    }
+
+    pub fn link(self: *Program, messages: c_int) !void {
+        if (c.glslang_program_link(@ptrCast(self), messages) != 0) return;
+        return error.GlslangFailed;
+    }
+
+    pub fn spirvGenerate(self: *Program, stage: c.glslang_stage_t) void {
+        c.glslang_program_SPIRV_generate(@ptrCast(self), stage);
+    }
+
+    pub fn spirvGetSize(self: *Program) usize {
+        return @intCast(c.glslang_program_SPIRV_get_size(@ptrCast(self)));
+    }
+
+    pub fn spirvGet(self: *Program, buf: []u32) void {
+        c.glslang_program_SPIRV_get(@ptrCast(self), buf.ptr);
+    }
+
+    pub fn spirvGetPtr(self: *Program) ![*]u32 {
+        return @ptrCast(c.glslang_program_SPIRV_get_ptr(@ptrCast(self)));
+    }
+
+    pub fn spirvGetMessages(self: *Program) ![:0]const u8 {
+        const ptr = c.glslang_program_SPIRV_get_messages(@ptrCast(self));
+        return std.mem.sliceTo(ptr, 0);
+    }
+
+    pub fn getInfoLog(self: *Program) ![:0]const u8 {
+        const ptr = c.glslang_program_get_info_log(@ptrCast(self));
+        return std.mem.sliceTo(ptr, 0);
+    }
+
+    pub fn getDebugInfoLog(self: *Program) ![:0]const u8 {
+        const ptr = c.glslang_program_get_info_debug_log(@ptrCast(self));
+        return std.mem.sliceTo(ptr, 0);
+    }
+};
+
+test {
+    var program = try Program.create();
+    defer program.delete();
+}

pkg/glslang/shader.zig

@@ -0,0 +1,58 @@
+const std = @import("std");
+const c = @import("c.zig");
+const testlib = @import("test.zig");
+
+pub const Shader = opaque {
+    pub fn create(input: *const c.glslang_input_t) !*Shader {
+        if (c.glslang_shader_create(input)) |ptr| return @ptrCast(ptr);
+        return error.OutOfMemory;
+    }
+
+    pub fn delete(self: *Shader) void {
+        c.glslang_shader_delete(@ptrCast(self));
+    }
+
+    pub fn preprocess(self: *Shader, input: *const c.glslang_input_t) !void {
+        if (c.glslang_shader_preprocess(@ptrCast(self), input) == 0)
+            return error.GlslangFailed;
+    }
+
+    pub fn parse(self: *Shader, input: *const c.glslang_input_t) !void {
+        if (c.glslang_shader_parse(@ptrCast(self), input) == 0)
+            return error.GlslangFailed;
+    }
+
+    pub fn getInfoLog(self: *Shader) ![:0]const u8 {
+        const ptr = c.glslang_shader_get_info_log(@ptrCast(self));
+        return std.mem.sliceTo(ptr, 0);
+    }
+
+    pub fn getDebugInfoLog(self: *Shader) ![:0]const u8 {
+        const ptr = c.glslang_shader_get_info_debug_log(@ptrCast(self));
+        return std.mem.sliceTo(ptr, 0);
+    }
+};
+
+test {
+    const input: c.glslang_input_t = .{
+        .language = c.GLSLANG_SOURCE_GLSL,
+        .stage = c.GLSLANG_STAGE_FRAGMENT,
+        .client = c.GLSLANG_CLIENT_VULKAN,
+        .client_version = c.GLSLANG_TARGET_VULKAN_1_2,
+        .target_language = c.GLSLANG_TARGET_SPV,
+        .target_language_version = c.GLSLANG_TARGET_SPV_1_5,
+        .code = @embedFile("test/simple.frag"),
+        .default_version = 100,
+        .default_profile = c.GLSLANG_NO_PROFILE,
+        .force_default_version_and_profile = 0,
+        .forward_compatible = 0,
+        .messages = c.GLSLANG_MSG_DEFAULT_BIT,
+        .resource = c.glslang_default_resource(),
+    };
+
+    try testlib.ensureInit();
+    const shader = try Shader.create(&input);
+    defer shader.delete();
+    try shader.preprocess(&input);
+    try shader.parse(&input);
+}

pkg/glslang/test.zig

@@ -0,0 +1,10 @@
+const glslang = @import("main.zig");
+
+var initialized: bool = false;
+
+/// Call this function before any other tests in this package to ensure that
+/// the glslang library is initialized.
+pub fn ensureInit() !void {
+    if (initialized) return;
+    try glslang.init();
+}

pkg/glslang/test/simple.frag

@@ -0,0 +1,56 @@
+#version 430 core
+
+layout(binding = 0) uniform Globals {
+    uniform vec3	iResolution;
+    uniform float	iTime;
+    uniform float	iTimeDelta;
+    uniform float	iFrameRate;
+    uniform int	    iFrame;
+    uniform float	iChannelTime[4];
+    uniform vec3	iChannelResolution[4];
+    uniform vec4	iMouse;
+    uniform vec4	iDate;
+    uniform float	iSampleRate;
+};
+
+layout(binding = 0) uniform sampler2D	iChannel0;
+layout(binding = 1) uniform sampler2D	iChannel1;
+layout(binding = 2) uniform sampler2D	iChannel2;
+layout(binding = 3) uniform sampler2D	iChannel3;
+
+layout(location = 0) in vec4 gl_FragCoord;
+layout(location = 0) out vec4 _fragColor;
+
+#define texture2D texture
+
+void mainImage( out vec4 fragColor, in vec2 fragCoord );
+void main() { mainImage (_fragColor, gl_FragCoord.xy); }
+
+#define t iTime
+
+void mainImage( out vec4 fragColor, in vec2 fragCoord )
+{
+    // Normalized pixel coordinates (from 0 to 1)
+    vec2 uv = ( fragCoord - .5*iResolution.xy) / iResolution.y;
+    vec3 col = vec3(0.);
+    float a = atan(uv.y,uv.x);
+    float r = 0.5*length(uv);
+    float counter = 100.;
+    a = 4.*a+20.*r+50.*cos(r)*cos(.1*t)+abs(a*r);
+    float f = 0.02*abs(cos(a))/(r*r);
+
+
+    vec2 v = vec2(0.);
+    for(float i=0.;i<counter;i++){
+    	v = mat2(v,-v.y,v.x) * v + vec2(2.*f+cos(0.5*t*(exp(-0.2* r))),-cos(t*r*r)*cos(0.5*t));
+        if(length(v)>2.){
+        	counter = i;
+            break;
+        }
+    }
+
+    col=vec3(min(0.9,1.2*exp(-pow(f,0.45)*counter)));
+
+	fragColor =    min(0.9,1.2*exp(-pow(f,0.45)*counter) )
+                * ( 0.7 + 0.3* cos(10.*r - 2.*t -vec4(.7,1.4,2.1,0) ) );
+}

pkg/macos/build.zig

@@ -28,6 +28,7 @@ pub fn build(b: *std.Build) !void {
     lib.linkFramework("CoreFoundation");
     lib.linkFramework("CoreGraphics");
     lib.linkFramework("CoreText");
+    lib.linkFramework("CoreVideo");
     if (!target.isNative()) try apple_sdk.addPaths(b, lib);
 
     b.installArtifact(lib);

pkg/macos/main.zig

@@ -2,6 +2,7 @@ pub const foundation = @import("foundation.zig");
 pub const graphics = @import("graphics.zig");
 pub const os = @import("os.zig");
 pub const text = @import("text.zig");
+pub const video = @import("video.zig");
 
 test {
     @import("std").testing.refAllDecls(@This());

pkg/macos/video.zig

@@ -0,0 +1,6 @@
+pub const c = @import("video/c.zig");
+pub usingnamespace @import("video/display_link.zig");
+
+test {
+    @import("std").testing.refAllDecls(@This());
+}

pkg/macos/video/c.zig

@@ -0,0 +1,3 @@
+pub usingnamespace @cImport({
+    @cInclude("CoreVideo/CoreVideo.h");
+});

pkg/macos/video/display_link.zig

@@ -0,0 +1,71 @@
+const std = @import("std");
+const assert = std.debug.assert;
+const Allocator = std.mem.Allocator;
+const c = @import("c.zig");
+
+pub const DisplayLink = opaque {
+    pub const Error = error{
+        InvalidOperation,
+    };
+
+    pub fn createWithActiveCGDisplays() Allocator.Error!*DisplayLink {
+        var result: ?*DisplayLink = null;
+        if (c.CVDisplayLinkCreateWithActiveCGDisplays(
+            @ptrCast(&result),
+        ) != c.kCVReturnSuccess)
+            return error.OutOfMemory;
+
+        return result orelse error.OutOfMemory;
+    }
+
+    pub fn release(self: *DisplayLink) void {
+        c.CVDisplayLinkRelease(@ptrCast(self));
+    }
+
+    pub fn start(self: *DisplayLink) Error!void {
+        if (c.CVDisplayLinkStart(@ptrCast(self)) != c.kCVReturnSuccess)
+            return error.InvalidOperation;
+    }
+
+    pub fn stop(self: *DisplayLink) Error!void {
+        if (c.CVDisplayLinkStop(@ptrCast(self)) != c.kCVReturnSuccess)
+            return error.InvalidOperation;
+    }
+
+    pub fn isRunning(self: *DisplayLink) bool {
+        return c.CVDisplayLinkIsRunning(@ptrCast(self)) != 0;
+    }
+
+    // Note: this purposely throws away a ton of arguments I didn't need.
+    // It would be trivial to refactor this into Zig types and properly
+    // pass this through.
+    pub fn setOutputCallback(
+        self: *DisplayLink,
+        comptime callbackFn: *const fn (*DisplayLink, ?*anyopaque) void,
+        userinfo: ?*anyopaque,
+    ) Error!void {
+        if (c.CVDisplayLinkSetOutputCallback(
+            @ptrCast(self),
+            @ptrCast(&(struct {
+                fn callback(
+                    displayLink: *DisplayLink,
+                    inNow: *const c.CVTimeStamp,
+                    inOutputTime: *const c.CVTimeStamp,
+                    flagsIn: c.CVOptionFlags,
+                    flagsOut: *c.CVOptionFlags,
+                    inner_userinfo: ?*anyopaque,
+                ) callconv(.C) c.CVReturn {
+                    _ = inNow;
+                    _ = inOutputTime;
+                    _ = flagsIn;
+                    _ = flagsOut;
+
+                    callbackFn(displayLink, inner_userinfo);
+                    return c.kCVReturnSuccess;
+                }
+            }).callback),
+            userinfo,
+        ) != c.kCVReturnSuccess)
+            return error.InvalidOperation;
+    }
+};

pkg/opengl/Buffer.zig

@@ -7,76 +7,106 @@ const glad = @import("glad.zig");
 
 id: c.GLuint,
 
-/// Enum for possible binding targets.
-pub const Target = enum(c_uint) {
-    ArrayBuffer = c.GL_ARRAY_BUFFER,
-    ElementArrayBuffer = c.GL_ELEMENT_ARRAY_BUFFER,
-    _,
-};
+/// Create a single buffer.
+pub fn create() !Buffer {
+    var vbo: c.GLuint = undefined;
+    glad.context.GenBuffers.?(1, &vbo);
+    return Buffer{ .id = vbo };
+}
 
-/// Enum for possible buffer usages.
-pub const Usage = enum(c_uint) {
-    StreamDraw = c.GL_STREAM_DRAW,
-    StreamRead = c.GL_STREAM_READ,
-    StreamCopy = c.GL_STREAM_COPY,
-    StaticDraw = c.GL_STATIC_DRAW,
-    StaticRead = c.GL_STATIC_READ,
-    StaticCopy = c.GL_STATIC_COPY,
-    DynamicDraw = c.GL_DYNAMIC_DRAW,
-    DynamicRead = c.GL_DYNAMIC_READ,
-    DynamicCopy = c.GL_DYNAMIC_COPY,
-    _,
-};
+/// glBindBuffer
+pub fn bind(self: Buffer, target: Target) !Binding {
+    glad.context.BindBuffer.?(@intFromEnum(target), self.id);
+    return Binding{ .id = self.id, .target = target };
+}
+
+pub fn destroy(self: Buffer) void {
+    glad.context.DeleteBuffers.?(1, &self.id);
+}
+
+pub fn bindBase(self: Buffer, target: Target, idx: c.GLuint) !void {
+    glad.context.BindBufferBase.?(
+        @intFromEnum(target),
+        idx,
+        self.id,
+    );
+    try errors.getError();
+}
 
 /// Binding is a bound buffer. By using this for functions that operate
 /// on bound buffers, you can easily defer unbinding and in safety-enabled
 /// modes verify that unbound buffers are never accessed.
 pub const Binding = struct {
+    id: c.GLuint,
     target: Target,
 
+    pub fn unbind(b: Binding) void {
+        glad.context.BindBuffer.?(@intFromEnum(b.target), 0);
+    }
+
     /// Sets the data of this bound buffer. The data can be any array-like
     /// type. The size of the data is automatically determined based on the type.
-    pub inline fn setData(
+    pub fn setData(
         b: Binding,
         data: anytype,
         usage: Usage,
     ) !void {
         const info = dataInfo(&data);
-        glad.context.BufferData.?(@intFromEnum(b.target), info.size, info.ptr, @intFromEnum(usage));
+        glad.context.BufferData.?(
+            @intFromEnum(b.target),
+            info.size,
+            info.ptr,
+            @intFromEnum(usage),
+        );
         try errors.getError();
     }
 
     /// Sets the data of this bound buffer. The data can be any array-like
     /// type. The size of the data is automatically determined based on the type.
-    pub inline fn setSubData(
+    pub fn setSubData(
         b: Binding,
         offset: usize,
         data: anytype,
     ) !void {
         const info = dataInfo(data);
-        glad.context.BufferSubData.?(@intFromEnum(b.target), @intCast(offset), info.size, info.ptr);
+        glad.context.BufferSubData.?(
+            @intFromEnum(b.target),
+            @intCast(offset),
+            info.size,
+            info.ptr,
+        );
         try errors.getError();
     }
 
     /// Sets the buffer data with a null buffer that is expected to be
     /// filled in the future using subData. This requires the type just so
     /// we can setup the data size.
-    pub inline fn setDataNull(
+    pub fn setDataNull(
         b: Binding,
         comptime T: type,
         usage: Usage,
     ) !void {
-        glad.context.BufferData.?(@intFromEnum(b.target), @sizeOf(T), null, @intFromEnum(usage));
+        glad.context.BufferData.?(
+            @intFromEnum(b.target),
+            @sizeOf(T),
+            null,
+            @intFromEnum(usage),
+        );
         try errors.getError();
     }
 
     /// Same as setDataNull but lets you manually specify the buffer size.
-    pub inline fn setDataNullManual(
+    pub fn setDataNullManual(
         b: Binding,
         size: usize,
         usage: Usage,
     ) !void {
-        glad.context.BufferData.?(@intFromEnum(b.target), @intCast(size), null, @intFromEnum(usage));
+        glad.context.BufferData.?(
+            @intFromEnum(b.target),
+            @intCast(size),
+            null,
+            @intFromEnum(usage),
+        );
         try errors.getError();
     }
 
@@ -87,7 +117,7 @@ pub const Binding = struct {
         return switch (@typeInfo(@TypeOf(data))) {
             .Pointer => |ptr| switch (ptr.size) {
                 .One => .{
-                    .size = @sizeOf(ptr.child) * data.len,
+                    .size = @sizeOf(ptr.child),
                     .ptr = data,
                 },
                 .Slice => .{
@@ -106,7 +136,7 @@ pub const Binding = struct {
         };
     }
 
-    pub inline fn enableAttribArray(_: Binding, idx: c.GLuint) !void {
+    pub fn enableAttribArray(_: Binding, idx: c.GLuint) !void {
         glad.context.EnableVertexAttribArray.?(idx);
     }
 
@@ -158,7 +188,7 @@ pub const Binding = struct {
         try errors.getError();
     }
 
-    pub inline fn attributeAdvanced(
+    pub fn attributeAdvanced(
         _: Binding,
         idx: c.GLuint,
         size: c.GLint,
@@ -177,7 +207,7 @@ pub const Binding = struct {
         try errors.getError();
     }
 
-    pub inline fn attributeIAdvanced(
+    pub fn attributeIAdvanced(
         _: Binding,
         idx: c.GLuint,
         size: c.GLint,
@@ -193,26 +223,26 @@ pub const Binding = struct {
         glad.context.VertexAttribIPointer.?(idx, size, typ, stride, offsetPtr);
         try errors.getError();
     }
-
-    pub inline fn unbind(b: *Binding) void {
-        glad.context.BindBuffer.?(@intFromEnum(b.target), 0);
-        b.* = undefined;
-    }
 };
 
-/// Create a single buffer.
-pub inline fn create() !Buffer {
-    var vbo: c.GLuint = undefined;
-    glad.context.GenBuffers.?(1, &vbo);
-    return Buffer{ .id = vbo };
-}
+/// Enum for possible binding targets.
+pub const Target = enum(c_uint) {
+    array = c.GL_ARRAY_BUFFER,
+    element_array = c.GL_ELEMENT_ARRAY_BUFFER,
+    uniform = c.GL_UNIFORM_BUFFER,
+    _,
+};
 
-/// glBindBuffer
-pub inline fn bind(v: Buffer, target: Target) !Binding {
-    glad.context.BindBuffer.?(@intFromEnum(target), v.id);
-    return Binding{ .target = target };
-}
-
-pub inline fn destroy(v: Buffer) void {
-    glad.context.DeleteBuffers.?(1, &v.id);
-}
+/// Enum for possible buffer usages.
+pub const Usage = enum(c_uint) {
+    stream_draw = c.GL_STREAM_DRAW,
+    stream_read = c.GL_STREAM_READ,
+    stream_copy = c.GL_STREAM_COPY,
+    static_draw = c.GL_STATIC_DRAW,
+    static_read = c.GL_STATIC_READ,
+    static_copy = c.GL_STATIC_COPY,
+    dynamic_draw = c.GL_DYNAMIC_DRAW,
+    dynamic_read = c.GL_DYNAMIC_READ,
+    dynamic_copy = c.GL_DYNAMIC_COPY,
+    _,
+};

pkg/opengl/Framebuffer.zig

@@ -0,0 +1,92 @@
+const Framebuffer = @This();
+
+const std = @import("std");
+const c = @import("c.zig");
+const errors = @import("errors.zig");
+const glad = @import("glad.zig");
+const Texture = @import("Texture.zig");
+
+id: c.GLuint,
+
+/// Create a single buffer.
+pub fn create() !Framebuffer {
+    var fbo: c.GLuint = undefined;
+    glad.context.GenFramebuffers.?(1, &fbo);
+    return .{ .id = fbo };
+}
+
+pub fn destroy(v: Framebuffer) void {
+    glad.context.DeleteFramebuffers.?(1, &v.id);
+}
+
+pub fn bind(v: Framebuffer, target: Target) !Binding {
+    // The default framebuffer is documented as being zero but
+    // on multiple OpenGL drivers its not zero, so we grab it
+    // at runtime.
+    var current: c.GLint = undefined;
+    glad.context.GetIntegerv.?(c.GL_FRAMEBUFFER_BINDING, &current);
+    glad.context.BindFramebuffer.?(@intFromEnum(target), v.id);
+    return .{ .target = target, .previous = @intCast(current) };
+}
+
+/// Enum for possible binding targets.
+pub const Target = enum(c_uint) {
+    framebuffer = c.GL_FRAMEBUFFER,
+    draw = c.GL_DRAW_FRAMEBUFFER,
+    read = c.GL_READ_FRAMEBUFFER,
+    _,
+};
+
+pub const Attachment = enum(c_uint) {
+    color0 = c.GL_COLOR_ATTACHMENT0,
+    depth = c.GL_DEPTH_ATTACHMENT,
+    stencil = c.GL_STENCIL_ATTACHMENT,
+    depth_stencil = c.GL_DEPTH_STENCIL_ATTACHMENT,
+    _,
+};
+
+pub const Status = enum(c_uint) {
+    complete = c.GL_FRAMEBUFFER_COMPLETE,
+    undefined = c.GL_FRAMEBUFFER_UNDEFINED,
+    incomplete_attachment = c.GL_FRAMEBUFFER_INCOMPLETE_ATTACHMENT,
+    incomplete_missing_attachment = c.GL_FRAMEBUFFER_INCOMPLETE_MISSING_ATTACHMENT,
+    incomplete_draw_buffer = c.GL_FRAMEBUFFER_INCOMPLETE_DRAW_BUFFER,
+    incomplete_read_buffer = c.GL_FRAMEBUFFER_INCOMPLETE_READ_BUFFER,
+    unsupported = c.GL_FRAMEBUFFER_UNSUPPORTED,
+    incomplete_multisample = c.GL_FRAMEBUFFER_INCOMPLETE_MULTISAMPLE,
+    incomplete_layer_targets = c.GL_FRAMEBUFFER_INCOMPLETE_LAYER_TARGETS,
+    _,
+};
+
+pub const Binding = struct {
+    target: Target,
+    previous: c.GLuint,
+
+    pub fn unbind(self: Binding) void {
+        glad.context.BindFramebuffer.?(
+            @intFromEnum(self.target),
+            self.previous,
+        );
+    }
+
+    pub fn texture2D(
+        self: Binding,
+        attachment: Attachment,
+        textarget: Texture.Target,
+        texture: Texture,
+        level: c.GLint,
+    ) !void {
+        glad.context.FramebufferTexture2D.?(
+            @intFromEnum(self.target),
+            @intFromEnum(attachment),
+            @intFromEnum(textarget),
+            texture.id,
+            level,
+        );
+        try errors.getError();
+    }
+
+    pub fn checkStatus(self: Binding) Status {
+        return @enumFromInt(glad.context.CheckFramebufferStatus.?(@intFromEnum(self.target)));
+    }
+};

pkg/opengl/Program.zig

@@ -11,23 +11,22 @@ const glad = @import("glad.zig");
 
 id: c.GLuint,
 
-const Binding = struct {
-    pub inline fn unbind(_: Binding) void {
+pub const Binding = struct {
+    pub fn unbind(_: Binding) void {
         glad.context.UseProgram.?(0);
     }
 };
 
-pub inline fn create() !Program {
+pub fn create() !Program {
     const id = glad.context.CreateProgram.?();
     if (id == 0) try errors.mustError();
-
     log.debug("program created id={}", .{id});
-    return Program{ .id = id };
+    return .{ .id = id };
 }
 
 /// Create a program from a vertex and fragment shader source. This will
 /// compile and link the vertex and fragment shader.
-pub inline fn createVF(vsrc: [:0]const u8, fsrc: [:0]const u8) !Program {
+pub fn createVF(vsrc: [:0]const u8, fsrc: [:0]const u8) !Program {
     const vs = try Shader.create(c.GL_VERTEX_SHADER);
     try vs.setSourceAndCompile(vsrc);
     defer vs.destroy();
@@ -44,12 +43,18 @@ pub inline fn createVF(vsrc: [:0]const u8, fsrc: [:0]const u8) !Program {
     return p;
 }
 
-pub inline fn attachShader(p: Program, s: Shader) !void {
+pub fn destroy(p: Program) void {
+    assert(p.id != 0);
+    glad.context.DeleteProgram.?(p.id);
+    log.debug("program destroyed id={}", .{p.id});
+}
+
+pub fn attachShader(p: Program, s: Shader) !void {
     glad.context.AttachShader.?(p.id, s.id);
     try errors.getError();
 }
 
-pub inline fn link(p: Program) !void {
+pub fn link(p: Program) !void {
     glad.context.LinkProgram.?(p.id);
 
     // Check if linking succeeded
@@ -67,14 +72,23 @@ pub inline fn link(p: Program) !void {
     return error.CompileFailed;
 }
 
-pub inline fn use(p: Program) !Binding {
+pub fn use(p: Program) !Binding {
     glad.context.UseProgram.?(p.id);
     try errors.getError();
-    return Binding{};
+    return .{};
+}
+
+pub fn uniformBlockBinding(
+    self: Program,
+    index: c.GLuint,
+    binding: c.GLuint,
+) !void {
+    glad.context.UniformBlockBinding.?(self.id, index, binding);
+    try errors.getError();
 }
 
 /// Requires the program is currently in use.
-pub inline fn setUniform(
+pub fn setUniform(
     p: Program,
     n: [:0]const u8,
     value: anytype,
@@ -115,14 +129,8 @@ pub inline fn setUniform(
 //
 // NOTE(mitchellh): we can add a dynamic version that uses an allocator
 // if we ever need it.
-pub inline fn getInfoLog(s: Program) [512]u8 {
+pub fn getInfoLog(s: Program) [512]u8 {
     var msg: [512]u8 = undefined;
     glad.context.GetProgramInfoLog.?(s.id, msg.len, null, &msg);
     return msg;
 }
-
-pub inline fn destroy(p: Program) void {
-    assert(p.id != 0);
-    glad.context.DeleteProgram.?(p.id);
-    log.debug("program destroyed id={}", .{p.id});
-}

pkg/opengl/Texture.zig

@@ -7,11 +7,29 @@ const glad = @import("glad.zig");
 
 id: c.GLuint,
 
-pub inline fn active(target: c.GLenum) !void {
+pub fn active(target: c.GLenum) !void {
     glad.context.ActiveTexture.?(target);
     try errors.getError();
 }
 
+/// Create a single texture.
+pub fn create() !Texture {
+    var id: c.GLuint = undefined;
+    glad.context.GenTextures.?(1, &id);
+    return .{ .id = id };
+}
+
+/// glBindTexture
+pub fn bind(v: Texture, target: Target) !Binding {
+    glad.context.BindTexture.?(@intFromEnum(target), v.id);
+    try errors.getError();
+    return .{ .target = target };
+}
+
+pub fn destroy(v: Texture) void {
+    glad.context.DeleteTextures.?(1, &v.id);
+}
+
 /// Enun for possible texture binding targets.
 pub const Target = enum(c_uint) {
     @"1D" = c.GL_TEXTURE_1D,
@@ -48,8 +66,9 @@ pub const Parameter = enum(c_uint) {
 
 /// Internal format enum for texture images.
 pub const InternalFormat = enum(c_int) {
-    Red = c.GL_RED,
-    RGBA = c.GL_RGBA,
+    red = c.GL_RED,
+    rgb = c.GL_RGB,
+    rgba = c.GL_RGBA,
 
     // There are so many more that I haven't filled in.
     _,
@@ -57,8 +76,9 @@ pub const InternalFormat = enum(c_int) {
 
 /// Format for texture images
 pub const Format = enum(c_uint) {
-    Red = c.GL_RED,
-    BGRA = c.GL_BGRA,
+    red = c.GL_RED,
+    rgb = c.GL_RGB,
+    bgra = c.GL_BGRA,
 
     // There are so many more that I haven't filled in.
     _,
@@ -75,9 +95,8 @@ pub const DataType = enum(c_uint) {
 pub const Binding = struct {
     target: Target,
 
-    pub inline fn unbind(b: *Binding) void {
+    pub fn unbind(b: *const Binding) void {
         glad.context.BindTexture.?(@intFromEnum(b.target), 0);
-        b.* = undefined;
     }
 
     pub fn generateMipmap(b: Binding) void {
@@ -143,21 +162,3 @@ pub const Binding = struct {
         );
     }
 };
-
-/// Create a single texture.
-pub inline fn create() !Texture {
-    var id: c.GLuint = undefined;
-    glad.context.GenTextures.?(1, &id);
-    return Texture{ .id = id };
-}
-
-/// glBindTexture
-pub inline fn bind(v: Texture, target: Target) !Binding {
-    glad.context.BindTexture.?(@intFromEnum(target), v.id);
-    try errors.getError();
-    return Binding{ .target = target };
-}
-
-pub inline fn destroy(v: Texture) void {
-    glad.context.DeleteTextures.?(1, &v.id);
-}

pkg/opengl/VertexArray.zig

@@ -7,23 +7,26 @@ const errors = @import("errors.zig");
 id: c.GLuint,
 
 /// Create a single vertex array object.
-pub inline fn create() !VertexArray {
+pub fn create() !VertexArray {
     var vao: c.GLuint = undefined;
     glad.context.GenVertexArrays.?(1, &vao);
     return VertexArray{ .id = vao };
 }
 
-// Unbind any active vertex array.
-pub inline fn unbind() !void {
-    glad.context.BindVertexArray.?(0);
-}
-
 /// glBindVertexArray
-pub inline fn bind(v: VertexArray) !void {
+pub fn bind(v: VertexArray) !Binding {
     glad.context.BindVertexArray.?(v.id);
     try errors.getError();
+    return .{};
 }
 
-pub inline fn destroy(v: VertexArray) void {
+pub fn destroy(v: VertexArray) void {
     glad.context.DeleteVertexArrays.?(1, &v.id);
 }
+
+pub const Binding = struct {
+    pub fn unbind(self: Binding) void {
+        _ = self;
+        glad.context.BindVertexArray.?(0);
+    }
+};

pkg/opengl/build.zig

@@ -0,0 +1,5 @@
+const std = @import("std");
+
+pub fn build(b: *std.Build) !void {
+    _ = b.addModule("opengl", .{ .source_file = .{ .path = "main.zig" } });
+}

pkg/opengl/main.zig

@@ -17,6 +17,7 @@ pub usingnamespace @import("draw.zig");
 
 pub const ext = @import("extensions.zig");
 pub const Buffer = @import("Buffer.zig");
+pub const Framebuffer = @import("Framebuffer.zig");
 pub const Program = @import("Program.zig");
 pub const Shader = @import("Shader.zig");
 pub const Texture = @import("Texture.zig");

pkg/spirv-cross/build.zig

@@ -0,0 +1,85 @@
+const std = @import("std");
+
+pub fn build(b: *std.Build) !void {
+    const target = b.standardTargetOptions(.{});
+    const optimize = b.standardOptimizeOption(.{});
+
+    _ = b.addModule("spirv_cross", .{ .source_file = .{ .path = "main.zig" } });
+
+    const upstream = b.dependency("spirv_cross", .{});
+    const lib = try buildSpirvCross(b, upstream, target, optimize);
+    b.installArtifact(lib);
+
+    {
+        const test_exe = b.addTest(.{
+            .name = "test",
+            .root_source_file = .{ .path = "main.zig" },
+            .target = target,
+            .optimize = optimize,
+        });
+        test_exe.linkLibrary(lib);
+        const tests_run = b.addRunArtifact(test_exe);
+        const test_step = b.step("test", "Run tests");
+        test_step.dependOn(&tests_run.step);
+
+        // Uncomment this if we're debugging tests
+        // b.installArtifact(test_exe);
+    }
+}
+
+fn buildSpirvCross(
+    b: *std.Build,
+    upstream: *std.Build.Dependency,
+    target: std.zig.CrossTarget,
+    optimize: std.builtin.OptimizeMode,
+) !*std.Build.Step.Compile {
+    const lib = b.addStaticLibrary(.{
+        .name = "spirv_cross",
+        .target = target,
+        .optimize = optimize,
+    });
+    lib.linkLibC();
+    lib.linkLibCpp();
+    //lib.addIncludePath(upstream.path(""));
+    //lib.addIncludePath(.{ .path = "override" });
+
+    var flags = std.ArrayList([]const u8).init(b.allocator);
+    defer flags.deinit();
+    try flags.appendSlice(&.{
+        "-DSPIRV_CROSS_C_API_GLSL=1",
+        "-DSPIRV_CROSS_C_API_MSL=1",
+
+        "-fno-sanitize=undefined",
+        "-fno-sanitize-trap=undefined",
+    });
+
+    lib.addCSourceFiles(.{
+        .dependency = upstream,
+        .flags = flags.items,
+        .files = &.{
+            // Core
+            "spirv_cross.cpp",
+            "spirv_parser.cpp",
+            "spirv_cross_parsed_ir.cpp",
+            "spirv_cfg.cpp",
+
+            // C
+            "spirv_cross_c.cpp",
+
+            // GLSL
+            "spirv_glsl.cpp",
+
+            // MSL
+            "spirv_msl.cpp",
+        },
+    });
+
+    lib.installHeadersDirectoryOptions(.{
+        .source_dir = upstream.path(""),
+        .install_dir = .header,
+        .install_subdir = "",
+        .include_extensions = &.{".h"},
+    });
+
+    return lib;
+}

pkg/spirv-cross/build.zig.zon

@@ -0,0 +1,11 @@
+.{
+    .name = "spirv-cross",
+    .version = "13.1.1",
+    .paths = .{""},
+    .dependencies = .{
+        .spirv_cross = .{
+            .url = "https://github.com/KhronosGroup/SPIRV-Cross/archive/4818f7e7ef7b7078a3a7a5a52c4a338e0dda22f4.tar.gz",
+            .hash = "1220b2d8a6cff1926ef28a29e312a0a503b555ebc2f082230b882410f49e672ac9c6",
+        },
+    },
+}

pkg/spirv-cross/c.zig

@@ -0,0 +1,3 @@
+pub usingnamespace @cImport({
+    @cInclude("spirv_cross_c.h");
+});

pkg/spirv-cross/main.zig

@@ -0,0 +1 @@
+pub const c = @import("c.zig");

src/apprt/glfw.zig

@@ -291,7 +291,7 @@ pub const App = struct {
         tabbing_id: *macos.foundation.String,
 
         pub fn init() !Darwin {
-            const NSWindow = objc.Class.getClass("NSWindow").?;
+            const NSWindow = objc.getClass("NSWindow").?;
             NSWindow.msgSend(void, objc.sel("setAllowsAutomaticWindowTabbing:"), .{true});
 
             // Our tabbing ID allows all of our windows to group together

src/apprt/gtk/ImguiWidget.zig

@@ -6,7 +6,7 @@ const assert = std.debug.assert;
 const cimgui = @import("cimgui");
 const c = @import("c.zig");
 const key = @import("key.zig");
-const gl = @import("../../renderer/opengl/main.zig");
+const gl = @import("opengl");
 const input = @import("../../input.zig");
 
 const log = std.log.scoped(.gtk_imgui_widget);

src/apprt/gtk/Surface.zig

@@ -252,7 +252,7 @@ pub fn deinit(self: *Surface) void {
 }
 
 fn render(self: *Surface) !void {
-    try self.core_surface.renderer.draw();
+    try self.core_surface.renderer.drawFrame(self);
 }
 
 /// Queue the inspector to render if we have one.

src/config/Config.zig

@@ -528,7 +528,7 @@ keybind: Keybinds = .{},
 ///
 /// Cycles are not allowed. If a cycle is detected, an error will be logged
 /// and the configuration file will be ignored.
-@"config-file": RepeatableString = .{},
+@"config-file": RepeatablePath = .{},
 
 /// Confirms that a surface should be closed before closing it. This defaults
 /// to true. If set to false, surfaces will close without any confirmation.
@@ -599,6 +599,46 @@ keybind: Keybinds = .{},
 /// need KAM, you don't need it.
 @"vt-kam-allowed": bool = false,
 
+/// Custom shaders to run after the default shaders. This is a file path
+/// to a GLSL-syntax shader for all platforms.
+///
+/// WARNING: Invalid shaders can cause Ghostty to become unusable such as by
+/// causing the window to be completely black. If this happens, you can
+/// unset this configuration to disable the shader.
+///
+/// On Linux, this requires OpenGL 4.2. Ghostty typically only requires
+/// OpenGL 3.3, but custom shaders push that requirement up to 4.2.
+///
+/// The shader API is identical to the Shadertoy API: you specify a `mainImage`
+/// function and the available uniforms match Shadertoy. The iChannel0 uniform
+/// is a texture containing the rendered terminal screen.
+///
+/// If the shader fails to compile, the shader will be ignored. Any errors
+/// related to shader compilation will not show up as configuration errors
+/// and only show up in the log, since shader compilation happens after
+/// configuration loading on the dedicated render thread.  For interactive
+/// development, use Shadertoy.com.
+///
+/// This can be repeated multiple times to load multiple shaders. The shaders
+/// will be run in the order they are specified.
+///
+/// Changing this value at runtime and reloading the configuration will only
+/// affect new windows, tabs, and splits.
+@"custom-shader": RepeatablePath = .{},
+
+/// If true (default), the focused terminal surface will run an animation
+/// loop when custom shaders are used. This uses slightly more CPU (generally
+/// less than 10%) but allows the shader to animate. This only runs if there
+/// are custom shaders.
+///
+/// If this is set to false, the terminal and custom shader will only render
+/// when the terminal is updated. This is more efficient but the shader will
+/// not animate.
+///
+/// This value can be changed at runtime and will affect all currently
+/// open terminals.
+@"custom-shader-animation": bool = true,
+
 /// If anything other than false, fullscreen mode on macOS will not use the
 /// native fullscreen, but make the window fullscreen without animations and
 /// using a new space. It's faster than the native fullscreen mode since it
@@ -1143,7 +1183,7 @@ pub fn loadDefaultFiles(self: *Config, alloc: Allocator) !void {
         var buf_reader = std.io.bufferedReader(file.reader());
         var iter = cli.args.lineIterator(buf_reader.reader());
         try cli.args.parse(Config, alloc, self, &iter);
-        try self.expandConfigFiles(std.fs.path.dirname(config_path).?);
+        try self.expandPaths(std.fs.path.dirname(config_path).?);
     } else |err| switch (err) {
         error.FileNotFound => std.log.info(
             "homedir config not found, not loading path={s}",
@@ -1172,15 +1212,15 @@ pub fn loadCliArgs(self: *Config, alloc_gpa: Allocator) !void {
     try cli.args.parse(Config, alloc_gpa, self, &iter);
 
     // Config files loaded from the CLI args are relative to pwd
-    if (self.@"config-file".list.items.len > 0) {
+    if (self.@"config-file".value.list.items.len > 0) {
         var buf: [std.fs.MAX_PATH_BYTES]u8 = undefined;
-        try self.expandConfigFiles(try std.fs.cwd().realpath(".", &buf));
+        try self.expandPaths(try std.fs.cwd().realpath(".", &buf));
     }
 }
 
 /// Load and parse the config files that were added in the "config-file" key.
 pub fn loadRecursiveFiles(self: *Config, alloc_gpa: Allocator) !void {
-    if (self.@"config-file".list.items.len == 0) return;
+    if (self.@"config-file".value.list.items.len == 0) return;
     const arena_alloc = self._arena.?.allocator();
 
     // Keeps track of loaded files to prevent cycles.
@@ -1189,8 +1229,8 @@ pub fn loadRecursiveFiles(self: *Config, alloc_gpa: Allocator) !void {
 
     const cwd = std.fs.cwd();
     var i: usize = 0;
-    while (i < self.@"config-file".list.items.len) : (i += 1) {
-        const path = self.@"config-file".list.items[i];
+    while (i < self.@"config-file".value.list.items.len) : (i += 1) {
+        const path = self.@"config-file".value.list.items[i];
 
         // Error paths
         if (path.len == 0) continue;
@@ -1227,37 +1267,22 @@ pub fn loadRecursiveFiles(self: *Config, alloc_gpa: Allocator) !void {
         var buf_reader = std.io.bufferedReader(file.reader());
         var iter = cli.args.lineIterator(buf_reader.reader());
         try cli.args.parse(Config, alloc_gpa, self, &iter);
-        try self.expandConfigFiles(std.fs.path.dirname(path).?);
+        try self.expandPaths(std.fs.path.dirname(path).?);
     }
 }
 
 /// Expand the relative paths in config-files to be absolute paths
 /// relative to the base directory.
-fn expandConfigFiles(self: *Config, base: []const u8) !void {
-    assert(std.fs.path.isAbsolute(base));
-    var dir = try std.fs.cwd().openDir(base, .{});
-    defer dir.close();
-
+fn expandPaths(self: *Config, base: []const u8) !void {
     const arena_alloc = self._arena.?.allocator();
-    for (self.@"config-file".list.items, 0..) |path, i| {
-        // If it is already absolute we can ignore it.
-        if (path.len == 0 or std.fs.path.isAbsolute(path)) continue;
-
-        // If it isn't absolute, we need to make it absolute relative to the base.
-        const abs = dir.realpathAlloc(arena_alloc, path) catch |err| {
-            try self._errors.add(arena_alloc, .{
-                .message = try std.fmt.allocPrintZ(
-                    arena_alloc,
-                    "error resolving config-file {s}: {}",
-                    .{ path, err },
-                ),
-            });
-            self.@"config-file".list.items[i] = "";
-            continue;
-        };
-
-        log.debug("expanding config-file path relative={s} abs={s}", .{ path, abs });
-        self.@"config-file".list.items[i] = abs;
+    inline for (@typeInfo(Config).Struct.fields) |field| {
+        if (field.type == RepeatablePath) {
+            try @field(self, field.name).expand(
+                arena_alloc,
+                base,
+                &self._errors,
+            );
+        }
     }
 }
 
@@ -1889,6 +1914,69 @@ pub const RepeatableString = struct {
     }
 };
 
+/// RepeatablePath is like repeatable string but represents a path value.
+/// The difference is that when loading the configuration any values for
+/// this will be automatically expanded relative to the path of the config
+/// file.
+pub const RepeatablePath = struct {
+    const Self = @This();
+
+    value: RepeatableString = .{},
+
+    pub fn parseCLI(self: *Self, alloc: Allocator, input: ?[]const u8) !void {
+        return self.value.parseCLI(alloc, input);
+    }
+
+    /// Deep copy of the struct. Required by Config.
+    pub fn clone(self: *const Self, alloc: Allocator) !Self {
+        return .{
+            .value = try self.value.clone(alloc),
+        };
+    }
+
+    /// Compare if two of our value are requal. Required by Config.
+    pub fn equal(self: Self, other: Self) bool {
+        return self.value.equal(other.value);
+    }
+
+    /// Expand all the paths relative to the base directory.
+    pub fn expand(
+        self: *Self,
+        alloc: Allocator,
+        base: []const u8,
+        errors: *ErrorList,
+    ) !void {
+        assert(std.fs.path.isAbsolute(base));
+        var dir = try std.fs.cwd().openDir(base, .{});
+        defer dir.close();
+
+        for (self.value.list.items, 0..) |path, i| {
+            // If it is already absolute we can ignore it.
+            if (path.len == 0 or std.fs.path.isAbsolute(path)) continue;
+
+            // If it isn't absolute, we need to make it absolute relative
+            // to the base.
+            const abs = dir.realpathAlloc(alloc, path) catch |err| {
+                try errors.add(alloc, .{
+                    .message = try std.fmt.allocPrintZ(
+                        alloc,
+                        "error resolving config-file {s}: {}",
+                        .{ path, err },
+                    ),
+                });
+                self.value.list.items[i] = "";
+                continue;
+            };
+
+            log.debug(
+                "expanding config-file path relative={s} abs={s}",
+                .{ path, abs },
+            );
+            self.value.list.items[i] = abs;
+        }
+    }
+};
+
 /// FontVariation is a repeatable configuration value that sets a single
 /// font variation value. Font variations are configurations for what
 /// are often called "variable fonts." The font files usually end in

src/main.zig

@@ -4,6 +4,7 @@ const Allocator = std.mem.Allocator;
 const build_config = @import("build_config.zig");
 const options = @import("build_options");
 const glfw = @import("glfw");
+const glslang = @import("glslang");
 const macos = @import("macos");
 const tracy = @import("tracy");
 const cli = @import("cli.zig");
@@ -267,6 +268,9 @@ pub const GlobalState = struct {
         // We need to make sure the process locale is set properly. Locale
         // affects a lot of behaviors in a shell.
         try internal_os.ensureLocale(self.alloc);
+
+        // Initialize glslang for shader compilation
+        try glslang.init();
     }
 
     /// Cleans up the global state. This doesn't _need_ to be called but

src/os/locale.zig

@@ -70,7 +70,7 @@ fn setLangFromCocoa() void {
     defer pool.deinit();
 
     // The classes we're going to need.
-    const NSLocale = objc.Class.getClass("NSLocale") orelse {
+    const NSLocale = objc.getClass("NSLocale") orelse {
         log.err("NSLocale class not found. Locale may be incorrect.", .{});
         return;
     };

src/os/macos_version.zig

@@ -7,7 +7,7 @@ const objc = @import("objc");
 pub fn macosVersionAtLeast(major: i64, minor: i64, patch: i64) bool {
     assert(builtin.target.isDarwin());
 
-    const NSProcessInfo = objc.Class.getClass("NSProcessInfo").?;
+    const NSProcessInfo = objc.getClass("NSProcessInfo").?;
     const info = NSProcessInfo.msgSend(objc.Object, objc.sel("processInfo"), .{});
     return info.msgSend(bool, objc.sel("isOperatingSystemAtLeastVersion:"), .{
         NSOperatingSystemVersion{ .major = major, .minor = minor, .patch = patch },

src/os/mouse.zig

@@ -9,7 +9,7 @@ const log = std.log.scoped(.os);
 pub fn clickInterval() ?u32 {
     // On macOS, we can ask the system.
     if (comptime builtin.target.isDarwin()) {
-        const NSEvent = objc.Class.getClass("NSEvent") orelse {
+        const NSEvent = objc.getClass("NSEvent") orelse {
             log.err("NSEvent class not found. Can't get click interval.", .{});
             return null;
         };

src/renderer.zig

@@ -15,6 +15,7 @@ const WasmTarget = @import("os/wasm/target.zig").Target;
 pub usingnamespace @import("renderer/cursor.zig");
 pub usingnamespace @import("renderer/message.zig");
 pub usingnamespace @import("renderer/size.zig");
+pub const shadertoy = @import("renderer/shadertoy.zig");
 pub const Metal = @import("renderer/Metal.zig");
 pub const OpenGL = @import("renderer/OpenGL.zig");
 pub const WebGL = @import("renderer/WebGL.zig");

src/renderer/Metal.zig

@@ -10,6 +10,7 @@ const glfw = @import("glfw");
 const objc = @import("objc");
 const macos = @import("macos");
 const imgui = @import("imgui");
+const glslang = @import("glslang");
 const apprt = @import("../apprt.zig");
 const configpkg = @import("../config.zig");
 const font = @import("../font/main.zig");
@@ -17,13 +18,16 @@ const terminal = @import("../terminal/main.zig");
 const renderer = @import("../renderer.zig");
 const math = @import("../math.zig");
 const Surface = @import("../Surface.zig");
+const shadertoy = @import("shadertoy.zig");
 const assert = std.debug.assert;
 const Allocator = std.mem.Allocator;
+const ArenaAllocator = std.heap.ArenaAllocator;
 const Terminal = terminal.Terminal;
 
 const mtl = @import("metal/api.zig");
 const mtl_buffer = @import("metal/buffer.zig");
 const mtl_image = @import("metal/image.zig");
+const mtl_sampler = @import("metal/sampler.zig");
 const mtl_shaders = @import("metal/shaders.zig");
 const Image = mtl_image.Image;
 const ImageMap = mtl_image.ImageMap;
@@ -76,6 +80,10 @@ background_color: terminal.color.RGB,
 /// by a terminal application
 cursor_color: ?terminal.color.RGB,
 
+/// The current frame background color. This is only updated during
+/// the updateFrame method.
+current_background_color: terminal.color.RGB,
+
 /// The current set of cells to render. This is rebuilt on every frame
 /// but we keep this around so that we don't reallocate. Each set of
 /// cells goes into a separate shader.
@@ -108,12 +116,31 @@ swapchain: objc.Object, // CAMetalLayer
 texture_greyscale: objc.Object, // MTLTexture
 texture_color: objc.Object, // MTLTexture
 
+/// Custom shader state. This is only set if we have custom shaders.
+custom_shader_state: ?CustomShaderState = null,
+
+pub const CustomShaderState = struct {
+    /// The screen texture that we render the terminal to. If we don't have
+    /// custom shaders, we render directly to the drawable.
+    screen_texture: objc.Object, // MTLTexture
+    sampler: mtl_sampler.Sampler,
+    uniforms: mtl_shaders.PostUniforms,
+    last_frame_time: std.time.Instant,
+
+    pub fn deinit(self: *CustomShaderState) void {
+        deinitMTLResource(self.screen_texture);
+        self.sampler.deinit();
+    }
+};
+
 /// The configuration for this renderer that is derived from the main
 /// configuration. This must be exported so that we don't need to
 /// pass around Config pointers which makes memory management a pain.
 pub const DerivedConfig = struct {
+    arena: ArenaAllocator,
+
     font_thicken: bool,
-    font_features: std.ArrayList([]const u8),
+    font_features: std.ArrayListUnmanaged([]const u8),
     font_styles: font.Group.StyleStatus,
     cursor_color: ?terminal.color.RGB,
     cursor_opacity: f64,
@@ -124,17 +151,22 @@ pub const DerivedConfig = struct {
     selection_background: ?terminal.color.RGB,
     selection_foreground: ?terminal.color.RGB,
     invert_selection_fg_bg: bool,
+    custom_shaders: std.ArrayListUnmanaged([]const u8),
+    custom_shader_animation: bool,
 
     pub fn init(
         alloc_gpa: Allocator,
         config: *const configpkg.Config,
     ) !DerivedConfig {
+        var arena = ArenaAllocator.init(alloc_gpa);
+        errdefer arena.deinit();
+        const alloc = arena.allocator();
+
+        // Copy our shaders
+        const custom_shaders = try config.@"custom-shader".value.list.clone(alloc);
+
         // Copy our font features
-        var font_features = features: {
-            var clone = try config.@"font-feature".list.clone(alloc_gpa);
-            break :features clone.toManaged(alloc_gpa);
-        };
-        errdefer font_features.deinit();
+        const font_features = try config.@"font-feature".list.clone(alloc);
 
         // Get our font styles
         var font_styles = font.Group.StyleStatus.initFill(true);
@@ -173,11 +205,16 @@ pub const DerivedConfig = struct {
                 bg.toTerminalRGB()
             else
                 null,
+
+            .custom_shaders = custom_shaders,
+            .custom_shader_animation = config.@"custom-shader-animation",
+
+            .arena = arena,
         };
     }
 
     pub fn deinit(self: *DerivedConfig) void {
-        self.font_features.deinit();
+        self.arena.deinit();
     }
 };
 
@@ -199,11 +236,15 @@ pub fn surfaceInit(surface: *apprt.Surface) !void {
 }
 
 pub fn init(alloc: Allocator, options: renderer.Options) !Metal {
+    var arena = ArenaAllocator.init(alloc);
+    defer arena.deinit();
+    const arena_alloc = arena.allocator();
+
     // Initialize our metal stuff
     const device = objc.Object.fromId(mtl.MTLCreateSystemDefaultDevice());
     const queue = device.msgSend(objc.Object, objc.sel("newCommandQueue"), .{});
     const swapchain = swapchain: {
-        const CAMetalLayer = objc.Class.getClass("CAMetalLayer").?;
+        const CAMetalLayer = objc.getClass("CAMetalLayer").?;
         const swapchain = CAMetalLayer.msgSend(objc.Object, objc.sel("layer"), .{});
         swapchain.setProperty("device", device.value);
         swapchain.setProperty("opaque", options.config.background_opacity >= 1);
@@ -252,9 +293,50 @@ pub fn init(alloc: Allocator, options: renderer.Options) !Metal {
     });
     errdefer buf_instance.deinit();
 
+    // Load our custom shaders
+    const custom_shaders: []const [:0]const u8 = shadertoy.loadFromFiles(
+        arena_alloc,
+        options.config.custom_shaders.items,
+        .msl,
+    ) catch |err| err: {
+        log.warn("error loading custom shaders err={}", .{err});
+        break :err &.{};
+    };
+
+    // If we have custom shaders then setup our state
+    var custom_shader_state: ?CustomShaderState = state: {
+        if (custom_shaders.len == 0) break :state null;
+
+        // Build our sampler for our texture
+        var sampler = try mtl_sampler.Sampler.init(device);
+        errdefer sampler.deinit();
+
+        break :state .{
+            // Resolution and screen texture will be fixed up by first
+            // call to setScreenSize. This happens before any draw call.
+            .screen_texture = undefined,
+            .sampler = sampler,
+            .uniforms = .{
+                .resolution = .{ 0, 0, 1 },
+                .time = 1,
+                .time_delta = 1,
+                .frame_rate = 1,
+                .frame = 1,
+                .channel_time = [1][4]f32{.{ 0, 0, 0, 0 }} ** 4,
+                .channel_resolution = [1][4]f32{.{ 0, 0, 0, 0 }} ** 4,
+                .mouse = .{ 0, 0, 0, 0 },
+                .date = .{ 0, 0, 0, 0 },
+                .sample_rate = 1,
+            },
+
+            .last_frame_time = try std.time.Instant.now(),
+        };
+    };
+    errdefer if (custom_shader_state) |*state| state.deinit();
+
     // Initialize our shaders
-    var shaders = try Shaders.init(device);
-    errdefer shaders.deinit();
+    var shaders = try Shaders.init(alloc, device, custom_shaders);
+    errdefer shaders.deinit(alloc);
 
     // Font atlas textures
     const texture_greyscale = try initAtlasTexture(device, &options.font_group.atlas_greyscale);
@@ -271,6 +353,7 @@ pub fn init(alloc: Allocator, options: renderer.Options) !Metal {
         .foreground_color = options.config.foreground,
         .background_color = options.config.background,
         .cursor_color = options.config.cursor_color,
+        .current_background_color = options.config.background,
 
         // Render state
         .cells_bg = .{},
@@ -298,6 +381,7 @@ pub fn init(alloc: Allocator, options: renderer.Options) !Metal {
         .swapchain = swapchain,
         .texture_greyscale = texture_greyscale,
         .texture_color = texture_color,
+        .custom_shader_state = custom_shader_state,
     };
 }
 
@@ -323,7 +407,9 @@ pub fn deinit(self: *Metal) void {
     deinitMTLResource(self.texture_color);
     self.queue.msgSend(void, objc.sel("release"), .{});
 
-    self.shaders.deinit();
+    if (self.custom_shader_state) |*state| state.deinit();
+
+    self.shaders.deinit(self.alloc);
 
     self.* = undefined;
 }
@@ -384,6 +470,13 @@ pub fn threadExit(self: *const Metal) void {
     // Metal requires no per-thread state.
 }
 
+/// True if our renderer has animations so that a higher frequency
+/// timer is used.
+pub fn hasAnimations(self: *const Metal) bool {
+    return self.custom_shader_state != null and
+        self.config.custom_shader_animation;
+}
+
 /// Returns the grid size for a given screen size. This is safe to call
 /// on any thread.
 fn gridSize(self: *Metal) ?renderer.GridSize {
@@ -448,8 +541,8 @@ pub fn setFontSize(self: *Metal, size: font.face.DesiredSize) !void {
     }, .{ .forever = {} });
 }
 
-/// The primary render callback that is completely thread-safe.
-pub fn render(
+/// Update the frame data.
+pub fn updateFrame(
     self: *Metal,
     surface: *apprt.Surface,
     state: *renderer.State,
@@ -535,10 +628,6 @@ pub fn render(
     };
     defer critical.screen.deinit();
 
-    // @autoreleasepool {}
-    const pool = objc.AutoreleasePool.init();
-    defer pool.deinit();
-
     // Build our GPU cells
     try self.rebuildCells(
         critical.selection,
@@ -547,18 +636,8 @@ pub fn render(
         critical.cursor_style,
     );
 
-    // Get our drawable (CAMetalDrawable)
-    const drawable = self.swapchain.msgSend(objc.Object, objc.sel("nextDrawable"), .{});
-
-    // If our font atlas changed, sync the texture data
-    if (self.font_group.atlas_greyscale.modified) {
-        try syncAtlasTexture(self.device, &self.font_group.atlas_greyscale, &self.texture_greyscale);
-        self.font_group.atlas_greyscale.modified = false;
-    }
-    if (self.font_group.atlas_color.modified) {
-        try syncAtlasTexture(self.device, &self.font_group.atlas_color, &self.texture_color);
-        self.font_group.atlas_color.modified = false;
-    }
+    // Update our background color
+    self.current_background_color = critical.bg;
 
     // Go through our images and see if we need to setup any textures.
     {
@@ -580,6 +659,45 @@ pub fn render(
             }
         }
     }
+}
+
+/// Draw the frame to the screen.
+pub fn drawFrame(self: *Metal, surface: *apprt.Surface) !void {
+    _ = surface;
+
+    // If we have custom shaders, update the animation time.
+    if (self.custom_shader_state) |*state| {
+        const now = std.time.Instant.now() catch state.last_frame_time;
+        const since_ns: f32 = @floatFromInt(now.since(state.last_frame_time));
+        state.uniforms.time = since_ns / std.time.ns_per_s;
+        state.uniforms.time_delta = since_ns / std.time.ns_per_s;
+    }
+
+    // @autoreleasepool {}
+    const pool = objc.AutoreleasePool.init();
+    defer pool.deinit();
+
+    // Get our drawable (CAMetalDrawable)
+    const drawable = self.swapchain.msgSend(objc.Object, objc.sel("nextDrawable"), .{});
+
+    // Get our screen texture. If we don't have a dedicated screen texture
+    // then we just use the drawable texture.
+    const screen_texture = if (self.custom_shader_state) |state|
+        state.screen_texture
+    else tex: {
+        const texture = drawable.msgSend(objc.c.id, objc.sel("texture"), .{});
+        break :tex objc.Object.fromId(texture);
+    };
+
+    // If our font atlas changed, sync the texture data
+    if (self.font_group.atlas_greyscale.modified) {
+        try syncAtlasTexture(self.device, &self.font_group.atlas_greyscale, &self.texture_greyscale);
+        self.font_group.atlas_greyscale.modified = false;
+    }
+    if (self.font_group.atlas_color.modified) {
+        try syncAtlasTexture(self.device, &self.font_group.atlas_color, &self.texture_color);
+        self.font_group.atlas_color.modified = false;
+    }
 
     // Command buffer (MTLCommandBuffer)
     const buffer = self.queue.msgSend(objc.Object, objc.sel("commandBuffer"), .{});
@@ -587,7 +705,7 @@ pub fn render(
     {
         // MTLRenderPassDescriptor
         const desc = desc: {
-            const MTLRenderPassDescriptor = objc.Class.getClass("MTLRenderPassDescriptor").?;
+            const MTLRenderPassDescriptor = objc.getClass("MTLRenderPassDescriptor").?;
             const desc = MTLRenderPassDescriptor.msgSend(
                 objc.Object,
                 objc.sel("renderPassDescriptor"),
@@ -607,14 +725,14 @@ pub fn render(
                 // Ghostty in XCode in debug mode it returns a CaptureMTLDrawable
                 // which ironically doesn't implement CAMetalDrawable as a
                 // property so we just send a message.
-                const texture = drawable.msgSend(objc.c.id, objc.sel("texture"), .{});
+                //const texture = drawable.msgSend(objc.c.id, objc.sel("texture"), .{});
                 attachment.setProperty("loadAction", @intFromEnum(mtl.MTLLoadAction.clear));
                 attachment.setProperty("storeAction", @intFromEnum(mtl.MTLStoreAction.store));
-                attachment.setProperty("texture", texture);
+                attachment.setProperty("texture", screen_texture.value);
                 attachment.setProperty("clearColor", mtl.MTLClearColor{
-                    .red = @as(f32, @floatFromInt(critical.bg.r)) / 255,
-                    .green = @as(f32, @floatFromInt(critical.bg.g)) / 255,
-                    .blue = @as(f32, @floatFromInt(critical.bg.b)) / 255,
+                    .red = @as(f32, @floatFromInt(self.current_background_color.r)) / 255,
+                    .green = @as(f32, @floatFromInt(self.current_background_color.g)) / 255,
+                    .blue = @as(f32, @floatFromInt(self.current_background_color.b)) / 255,
                     .alpha = self.config.background_opacity,
                 });
             }
@@ -646,10 +764,118 @@ pub fn render(
         try self.drawImagePlacements(encoder, self.image_placements.items[self.image_text_end..]);
     }
 
+    // If we have custom shaders AND we have a screen texture, then we
+    // render the custom shaders.
+    if (self.custom_shader_state) |state| {
+        // MTLRenderPassDescriptor
+        const desc = desc: {
+            const MTLRenderPassDescriptor = objc.getClass("MTLRenderPassDescriptor").?;
+            const desc = MTLRenderPassDescriptor.msgSend(
+                objc.Object,
+                objc.sel("renderPassDescriptor"),
+                .{},
+            );
+
+            // Set our color attachment to be our drawable surface.
+            const attachments = objc.Object.fromId(desc.getProperty(?*anyopaque, "colorAttachments"));
+            {
+                const attachment = attachments.msgSend(
+                    objc.Object,
+                    objc.sel("objectAtIndexedSubscript:"),
+                    .{@as(c_ulong, 0)},
+                );
+
+                // Texture is a property of CAMetalDrawable but if you run
+                // Ghostty in XCode in debug mode it returns a CaptureMTLDrawable
+                // which ironically doesn't implement CAMetalDrawable as a
+                // property so we just send a message.
+                const texture = drawable.msgSend(objc.c.id, objc.sel("texture"), .{});
+                attachment.setProperty("loadAction", @intFromEnum(mtl.MTLLoadAction.clear));
+                attachment.setProperty("storeAction", @intFromEnum(mtl.MTLStoreAction.store));
+                attachment.setProperty("texture", texture);
+                attachment.setProperty("clearColor", mtl.MTLClearColor{
+                    .red = 0,
+                    .green = 0,
+                    .blue = 0,
+                    .alpha = 1,
+                });
+            }
+
+            break :desc desc;
+        };
+
+        // MTLRenderCommandEncoder
+        const encoder = buffer.msgSend(
+            objc.Object,
+            objc.sel("renderCommandEncoderWithDescriptor:"),
+            .{desc.value},
+        );
+        defer encoder.msgSend(void, objc.sel("endEncoding"), .{});
+
+        for (self.shaders.post_pipelines) |pipeline| {
+            try self.drawPostShader(encoder, pipeline, &state);
+        }
+    }
+
     buffer.msgSend(void, objc.sel("presentDrawable:"), .{drawable.value});
     buffer.msgSend(void, objc.sel("commit"), .{});
 }
 
+fn drawPostShader(
+    self: *Metal,
+    encoder: objc.Object,
+    pipeline: objc.Object,
+    state: *const CustomShaderState,
+) !void {
+    _ = self;
+
+    // Use our custom shader pipeline
+    encoder.msgSend(
+        void,
+        objc.sel("setRenderPipelineState:"),
+        .{pipeline.value},
+    );
+
+    // Set our sampler
+    encoder.msgSend(
+        void,
+        objc.sel("setFragmentSamplerState:atIndex:"),
+        .{ state.sampler.sampler.value, @as(c_ulong, 0) },
+    );
+
+    // Set our uniforms
+    encoder.msgSend(
+        void,
+        objc.sel("setFragmentBytes:length:atIndex:"),
+        .{
+            @as(*const anyopaque, @ptrCast(&state.uniforms)),
+            @as(c_ulong, @sizeOf(@TypeOf(state.uniforms))),
+            @as(c_ulong, 0),
+        },
+    );
+
+    // Screen texture
+    encoder.msgSend(
+        void,
+        objc.sel("setFragmentTexture:atIndex:"),
+        .{
+            state.screen_texture.value,
+            @as(c_ulong, 0),
+        },
+    );
+
+    // Draw!
+    encoder.msgSend(
+        void,
+        objc.sel("drawPrimitives:vertexStart:vertexCount:"),
+        .{
+            @intFromEnum(mtl.MTLPrimitiveType.triangle_strip),
+            @as(c_ulong, 0),
+            @as(c_ulong, 4),
+        },
+    );
+}
+
 fn drawImagePlacements(
     self: *Metal,
     encoder: objc.Object,
@@ -1070,6 +1296,51 @@ pub fn setScreenSize(
     self.cells.clearAndFree(self.alloc);
     self.cells_bg.clearAndFree(self.alloc);
 
+    // If we have custom shaders then we update the state
+    if (self.custom_shader_state) |*state| {
+        // Only free our previous texture if this isn't our first
+        // time setting the custom shader state.
+        if (state.uniforms.resolution[0] > 0) {
+            deinitMTLResource(state.screen_texture);
+        }
+
+        state.uniforms.resolution = .{
+            @floatFromInt(dim.width),
+            @floatFromInt(dim.height),
+            1,
+        };
+
+        state.screen_texture = screen_texture: {
+            // This texture is the size of our drawable but supports being a
+            // render target AND reading so that the custom shaders can read from it.
+            const desc = init: {
+                const Class = objc.getClass("MTLTextureDescriptor").?;
+                const id_alloc = Class.msgSend(objc.Object, objc.sel("alloc"), .{});
+                const id_init = id_alloc.msgSend(objc.Object, objc.sel("init"), .{});
+                break :init id_init;
+            };
+            desc.setProperty("pixelFormat", @intFromEnum(mtl.MTLPixelFormat.bgra8unorm));
+            desc.setProperty("width", @as(c_ulong, @intCast(dim.width)));
+            desc.setProperty("height", @as(c_ulong, @intCast(dim.height)));
+            desc.setProperty(
+                "usage",
+                @intFromEnum(mtl.MTLTextureUsage.render_target) |
+                    @intFromEnum(mtl.MTLTextureUsage.shader_read) |
+                    @intFromEnum(mtl.MTLTextureUsage.shader_write),
+            );
+
+            // If we fail to create the texture, then we just don't have a screen
+            // texture and our custom shaders won't run.
+            const id = self.device.msgSend(
+                ?*anyopaque,
+                objc.sel("newTextureWithDescriptor:"),
+                .{desc},
+            ) orelse return error.MetalFailed;
+
+            break :screen_texture objc.Object.fromId(id);
+        };
+    }
+
     log.debug("screen size screen={} grid={}, cell={}", .{ dim, grid_size, self.cell_size });
 }
 
@@ -1625,7 +1896,7 @@ fn initAtlasTexture(device: objc.Object, atlas: *const font.Atlas) !objc.Object
 
     // Create our descriptor
     const desc = init: {
-        const Class = objc.Class.getClass("MTLTextureDescriptor").?;
+        const Class = objc.getClass("MTLTextureDescriptor").?;
         const id_alloc = Class.msgSend(objc.Object, objc.sel("alloc"), .{});
         const id_init = id_alloc.msgSend(objc.Object, objc.sel("init"), .{});
         break :init id_init;

src/renderer/OpenGL.zig

@@ -7,6 +7,8 @@ const glfw = @import("glfw");
 const assert = std.debug.assert;
 const testing = std.testing;
 const Allocator = std.mem.Allocator;
+const ArenaAllocator = std.heap.ArenaAllocator;
+const shadertoy = @import("shadertoy.zig");
 const apprt = @import("../apprt.zig");
 const configpkg = @import("../config.zig");
 const font = @import("../font/main.zig");
@@ -14,11 +16,14 @@ const imgui = @import("imgui");
 const renderer = @import("../renderer.zig");
 const terminal = @import("../terminal/main.zig");
 const Terminal = terminal.Terminal;
-const gl = @import("opengl/main.zig");
+const gl = @import("opengl");
 const trace = @import("tracy").trace;
 const math = @import("../math.zig");
 const Surface = @import("../Surface.zig");
 
+const CellProgram = @import("opengl/CellProgram.zig");
+const custom = @import("opengl/custom.zig");
+
 const log = std.log.scoped(.grid);
 
 /// The runtime can request a single-threaded draw by setting this boolean
@@ -45,8 +50,8 @@ screen_size: ?renderer.ScreenSize,
 
 /// The current set of cells to render. Each set of cells goes into
 /// a separate shader call.
-cells_bg: std.ArrayListUnmanaged(GPUCell),
-cells: std.ArrayListUnmanaged(GPUCell),
+cells_bg: std.ArrayListUnmanaged(CellProgram.Cell),
+cells: std.ArrayListUnmanaged(CellProgram.Cell),
 
 /// The size of the cells list that was sent to the GPU. This is used
 /// to detect when the cells array was reallocated/resized and handle that
@@ -102,8 +107,11 @@ draw_background: terminal.color.RGB,
 const SetScreenSize = struct {
     size: renderer.ScreenSize,
 
-    fn apply(self: SetScreenSize, r: *const OpenGL) !void {
-        const gl_state = r.gl_state orelse return error.OpenGLUninitialized;
+    fn apply(self: SetScreenSize, r: *OpenGL) !void {
+        const gl_state: *GLState = if (r.gl_state) |*v|
+            v
+        else
+            return error.OpenGLUninitialized;
 
         // Apply our padding
         const padding = if (r.padding.balance)
@@ -130,7 +138,7 @@ const SetScreenSize = struct {
         );
 
         // Update the projection uniform within our shader
-        try gl_state.program.setUniform(
+        try gl_state.cell_program.program.setUniform(
             "projection",
 
             // 2D orthographic projection with the full w/h
@@ -141,6 +149,11 @@ const SetScreenSize = struct {
                 -1 * @as(f32, @floatFromInt(padding.top)),
             ),
         );
+
+        // Update our custom shader resolution
+        if (gl_state.custom) |*custom_state| {
+            try custom_state.setScreenSize(self.size);
+        }
     }
 };
 
@@ -150,84 +163,32 @@ const SetFontSize = struct {
     fn apply(self: SetFontSize, r: *const OpenGL) !void {
         const gl_state = r.gl_state orelse return error.OpenGLUninitialized;
 
-        try gl_state.program.setUniform(
+        try gl_state.cell_program.program.setUniform(
             "cell_size",
             @Vector(2, f32){
                 @floatFromInt(self.metrics.cell_width),
                 @floatFromInt(self.metrics.cell_height),
             },
         );
-        try gl_state.program.setUniform(
+        try gl_state.cell_program.program.setUniform(
             "strikethrough_position",
             @as(f32, @floatFromInt(self.metrics.strikethrough_position)),
         );
-        try gl_state.program.setUniform(
+        try gl_state.cell_program.program.setUniform(
             "strikethrough_thickness",
             @as(f32, @floatFromInt(self.metrics.strikethrough_thickness)),
         );
     }
 };
 
-/// The raw structure that maps directly to the buffer sent to the vertex shader.
-/// This must be "extern" so that the field order is not reordered by the
-/// Zig compiler.
-const GPUCell = extern struct {
-    /// vec2 grid_coord
-    grid_col: u16,
-    grid_row: u16,
-
-    /// vec2 glyph_pos
-    glyph_x: u32 = 0,
-    glyph_y: u32 = 0,
-
-    /// vec2 glyph_size
-    glyph_width: u32 = 0,
-    glyph_height: u32 = 0,
-
-    /// vec2 glyph_size
-    glyph_offset_x: i32 = 0,
-    glyph_offset_y: i32 = 0,
-
-    /// vec4 fg_color_in
-    fg_r: u8,
-    fg_g: u8,
-    fg_b: u8,
-    fg_a: u8,
-
-    /// vec4 bg_color_in
-    bg_r: u8,
-    bg_g: u8,
-    bg_b: u8,
-    bg_a: u8,
-
-    /// uint mode
-    mode: GPUCellMode,
-
-    /// The width in grid cells that a rendering takes.
-    grid_width: u8,
-};
-
-const GPUCellMode = enum(u8) {
-    bg = 1,
-    fg = 2,
-    fg_color = 7,
-    strikethrough = 8,
-
-    // Non-exhaustive because masks change it
-    _,
-
-    /// Apply a mask to the mode.
-    pub fn mask(self: GPUCellMode, m: GPUCellMode) GPUCellMode {
-        return @enumFromInt(@intFromEnum(self) | @intFromEnum(m));
-    }
-};
-
 /// The configuration for this renderer that is derived from the main
 /// configuration. This must be exported so that we don't need to
 /// pass around Config pointers which makes memory management a pain.
 pub const DerivedConfig = struct {
+    arena: ArenaAllocator,
+
     font_thicken: bool,
-    font_features: std.ArrayList([]const u8),
+    font_features: std.ArrayListUnmanaged([]const u8),
     font_styles: font.Group.StyleStatus,
     cursor_color: ?terminal.color.RGB,
     cursor_text: ?terminal.color.RGB,
@@ -238,17 +199,22 @@ pub const DerivedConfig = struct {
     selection_background: ?terminal.color.RGB,
     selection_foreground: ?terminal.color.RGB,
     invert_selection_fg_bg: bool,
+    custom_shaders: std.ArrayListUnmanaged([]const u8),
+    custom_shader_animation: bool,
 
     pub fn init(
         alloc_gpa: Allocator,
         config: *const configpkg.Config,
     ) !DerivedConfig {
+        var arena = ArenaAllocator.init(alloc_gpa);
+        errdefer arena.deinit();
+        const alloc = arena.allocator();
+
+        // Copy our shaders
+        const custom_shaders = try config.@"custom-shader".value.list.clone(alloc);
+
         // Copy our font features
-        var font_features = features: {
-            var clone = try config.@"font-feature".list.clone(alloc_gpa);
-            break :features clone.toManaged(alloc_gpa);
-        };
-        errdefer font_features.deinit();
+        const font_features = try config.@"font-feature".list.clone(alloc);
 
         // Get our font styles
         var font_styles = font.Group.StyleStatus.initFill(true);
@@ -287,11 +253,16 @@ pub const DerivedConfig = struct {
                 bg.toTerminalRGB()
             else
                 null,
+
+            .custom_shaders = custom_shaders,
+            .custom_shader_animation = config.@"custom-shader-animation",
+
+            .arena = arena,
         };
     }
 
     pub fn deinit(self: *DerivedConfig) void {
-        self.font_features.deinit();
+        self.arena.deinit();
     }
 };
 
@@ -309,7 +280,7 @@ pub fn init(alloc: Allocator, options: renderer.Options) !OpenGL {
         options.config.font_thicken,
     );
 
-    var gl_state = try GLState.init(options.font_group);
+    var gl_state = try GLState.init(alloc, options.config, options.font_group);
     errdefer gl_state.deinit();
 
     return OpenGL{
@@ -336,7 +307,7 @@ pub fn init(alloc: Allocator, options: renderer.Options) !OpenGL {
 pub fn deinit(self: *OpenGL) void {
     self.font_shaper.deinit();
 
-    if (self.gl_state) |*v| v.deinit();
+    if (self.gl_state) |*v| v.deinit(self.alloc);
 
     self.cells.deinit(self.alloc);
     self.cells_bg.deinit(self.alloc);
@@ -410,7 +381,7 @@ pub fn displayUnrealized(self: *OpenGL) void {
     defer if (single_threaded_draw) self.draw_mutex.unlock();
 
     if (self.gl_state) |*v| {
-        v.deinit();
+        v.deinit(self.alloc);
         self.gl_state = null;
     }
 }
@@ -428,11 +399,11 @@ pub fn displayRealize(self: *OpenGL) !void {
     );
 
     // Make our new state
-    var gl_state = try GLState.init(self.font_group);
+    var gl_state = try GLState.init(self.alloc, self.config, self.font_group);
     errdefer gl_state.deinit();
 
     // Unrealize if we have to
-    if (self.gl_state) |*v| v.deinit();
+    if (self.gl_state) |*v| v.deinit(self.alloc);
 
     // Set our new state
     self.gl_state = gl_state;
@@ -506,6 +477,13 @@ pub fn threadExit(self: *const OpenGL) void {
     }
 }
 
+/// True if our renderer has animations so that a higher frequency
+/// timer is used.
+pub fn hasAnimations(self: *const OpenGL) bool {
+    const state = self.gl_state orelse return false;
+    return state.custom != null and self.config.custom_shader_animation;
+}
+
 /// Callback when the focus changes for the terminal this is rendering.
 ///
 /// Must be called on the render thread.
@@ -576,12 +554,14 @@ fn resetFontMetrics(
 }
 
 /// The primary render callback that is completely thread-safe.
-pub fn render(
+pub fn updateFrame(
     self: *OpenGL,
     surface: *apprt.Surface,
     state: *renderer.State,
     cursor_blink_visible: bool,
 ) !void {
+    _ = surface;
+
     // Data we extract out of the critical area.
     const Critical = struct {
         gl_bg: terminal.color.RGB,
@@ -669,19 +649,6 @@ pub fn render(
             critical.cursor_style,
         );
     }
-
-    // We're out of the critical path now. Let's render. We only render if
-    // we're not single threaded. If we're single threaded we expect the
-    // runtime to call draw.
-    if (single_threaded_draw) return;
-
-    try self.draw();
-
-    // Swap our window buffers
-    switch (apprt.runtime) {
-        else => @compileError("unsupported runtime"),
-        apprt.glfw => surface.window.swapBuffers(),
-    }
 }
 
 /// rebuildCells rebuilds all the GPU cells from our CPU state. This is a
@@ -735,7 +702,7 @@ pub fn rebuildCells(
     // This is the cell that has [mode == .fg] and is underneath our cursor.
     // We keep track of it so that we can invert the colors so the character
     // remains visible.
-    var cursor_cell: ?GPUCell = null;
+    var cursor_cell: ?CellProgram.Cell = null;
 
     // Build each cell
     var rowIter = screen.rowIterator(.viewport);
@@ -868,15 +835,15 @@ pub fn rebuildCells(
         if (cursor_cell) |*cell| {
             if (cell.mode == .fg) {
                 if (self.config.cursor_text) |txt| {
-                    cell.fg_r = txt.r;
-                    cell.fg_g = txt.g;
-                    cell.fg_b = txt.b;
-                    cell.fg_a = 255;
+                    cell.r = txt.r;
+                    cell.g = txt.g;
+                    cell.b = txt.b;
+                    cell.a = 255;
                 } else {
-                    cell.fg_r = 0;
-                    cell.fg_g = 0;
-                    cell.fg_b = 0;
-                    cell.fg_a = 255;
+                    cell.r = 0;
+                    cell.g = 0;
+                    cell.b = 0;
+                    cell.a = 255;
                 }
             }
             self.cells.appendAssumeCapacity(cell.*);
@@ -940,14 +907,10 @@ fn addPreeditCell(
         .glyph_height = 0,
         .glyph_offset_x = 0,
         .glyph_offset_y = 0,
-        .fg_r = 0,
-        .fg_g = 0,
-        .fg_b = 0,
-        .fg_a = 0,
-        .bg_r = bg.r,
-        .bg_g = bg.g,
-        .bg_b = bg.b,
-        .bg_a = 255,
+        .r = bg.r,
+        .g = bg.g,
+        .b = bg.b,
+        .a = 255,
     });
 
     // Add our text
@@ -962,14 +925,10 @@ fn addPreeditCell(
         .glyph_height = glyph.height,
         .glyph_offset_x = glyph.offset_x,
         .glyph_offset_y = glyph.offset_y,
-        .fg_r = fg.r,
-        .fg_g = fg.g,
-        .fg_b = fg.b,
-        .fg_a = 255,
-        .bg_r = 0,
-        .bg_g = 0,
-        .bg_b = 0,
-        .bg_a = 0,
+        .r = fg.r,
+        .g = fg.g,
+        .b = fg.b,
+        .a = 255,
     });
 }
 
@@ -977,7 +936,7 @@ fn addCursor(
     self: *OpenGL,
     screen: *terminal.Screen,
     cursor_style: renderer.CursorStyle,
-) ?*const GPUCell {
+) ?*const CellProgram.Cell {
     // Add the cursor. We render the cursor over the wide character if
     // we're on the wide characer tail.
     const wide, const x = cell: {
@@ -1027,14 +986,10 @@ fn addCursor(
         .grid_col = @intCast(x),
         .grid_row = @intCast(screen.cursor.y),
         .grid_width = if (wide) 2 else 1,
-        .fg_r = color.r,
-        .fg_g = color.g,
-        .fg_b = color.b,
-        .fg_a = alpha,
-        .bg_r = 0,
-        .bg_g = 0,
-        .bg_b = 0,
-        .bg_a = 0,
+        .r = color.r,
+        .g = color.g,
+        .b = color.b,
+        .a = alpha,
         .glyph_x = glyph.atlas_x,
         .glyph_y = glyph.atlas_y,
         .glyph_width = glyph.width,
@@ -1182,14 +1137,10 @@ pub fn updateCell(
             .glyph_height = 0,
             .glyph_offset_x = 0,
             .glyph_offset_y = 0,
-            .fg_r = 0,
-            .fg_g = 0,
-            .fg_b = 0,
-            .fg_a = 0,
-            .bg_r = rgb.r,
-            .bg_g = rgb.g,
-            .bg_b = rgb.b,
-            .bg_a = bg_alpha,
+            .r = rgb.r,
+            .g = rgb.g,
+            .b = rgb.b,
+            .a = bg_alpha,
         });
     }
 
@@ -1208,7 +1159,7 @@ pub fn updateCell(
 
         // If we're rendering a color font, we use the color atlas
         const presentation = try self.font_group.group.presentationFromIndex(shaper_run.font_index);
-        const mode: GPUCellMode = switch (presentation) {
+        const mode: CellProgram.CellMode = switch (presentation) {
             .text => .fg,
             .emoji => .fg_color,
         };
@@ -1224,14 +1175,10 @@ pub fn updateCell(
             .glyph_height = glyph.height,
             .glyph_offset_x = glyph.offset_x,
             .glyph_offset_y = glyph.offset_y,
-            .fg_r = colors.fg.r,
-            .fg_g = colors.fg.g,
-            .fg_b = colors.fg.b,
-            .fg_a = alpha,
-            .bg_r = 0,
-            .bg_g = 0,
-            .bg_b = 0,
-            .bg_a = 0,
+            .r = colors.fg.r,
+            .g = colors.fg.g,
+            .b = colors.fg.b,
+            .a = alpha,
         });
     }
 
@@ -1265,14 +1212,10 @@ pub fn updateCell(
             .glyph_height = underline_glyph.height,
             .glyph_offset_x = underline_glyph.offset_x,
             .glyph_offset_y = underline_glyph.offset_y,
-            .fg_r = color.r,
-            .fg_g = color.g,
-            .fg_b = color.b,
-            .fg_a = alpha,
-            .bg_r = 0,
-            .bg_g = 0,
-            .bg_b = 0,
-            .bg_a = 0,
+            .r = color.r,
+            .g = color.g,
+            .b = color.b,
+            .a = alpha,
         });
     }
 
@@ -1288,14 +1231,10 @@ pub fn updateCell(
             .glyph_height = 0,
             .glyph_offset_x = 0,
             .glyph_offset_y = 0,
-            .fg_r = colors.fg.r,
-            .fg_g = colors.fg.g,
-            .fg_b = colors.fg.b,
-            .fg_a = alpha,
-            .bg_r = 0,
-            .bg_g = 0,
-            .bg_b = 0,
-            .bg_a = 0,
+            .r = colors.fg.r,
+            .g = colors.fg.g,
+            .b = colors.fg.b,
+            .a = alpha,
         });
     }
 
@@ -1388,11 +1327,11 @@ fn flushAtlas(self: *OpenGL) !void {
                 atlas.resized = false;
                 try texbind.image2D(
                     0,
-                    .Red,
+                    .red,
                     @intCast(atlas.size),
                     @intCast(atlas.size),
                     0,
-                    .Red,
+                    .red,
                     .UnsignedByte,
                     atlas.data.ptr,
                 );
@@ -1403,7 +1342,7 @@ fn flushAtlas(self: *OpenGL) !void {
                     0,
                     @intCast(atlas.size),
                     @intCast(atlas.size),
-                    .Red,
+                    .red,
                     .UnsignedByte,
                     atlas.data.ptr,
                 );
@@ -1422,11 +1361,11 @@ fn flushAtlas(self: *OpenGL) !void {
                 atlas.resized = false;
                 try texbind.image2D(
                     0,
-                    .RGBA,
+                    .rgba,
                     @intCast(atlas.size),
                     @intCast(atlas.size),
                     0,
-                    .BGRA,
+                    .bgra,
                     .UnsignedByte,
                     atlas.data.ptr,
                 );
@@ -1437,7 +1376,7 @@ fn flushAtlas(self: *OpenGL) !void {
                     0,
                     @intCast(atlas.size),
                     @intCast(atlas.size),
-                    .BGRA,
+                    .bgra,
                     .UnsignedByte,
                     atlas.data.ptr,
                 );
@@ -1448,19 +1387,71 @@ fn flushAtlas(self: *OpenGL) !void {
 
 /// Render renders the current cell state. This will not modify any of
 /// the cells.
-pub fn draw(self: *OpenGL) !void {
+pub fn drawFrame(self: *OpenGL, surface: *apprt.Surface) !void {
     const t = trace(@src());
     defer t.end();
 
     // If we're in single-threaded more we grab a lock since we use shared data.
     if (single_threaded_draw) self.draw_mutex.lock();
     defer if (single_threaded_draw) self.draw_mutex.unlock();
-    const gl_state = self.gl_state orelse return;
+    const gl_state: *GLState = if (self.gl_state) |*v| v else return;
 
+    // Draw our terminal cells
+    try self.drawCellProgram(gl_state);
+
+    // Draw our custom shaders
+    if (gl_state.custom) |*custom_state| {
+        try self.drawCustomPrograms(custom_state);
+    }
+
+    // Swap our window buffers
+    switch (apprt.runtime) {
+        apprt.glfw => surface.window.swapBuffers(),
+        apprt.gtk => {},
+        else => @compileError("unsupported runtime"),
+    }
+}
+
+/// Draw the custom shaders.
+fn drawCustomPrograms(
+    self: *OpenGL,
+    custom_state: *custom.State,
+) !void {
+    _ = self;
+
+    // Bind our state that is global to all custom shaders
+    const custom_bind = try custom_state.bind();
+    defer custom_bind.unbind();
+
+    // Setup the new frame
+    try custom_state.newFrame();
+
+    // Go through each custom shader and draw it.
+    for (custom_state.programs) |program| {
+        // Bind our cell program state, buffers
+        const bind = try program.bind();
+        defer bind.unbind();
+        try bind.draw();
+    }
+}
+
+/// Runs the cell program (shaders) to draw the terminal grid.
+fn drawCellProgram(
+    self: *OpenGL,
+    gl_state: *const GLState,
+) !void {
     // Try to flush our atlas, this will only do something if there
     // are changes to the atlas.
     try self.flushAtlas();
 
+    // If we have custom shaders, then we draw to the custom
+    // shader framebuffer.
+    const fbobind: ?gl.Framebuffer.Binding = fbobind: {
+        const state = gl_state.custom orelse break :fbobind null;
+        break :fbobind try state.fbo.bind(.framebuffer);
+    };
+    defer if (fbobind) |v| v.unbind();
+
     // Clear the surface
     gl.clearColor(
         @as(f32, @floatFromInt(self.draw_background.r)) / 255,
@@ -1470,17 +1461,9 @@ pub fn draw(self: *OpenGL) !void {
     );
     gl.clear(gl.c.GL_COLOR_BUFFER_BIT);
 
-    // Setup our VAO
-    try gl_state.vao.bind();
-    defer gl.VertexArray.unbind() catch null;
-
-    // Bind EBO
-    var ebobind = try gl_state.ebo.bind(.ElementArrayBuffer);
-    defer ebobind.unbind();
-
-    // Bind VBO and set data
-    var binding = try gl_state.vbo.bind(.ArrayBuffer);
-    defer binding.unbind();
+    // Bind our cell program state, buffers
+    const bind = try gl_state.cell_program.bind();
+    defer bind.unbind();
 
     // Bind our textures
     try gl.Texture.active(gl.c.GL_TEXTURE0);
@@ -1491,10 +1474,6 @@ pub fn draw(self: *OpenGL) !void {
     var texbind1 = try gl_state.texture_color.bind(.@"2D");
     defer texbind1.unbind();
 
-    // Pick our shader to use
-    const pbind = try gl_state.program.use();
-    defer pbind.unbind();
-
     // If we have deferred operations, run them.
     if (self.deferred_screen_size) |v| {
         try v.apply(self);
@@ -1505,8 +1484,9 @@ pub fn draw(self: *OpenGL) !void {
         self.deferred_font_size = null;
     }
 
-    try self.drawCells(binding, self.cells_bg);
-    try self.drawCells(binding, self.cells);
+    // Draw our background, then draw the fg on top of it.
+    try self.drawCells(bind.vbo, self.cells_bg);
+    try self.drawCells(bind.vbo, self.cells);
 }
 
 /// Loads some set of cell data into our buffer and issues a draw call.
@@ -1517,7 +1497,7 @@ pub fn draw(self: *OpenGL) !void {
 fn drawCells(
     self: *OpenGL,
     binding: gl.Buffer.Binding,
-    cells: std.ArrayListUnmanaged(GPUCell),
+    cells: std.ArrayListUnmanaged(CellProgram.Cell),
 ) !void {
     // If we have no cells to render, then we render nothing.
     if (cells.items.len == 0) return;
@@ -1534,8 +1514,8 @@ fn drawCells(
         });
 
         try binding.setDataNullManual(
-            @sizeOf(GPUCell) * cells.capacity,
-            .StaticDraw,
+            @sizeOf(CellProgram.Cell) * cells.capacity,
+            .static_draw,
         );
 
         self.gl_cells_size = cells.capacity;
@@ -1546,7 +1526,7 @@ fn drawCells(
     if (self.gl_cells_written < cells.items.len) {
         const data = cells.items[self.gl_cells_written..];
         // log.info("sending {} cells to GPU", .{data.len});
-        try binding.setSubData(self.gl_cells_written * @sizeOf(GPUCell), data);
+        try binding.setSubData(self.gl_cells_written * @sizeOf(CellProgram.Cell), data);
 
         self.gl_cells_written += data.len;
         assert(data.len > 0);
@@ -1565,14 +1545,41 @@ fn drawCells(
 /// easy to create/destroy these as a set in situations i.e. where the
 /// OpenGL context is replaced.
 const GLState = struct {
-    program: gl.Program,
-    vao: gl.VertexArray,
-    ebo: gl.Buffer,
-    vbo: gl.Buffer,
+    cell_program: CellProgram,
     texture: gl.Texture,
     texture_color: gl.Texture,
+    custom: ?custom.State,
+
+    pub fn init(
+        alloc: Allocator,
+        config: DerivedConfig,
+        font_group: *font.GroupCache,
+    ) !GLState {
+        var arena = ArenaAllocator.init(alloc);
+        defer arena.deinit();
+        const arena_alloc = arena.allocator();
+
+        // Load our custom shaders
+        const custom_state: ?custom.State = custom: {
+            const shaders: []const [:0]const u8 = shadertoy.loadFromFiles(
+                arena_alloc,
+                config.custom_shaders.items,
+                .glsl,
+            ) catch |err| err: {
+                log.warn("error loading custom shaders err={}", .{err});
+                break :err &.{};
+            };
+            if (shaders.len == 0) break :custom null;
+
+            break :custom custom.State.init(
+                alloc,
+                shaders,
+            ) catch |err| err: {
+                log.warn("error initializing custom shaders err={}", .{err});
+                break :err null;
+            };
+        };
 
-    pub fn init(font_group: *font.GroupCache) !GLState {
         // Blending for text. We use GL_ONE here because we should be using
         // premultiplied alpha for all our colors in our fragment shaders.
         // This avoids having a blurry border where transparency is expected on
@@ -1580,74 +1587,6 @@ const GLState = struct {
         try gl.enable(gl.c.GL_BLEND);
         try gl.blendFunc(gl.c.GL_ONE, gl.c.GL_ONE_MINUS_SRC_ALPHA);
 
-        // Shader
-        const program = try gl.Program.createVF(
-            @embedFile("shaders/cell.v.glsl"),
-            @embedFile("shaders/cell.f.glsl"),
-        );
-
-        // Set our cell dimensions
-        const pbind = try program.use();
-        defer pbind.unbind();
-
-        // Set all of our texture indexes
-        try program.setUniform("text", 0);
-        try program.setUniform("text_color", 1);
-
-        // Setup our VAO
-        const vao = try gl.VertexArray.create();
-        errdefer vao.destroy();
-        try vao.bind();
-        defer gl.VertexArray.unbind() catch null;
-
-        // Element buffer (EBO)
-        const ebo = try gl.Buffer.create();
-        errdefer ebo.destroy();
-        var ebobind = try ebo.bind(.ElementArrayBuffer);
-        defer ebobind.unbind();
-        try ebobind.setData([6]u8{
-            0, 1, 3, // Top-left triangle
-            1, 2, 3, // Bottom-right triangle
-        }, .StaticDraw);
-
-        // Vertex buffer (VBO)
-        const vbo = try gl.Buffer.create();
-        errdefer vbo.destroy();
-        var vbobind = try vbo.bind(.ArrayBuffer);
-        defer vbobind.unbind();
-        var offset: usize = 0;
-        try vbobind.attributeAdvanced(0, 2, gl.c.GL_UNSIGNED_SHORT, false, @sizeOf(GPUCell), offset);
-        offset += 2 * @sizeOf(u16);
-        try vbobind.attributeAdvanced(1, 2, gl.c.GL_UNSIGNED_INT, false, @sizeOf(GPUCell), offset);
-        offset += 2 * @sizeOf(u32);
-        try vbobind.attributeAdvanced(2, 2, gl.c.GL_UNSIGNED_INT, false, @sizeOf(GPUCell), offset);
-        offset += 2 * @sizeOf(u32);
-        try vbobind.attributeAdvanced(3, 2, gl.c.GL_INT, false, @sizeOf(GPUCell), offset);
-        offset += 2 * @sizeOf(i32);
-        try vbobind.attributeAdvanced(4, 4, gl.c.GL_UNSIGNED_BYTE, false, @sizeOf(GPUCell), offset);
-        offset += 4 * @sizeOf(u8);
-        try vbobind.attributeAdvanced(5, 4, gl.c.GL_UNSIGNED_BYTE, false, @sizeOf(GPUCell), offset);
-        offset += 4 * @sizeOf(u8);
-        try vbobind.attributeIAdvanced(6, 1, gl.c.GL_UNSIGNED_BYTE, @sizeOf(GPUCell), offset);
-        offset += 1 * @sizeOf(u8);
-        try vbobind.attributeIAdvanced(7, 1, gl.c.GL_UNSIGNED_BYTE, @sizeOf(GPUCell), offset);
-        try vbobind.enableAttribArray(0);
-        try vbobind.enableAttribArray(1);
-        try vbobind.enableAttribArray(2);
-        try vbobind.enableAttribArray(3);
-        try vbobind.enableAttribArray(4);
-        try vbobind.enableAttribArray(5);
-        try vbobind.enableAttribArray(6);
-        try vbobind.enableAttribArray(7);
-        try vbobind.attributeDivisor(0, 1);
-        try vbobind.attributeDivisor(1, 1);
-        try vbobind.attributeDivisor(2, 1);
-        try vbobind.attributeDivisor(3, 1);
-        try vbobind.attributeDivisor(4, 1);
-        try vbobind.attributeDivisor(5, 1);
-        try vbobind.attributeDivisor(6, 1);
-        try vbobind.attributeDivisor(7, 1);
-
         // Build our texture
         const tex = try gl.Texture.create();
         errdefer tex.destroy();
@@ -1659,11 +1598,11 @@ const GLState = struct {
             try texbind.parameter(.MagFilter, gl.c.GL_LINEAR);
             try texbind.image2D(
                 0,
-                .Red,
+                .red,
                 @intCast(font_group.atlas_greyscale.size),
                 @intCast(font_group.atlas_greyscale.size),
                 0,
-                .Red,
+                .red,
                 .UnsignedByte,
                 font_group.atlas_greyscale.data.ptr,
             );
@@ -1680,32 +1619,32 @@ const GLState = struct {
             try texbind.parameter(.MagFilter, gl.c.GL_LINEAR);
             try texbind.image2D(
                 0,
-                .RGBA,
+                .rgba,
                 @intCast(font_group.atlas_color.size),
                 @intCast(font_group.atlas_color.size),
                 0,
-                .BGRA,
+                .bgra,
                 .UnsignedByte,
                 font_group.atlas_color.data.ptr,
             );
         }
 
+        // Build our cell renderer
+        const cell_program = try CellProgram.init();
+        errdefer cell_program.deinit();
+
         return .{
-            .program = program,
-            .vao = vao,
-            .ebo = ebo,
-            .vbo = vbo,
+            .cell_program = cell_program,
             .texture = tex,
             .texture_color = tex_color,
+            .custom = custom_state,
         };
     }
 
-    pub fn deinit(self: *GLState) void {
+    pub fn deinit(self: *GLState, alloc: Allocator) void {
+        if (self.custom) |v| v.deinit(alloc);
         self.texture.destroy();
         self.texture_color.destroy();
-        self.vbo.destroy();
-        self.ebo.destroy();
-        self.vao.destroy();
-        self.program.destroy();
+        self.cell_program.deinit();
     }
 };

src/renderer/Thread.zig

@@ -15,6 +15,7 @@ const App = @import("../App.zig");
 const Allocator = std.mem.Allocator;
 const log = std.log.scoped(.renderer_thread);
 
+const DRAW_INTERVAL = 33; // 30 FPS
 const CURSOR_BLINK_INTERVAL = 600;
 
 /// The type used for sending messages to the IO thread. For now this is
@@ -43,6 +44,13 @@ stop_c: xev.Completion = .{},
 render_h: xev.Timer,
 render_c: xev.Completion = .{},
 
+/// The timer used for draw calls. Draw calls don't update from the
+/// terminal state so they're much cheaper. They're used for animation
+/// and are paused when the terminal is not focused.
+draw_h: xev.Timer,
+draw_c: xev.Completion = .{},
+draw_active: bool = false,
+
 /// The timer used for cursor blinking
 cursor_h: xev.Timer,
 cursor_c: xev.Completion = .{},
@@ -100,6 +108,10 @@ pub fn init(
     var render_h = try xev.Timer.init();
     errdefer render_h.deinit();
 
+    // Draw timer, see comments.
+    var draw_h = try xev.Timer.init();
+    errdefer draw_h.deinit();
+
     // Setup a timer for blinking the cursor
     var cursor_timer = try xev.Timer.init();
     errdefer cursor_timer.deinit();
@@ -114,6 +126,7 @@ pub fn init(
         .wakeup = wakeup_h,
         .stop = stop_h,
         .render_h = render_h,
+        .draw_h = draw_h,
         .cursor_h = cursor_timer,
         .surface = surface,
         .renderer = renderer_impl,
@@ -129,6 +142,7 @@ pub fn deinit(self: *Thread) void {
     self.stop.deinit();
     self.wakeup.deinit();
     self.render_h.deinit();
+    self.draw_h.deinit();
     self.cursor_h.deinit();
     self.loop.deinit();
 
@@ -172,27 +186,8 @@ fn threadMain_(self: *Thread) !void {
         cursorTimerCallback,
     );
 
-    // If we are using tracy, then we setup a prepare handle so that
-    // we can mark the frame.
-    // TODO
-    // var frame_h: libuv.Prepare = if (!tracy.enabled) undefined else frame_h: {
-    //     const alloc_ptr = self.loop.getData(Allocator).?;
-    //     const alloc = alloc_ptr.*;
-    //     const h = try libuv.Prepare.init(alloc, self.loop);
-    //     h.setData(self);
-    //     try h.start(prepFrameCallback);
-    //
-    //     break :frame_h h;
-    // };
-    // defer if (tracy.enabled) {
-    //     frame_h.close((struct {
-    //         fn callback(h: *libuv.Prepare) void {
-    //             const alloc_h = h.loop().getData(Allocator).?.*;
-    //             h.deinit(alloc_h);
-    //         }
-    //     }).callback);
-    //     _ = self.loop.run(.nowait) catch {};
-    // };
+    // Start the draw timer
+    self.startDrawTimer();
 
     // Run
     log.debug("starting renderer thread", .{});
@@ -200,6 +195,34 @@ fn threadMain_(self: *Thread) !void {
     _ = try self.loop.run(.until_done);
 }
 
+fn startDrawTimer(self: *Thread) void {
+    // If our renderer doesn't suppoort animations then we never run this.
+    if (!@hasDecl(renderer.Renderer, "hasAnimations")) return;
+    if (!self.renderer.hasAnimations()) return;
+
+    // Set our active state so it knows we're running. We set this before
+    // even checking the active state in case we have a pending shutdown.
+    self.draw_active = true;
+
+    // If our draw timer is already active, then we don't have to do anything.
+    if (self.draw_c.state() == .active) return;
+
+    // Start the timer which loops
+    self.draw_h.run(
+        &self.loop,
+        &self.draw_c,
+        DRAW_INTERVAL,
+        Thread,
+        self,
+        drawCallback,
+    );
+}
+
+fn stopDrawTimer(self: *Thread) void {
+    // This will stop the draw on the next iteration.
+    self.draw_active = false;
+}
+
 /// Drain the mailbox.
 fn drainMailbox(self: *Thread) !void {
     const zone = trace(@src());
@@ -213,6 +236,9 @@ fn drainMailbox(self: *Thread) !void {
                 try self.renderer.setFocus(v);
 
                 if (!v) {
+                    // Stop the draw timer
+                    self.stopDrawTimer();
+
                     // If we're not focused, then we stop the cursor blink
                     if (self.cursor_c.state() == .active and
                         self.cursor_c_cancel.state() == .dead)
@@ -227,6 +253,9 @@ fn drainMailbox(self: *Thread) !void {
                         );
                     }
                 } else {
+                    // Start the draw timer
+                    self.startDrawTimer();
+
                     // If we're focused, we immediately show the cursor again
                     // and then restart the timer.
                     if (self.cursor_c.state() != .active) {
@@ -281,6 +310,11 @@ fn drainMailbox(self: *Thread) !void {
             .change_config => |config| {
                 defer config.alloc.destroy(config.ptr);
                 try self.renderer.changeConfig(config.ptr);
+
+                // Stop and start the draw timer to capture the new
+                // hasAnimations value.
+                self.stopDrawTimer();
+                self.startDrawTimer();
             },
 
             .inspector => |v| self.flags.has_inspector = v,
@@ -325,6 +359,41 @@ fn wakeupCallback(
     return .rearm;
 }
 
+fn drawCallback(
+    self_: ?*Thread,
+    _: *xev.Loop,
+    _: *xev.Completion,
+    r: xev.Timer.RunError!void,
+) xev.CallbackAction {
+    _ = r catch unreachable;
+    const t = self_ orelse {
+        // This shouldn't happen so we log it.
+        log.warn("render callback fired without data set", .{});
+        return .disarm;
+    };
+
+    // If we're doing single-threaded GPU calls then we just wake up the
+    // app thread to redraw at this point.
+    if (renderer.Renderer == renderer.OpenGL and
+        renderer.OpenGL.single_threaded_draw)
+    {
+        _ = t.app_mailbox.push(
+            .{ .redraw_surface = t.surface },
+            .{ .instant = {} },
+        );
+    } else {
+        t.renderer.drawFrame(t.surface) catch |err|
+            log.warn("error drawing err={}", .{err});
+    }
+
+    // Only continue if we're still active
+    if (t.draw_active) {
+        t.draw_h.run(&t.loop, &t.draw_c, DRAW_INTERVAL, Thread, t, drawCallback);
+    }
+
+    return .disarm;
+}
+
 fn renderCallback(
     self_: ?*Thread,
     _: *xev.Loop,
@@ -346,7 +415,8 @@ fn renderCallback(
         _ = t.app_mailbox.push(.{ .redraw_inspector = t.surface }, .{ .instant = {} });
     }
 
-    t.renderer.render(
+    // Update our frame data
+    t.renderer.updateFrame(
         t.surface,
         t.state,
         t.flags.cursor_blink_visible,
@@ -359,8 +429,13 @@ fn renderCallback(
         renderer.OpenGL.single_threaded_draw)
     {
         _ = t.app_mailbox.push(.{ .redraw_surface = t.surface }, .{ .instant = {} });
+        return .disarm;
     }
 
+    // Draw
+    t.renderer.drawFrame(t.surface) catch |err|
+        log.warn("error drawing err={}", .{err});
+
     return .disarm;
 }
 

src/renderer/metal/api.zig

@@ -57,6 +57,7 @@ pub const MTLVertexStepFunction = enum(c_ulong) {
 /// https://developer.apple.com/documentation/metal/mtlpixelformat?language=objc
 pub const MTLPixelFormat = enum(c_ulong) {
     r8unorm = 10,
+    rgba8unorm = 70,
     rgba8uint = 73,
     bgra8unorm = 80,
 };
@@ -66,6 +67,22 @@ pub const MTLPurgeableState = enum(c_ulong) {
     empty = 4,
 };
 
+/// https://developer.apple.com/documentation/metal/mtlsamplerminmagfilter?language=objc
+pub const MTLSamplerMinMagFilter = enum(c_ulong) {
+    nearest = 0,
+    linear = 1,
+};
+
+/// https://developer.apple.com/documentation/metal/mtlsampleraddressmode?language=objc
+pub const MTLSamplerAddressMode = enum(c_ulong) {
+    clamp_to_edge = 0,
+    mirror_clamp_to_edge = 1,
+    repeat = 2,
+    mirror_repeat = 3,
+    clamp_to_zero = 4,
+    clamp_to_border_color = 5,
+};
+
 /// https://developer.apple.com/documentation/metal/mtlblendfactor?language=objc
 pub const MTLBlendFactor = enum(c_ulong) {
     zero = 0,
@@ -98,6 +115,15 @@ pub const MTLBlendOperation = enum(c_ulong) {
     max = 4,
 };
 
+/// https://developer.apple.com/documentation/metal/mtltextureusage?language=objc<D-j>
+pub const MTLTextureUsage = enum(c_ulong) {
+    unknown = 0,
+    shader_read = 1,
+    shader_write = 2,
+    render_target = 4,
+    pixel_format_view = 8,
+};
+
 /// https://developer.apple.com/documentation/metal/mtlresourceoptions?language=objc
 /// (incomplete, we only use this mode so we just hardcode it)
 pub const MTLResourceStorageModeShared: c_ulong = @intFromEnum(MTLStorageMode.shared) << 4;

src/renderer/metal/image.zig

@@ -218,7 +218,7 @@ pub const Image = union(enum) {
     fn initTexture(p: Pending, device: objc.Object) !objc.Object {
         // Create our descriptor
         const desc = init: {
-            const Class = objc.Class.getClass("MTLTextureDescriptor").?;
+            const Class = objc.getClass("MTLTextureDescriptor").?;
             const id_alloc = Class.msgSend(objc.Object, objc.sel("alloc"), .{});
             const id_init = id_alloc.msgSend(objc.Object, objc.sel("init"), .{});
             break :init id_init;

src/renderer/metal/sampler.zig

@@ -0,0 +1,38 @@
+const std = @import("std");
+const Allocator = std.mem.Allocator;
+const assert = std.debug.assert;
+const objc = @import("objc");
+
+const mtl = @import("api.zig");
+
+pub const Sampler = struct {
+    sampler: objc.Object,
+
+    pub fn init(device: objc.Object) !Sampler {
+        const desc = init: {
+            const Class = objc.getClass("MTLSamplerDescriptor").?;
+            const id_alloc = Class.msgSend(objc.Object, objc.sel("alloc"), .{});
+            const id_init = id_alloc.msgSend(objc.Object, objc.sel("init"), .{});
+            break :init id_init;
+        };
+        defer desc.msgSend(void, objc.sel("release"), .{});
+        desc.setProperty("rAddressMode", @intFromEnum(mtl.MTLSamplerAddressMode.clamp_to_edge));
+        desc.setProperty("sAddressMode", @intFromEnum(mtl.MTLSamplerAddressMode.clamp_to_edge));
+        desc.setProperty("tAddressMode", @intFromEnum(mtl.MTLSamplerAddressMode.clamp_to_edge));
+        desc.setProperty("minFilter", @intFromEnum(mtl.MTLSamplerMinMagFilter.linear));
+        desc.setProperty("magFilter", @intFromEnum(mtl.MTLSamplerMinMagFilter.linear));
+
+        const sampler = device.msgSend(
+            objc.Object,
+            objc.sel("newSamplerStateWithDescriptor:"),
+            .{desc},
+        );
+        errdefer sampler.msgSend(void, objc.sel("release"), .{});
+
+        return .{ .sampler = sampler };
+    }
+
+    pub fn deinit(self: *Sampler) void {
+        self.sampler.msgSend(void, objc.sel("release"), .{});
+    }
+};

src/renderer/metal/shaders.zig

@@ -12,10 +12,31 @@ const log = std.log.scoped(.metal);
 /// This contains the state for the shaders used by the Metal renderer.
 pub const Shaders = struct {
     library: objc.Object,
+
+    /// The cell shader is the shader used to render the terminal cells.
+    /// It is a single shader that is used for both the background and
+    /// foreground.
     cell_pipeline: objc.Object,
+
+    /// The image shader is the shader used to render images for things
+    /// like the Kitty image protocol.
     image_pipeline: objc.Object,
 
-    pub fn init(device: objc.Object) !Shaders {
+    /// Custom shaders to run against the final drawable texture. This
+    /// can be used to apply a lot of effects. Each shader is run in sequence
+    /// against the output of the previous shader.
+    post_pipelines: []const objc.Object,
+
+    /// Initialize our shader set.
+    ///
+    /// "post_shaders" is an optional list of postprocess shaders to run
+    /// against the final drawable texture. This is an array of shader source
+    /// code, not file paths.
+    pub fn init(
+        alloc: Allocator,
+        device: objc.Object,
+        post_shaders: []const [:0]const u8,
+    ) !Shaders {
         const library = try initLibrary(device);
         errdefer library.msgSend(void, objc.sel("release"), .{});
 
@@ -25,17 +46,44 @@ pub const Shaders = struct {
         const image_pipeline = try initImagePipeline(device, library);
         errdefer image_pipeline.msgSend(void, objc.sel("release"), .{});
 
+        const post_pipelines: []const objc.Object = initPostPipelines(
+            alloc,
+            device,
+            library,
+            post_shaders,
+        ) catch |err| err: {
+            // If an error happens while building postprocess shaders we
+            // want to just not use any postprocess shaders since we don't
+            // want to block Ghostty from working.
+            log.warn("error initializing postprocess shaders err={}", .{err});
+            break :err &.{};
+        };
+        errdefer if (post_pipelines.len > 0) {
+            for (post_pipelines) |pipeline| pipeline.msgSend(void, objc.sel("release"), .{});
+            alloc.free(post_pipelines);
+        };
+
         return .{
             .library = library,
             .cell_pipeline = cell_pipeline,
             .image_pipeline = image_pipeline,
+            .post_pipelines = post_pipelines,
         };
     }
 
-    pub fn deinit(self: *Shaders) void {
+    pub fn deinit(self: *Shaders, alloc: Allocator) void {
+        // Release our primary shaders
         self.cell_pipeline.msgSend(void, objc.sel("release"), .{});
         self.image_pipeline.msgSend(void, objc.sel("release"), .{});
         self.library.msgSend(void, objc.sel("release"), .{});
+
+        // Release our postprocess shaders
+        if (self.post_pipelines.len > 0) {
+            for (self.post_pipelines) |pipeline| {
+                pipeline.msgSend(void, objc.sel("release"), .{});
+            }
+            alloc.free(self.post_pipelines);
+        }
     }
 };
 
@@ -79,6 +127,23 @@ pub const Uniforms = extern struct {
     strikethrough_thickness: f32,
 };
 
+/// The uniforms used for custom postprocess shaders.
+pub const PostUniforms = extern struct {
+    // Note: all of the explicit aligmnments are copied from the
+    // MSL developer reference just so that we can be sure that we got
+    // it all exactly right.
+    resolution: [3]f32 align(16),
+    time: f32 align(4),
+    time_delta: f32 align(4),
+    frame_rate: f32 align(4),
+    frame: i32 align(4),
+    channel_time: [4][4]f32 align(16),
+    channel_resolution: [4][4]f32 align(16),
+    mouse: [4]f32 align(16),
+    date: [4]f32 align(16),
+    sample_rate: f32 align(4),
+};
+
 /// Initialize the MTLLibrary. A MTLLibrary is a collection of shaders.
 fn initLibrary(device: objc.Object) !objc.Object {
     // Hardcoded since this file isn't meant to be reusable.
@@ -105,6 +170,129 @@ fn initLibrary(device: objc.Object) !objc.Object {
     return library;
 }
 
+/// Initialize our custom shader pipelines. The shaders argument is a
+/// set of shader source code, not file paths.
+fn initPostPipelines(
+    alloc: Allocator,
+    device: objc.Object,
+    library: objc.Object,
+    shaders: []const [:0]const u8,
+) ![]const objc.Object {
+    // If we have no shaders, do nothing.
+    if (shaders.len == 0) return &.{};
+
+    // Keeps track of how many shaders we successfully wrote.
+    var i: usize = 0;
+
+    // Initialize our result set. If any error happens, we undo everything.
+    var pipelines = try alloc.alloc(objc.Object, shaders.len);
+    errdefer {
+        for (pipelines[0..i]) |pipeline| {
+            pipeline.msgSend(void, objc.sel("release"), .{});
+        }
+        alloc.free(pipelines);
+    }
+
+    // Build each shader. Note we don't use "0.." to build our index
+    // because we need to keep track of our length to clean up above.
+    for (shaders) |source| {
+        pipelines[i] = try initPostPipeline(device, library, source);
+        i += 1;
+    }
+
+    return pipelines;
+}
+
+/// Initialize a single custom shader pipeline from shader source.
+fn initPostPipeline(
+    device: objc.Object,
+    library: objc.Object,
+    data: [:0]const u8,
+) !objc.Object {
+    // Create our library which has the shader source
+    const post_library = library: {
+        const source = try macos.foundation.String.createWithBytes(
+            data,
+            .utf8,
+            false,
+        );
+        defer source.release();
+
+        var err: ?*anyopaque = null;
+        const post_library = device.msgSend(
+            objc.Object,
+            objc.sel("newLibraryWithSource:options:error:"),
+            .{ source, @as(?*anyopaque, null), &err },
+        );
+        try checkError(err);
+        errdefer post_library.msgSend(void, objc.sel("release"), .{});
+
+        break :library post_library;
+    };
+    defer post_library.msgSend(void, objc.sel("release"), .{});
+
+    // Get our vertex and fragment functions
+    const func_vert = func_vert: {
+        const str = try macos.foundation.String.createWithBytes(
+            "post_vertex",
+            .utf8,
+            false,
+        );
+        defer str.release();
+
+        const ptr = library.msgSend(?*anyopaque, objc.sel("newFunctionWithName:"), .{str});
+        break :func_vert objc.Object.fromId(ptr.?);
+    };
+    const func_frag = func_frag: {
+        const str = try macos.foundation.String.createWithBytes(
+            "main0",
+            .utf8,
+            false,
+        );
+        defer str.release();
+
+        const ptr = post_library.msgSend(?*anyopaque, objc.sel("newFunctionWithName:"), .{str});
+        break :func_frag objc.Object.fromId(ptr.?);
+    };
+    defer func_vert.msgSend(void, objc.sel("release"), .{});
+    defer func_frag.msgSend(void, objc.sel("release"), .{});
+
+    // Create our descriptor
+    const desc = init: {
+        const Class = objc.getClass("MTLRenderPipelineDescriptor").?;
+        const id_alloc = Class.msgSend(objc.Object, objc.sel("alloc"), .{});
+        const id_init = id_alloc.msgSend(objc.Object, objc.sel("init"), .{});
+        break :init id_init;
+    };
+    defer desc.msgSend(void, objc.sel("release"), .{});
+    desc.setProperty("vertexFunction", func_vert);
+    desc.setProperty("fragmentFunction", func_frag);
+
+    // Set our color attachment
+    const attachments = objc.Object.fromId(desc.getProperty(?*anyopaque, "colorAttachments"));
+    {
+        const attachment = attachments.msgSend(
+            objc.Object,
+            objc.sel("objectAtIndexedSubscript:"),
+            .{@as(c_ulong, 0)},
+        );
+
+        // Value is MTLPixelFormatBGRA8Unorm
+        attachment.setProperty("pixelFormat", @as(c_ulong, 80));
+    }
+
+    // Make our state
+    var err: ?*anyopaque = null;
+    const pipeline_state = device.msgSend(
+        objc.Object,
+        objc.sel("newRenderPipelineStateWithDescriptor:error:"),
+        .{ desc, &err },
+    );
+    try checkError(err);
+
+    return pipeline_state;
+}
+
 /// Initialize the cell render pipeline for our shader library.
 fn initCellPipeline(device: objc.Object, library: objc.Object) !objc.Object {
     // Get our vertex and fragment functions
@@ -130,6 +318,8 @@ fn initCellPipeline(device: objc.Object, library: objc.Object) !objc.Object {
         const ptr = library.msgSend(?*anyopaque, objc.sel("newFunctionWithName:"), .{str});
         break :func_frag objc.Object.fromId(ptr.?);
     };
+    defer func_vert.msgSend(void, objc.sel("release"), .{});
+    defer func_frag.msgSend(void, objc.sel("release"), .{});
 
     // Create the vertex descriptor. The vertex descriptor describes the
     // data layout of the vertex inputs. We use indexed (or "instanced")
@@ -137,7 +327,7 @@ fn initCellPipeline(device: objc.Object, library: objc.Object) !objc.Object {
     // Cell as input.
     const vertex_desc = vertex_desc: {
         const desc = init: {
-            const Class = objc.Class.getClass("MTLVertexDescriptor").?;
+            const Class = objc.getClass("MTLVertexDescriptor").?;
             const id_alloc = Class.msgSend(objc.Object, objc.sel("alloc"), .{});
             const id_init = id_alloc.msgSend(objc.Object, objc.sel("init"), .{});
             break :init id_init;
@@ -239,14 +429,16 @@ fn initCellPipeline(device: objc.Object, library: objc.Object) !objc.Object {
 
         break :vertex_desc desc;
     };
+    defer vertex_desc.msgSend(void, objc.sel("release"), .{});
 
     // Create our descriptor
     const desc = init: {
-        const Class = objc.Class.getClass("MTLRenderPipelineDescriptor").?;
+        const Class = objc.getClass("MTLRenderPipelineDescriptor").?;
         const id_alloc = Class.msgSend(objc.Object, objc.sel("alloc"), .{});
         const id_init = id_alloc.msgSend(objc.Object, objc.sel("init"), .{});
         break :init id_init;
     };
+    defer desc.msgSend(void, objc.sel("release"), .{});
 
     // Set our properties
     desc.setProperty("vertexFunction", func_vert);
@@ -284,6 +476,7 @@ fn initCellPipeline(device: objc.Object, library: objc.Object) !objc.Object {
         .{ desc, &err },
     );
     try checkError(err);
+    errdefer pipeline_state.msgSend(void, objc.sel("release"), .{});
 
     return pipeline_state;
 }
@@ -313,6 +506,8 @@ fn initImagePipeline(device: objc.Object, library: objc.Object) !objc.Object {
         const ptr = library.msgSend(?*anyopaque, objc.sel("newFunctionWithName:"), .{str});
         break :func_frag objc.Object.fromId(ptr.?);
     };
+    defer func_vert.msgSend(void, objc.sel("release"), .{});
+    defer func_frag.msgSend(void, objc.sel("release"), .{});
 
     // Create the vertex descriptor. The vertex descriptor describes the
     // data layout of the vertex inputs. We use indexed (or "instanced")
@@ -320,7 +515,7 @@ fn initImagePipeline(device: objc.Object, library: objc.Object) !objc.Object {
     // Image as input.
     const vertex_desc = vertex_desc: {
         const desc = init: {
-            const Class = objc.Class.getClass("MTLVertexDescriptor").?;
+            const Class = objc.getClass("MTLVertexDescriptor").?;
             const id_alloc = Class.msgSend(objc.Object, objc.sel("alloc"), .{});
             const id_init = id_alloc.msgSend(objc.Object, objc.sel("init"), .{});
             break :init id_init;
@@ -389,14 +584,16 @@ fn initImagePipeline(device: objc.Object, library: objc.Object) !objc.Object {
 
         break :vertex_desc desc;
     };
+    defer vertex_desc.msgSend(void, objc.sel("release"), .{});
 
     // Create our descriptor
     const desc = init: {
-        const Class = objc.Class.getClass("MTLRenderPipelineDescriptor").?;
+        const Class = objc.getClass("MTLRenderPipelineDescriptor").?;
         const id_alloc = Class.msgSend(objc.Object, objc.sel("alloc"), .{});
         const id_init = id_alloc.msgSend(objc.Object, objc.sel("init"), .{});
         break :init id_init;
     };
+    defer desc.msgSend(void, objc.sel("release"), .{});
 
     // Set our properties
     desc.setProperty("vertexFunction", func_vert);

src/renderer/opengl/CellProgram.zig

@@ -0,0 +1,174 @@
+/// The OpenGL program for rendering terminal cells.
+const CellProgram = @This();
+
+const std = @import("std");
+const gl = @import("opengl");
+
+program: gl.Program,
+vao: gl.VertexArray,
+ebo: gl.Buffer,
+vbo: gl.Buffer,
+
+/// The raw structure that maps directly to the buffer sent to the vertex shader.
+/// This must be "extern" so that the field order is not reordered by the
+/// Zig compiler.
+pub const Cell = extern struct {
+    /// vec2 grid_coord
+    grid_col: u16,
+    grid_row: u16,
+
+    /// vec2 glyph_pos
+    glyph_x: u32 = 0,
+    glyph_y: u32 = 0,
+
+    /// vec2 glyph_size
+    glyph_width: u32 = 0,
+    glyph_height: u32 = 0,
+
+    /// vec2 glyph_offset
+    glyph_offset_x: i32 = 0,
+    glyph_offset_y: i32 = 0,
+
+    /// vec4 fg_color_in
+    r: u8,
+    g: u8,
+    b: u8,
+    a: u8,
+
+    /// uint mode
+    mode: CellMode,
+
+    /// The width in grid cells that a rendering takes.
+    grid_width: u8,
+};
+
+pub const CellMode = enum(u8) {
+    bg = 1,
+    fg = 2,
+    fg_color = 7,
+    strikethrough = 8,
+
+    // Non-exhaustive because masks change it
+    _,
+
+    /// Apply a mask to the mode.
+    pub fn mask(self: CellMode, m: CellMode) CellMode {
+        return @enumFromInt(@intFromEnum(self) | @intFromEnum(m));
+    }
+};
+
+pub fn init() !CellProgram {
+    // Load and compile our shaders.
+    const program = try gl.Program.createVF(
+        @embedFile("../shaders/cell.v.glsl"),
+        @embedFile("../shaders/cell.f.glsl"),
+    );
+    errdefer program.destroy();
+
+    // Set our cell dimensions
+    const pbind = try program.use();
+    defer pbind.unbind();
+
+    // Set all of our texture indexes
+    try program.setUniform("text", 0);
+    try program.setUniform("text_color", 1);
+
+    // Setup our VAO
+    const vao = try gl.VertexArray.create();
+    errdefer vao.destroy();
+    const vaobind = try vao.bind();
+    defer vaobind.unbind();
+
+    // Element buffer (EBO)
+    const ebo = try gl.Buffer.create();
+    errdefer ebo.destroy();
+    var ebobind = try ebo.bind(.element_array);
+    defer ebobind.unbind();
+    try ebobind.setData([6]u8{
+        0, 1, 3, // Top-left triangle
+        1, 2, 3, // Bottom-right triangle
+    }, .static_draw);
+
+    // Vertex buffer (VBO)
+    const vbo = try gl.Buffer.create();
+    errdefer vbo.destroy();
+    var vbobind = try vbo.bind(.array);
+    defer vbobind.unbind();
+    var offset: usize = 0;
+    try vbobind.attributeAdvanced(0, 2, gl.c.GL_UNSIGNED_SHORT, false, @sizeOf(Cell), offset);
+    offset += 2 * @sizeOf(u16);
+    try vbobind.attributeAdvanced(1, 2, gl.c.GL_UNSIGNED_INT, false, @sizeOf(Cell), offset);
+    offset += 2 * @sizeOf(u32);
+    try vbobind.attributeAdvanced(2, 2, gl.c.GL_UNSIGNED_INT, false, @sizeOf(Cell), offset);
+    offset += 2 * @sizeOf(u32);
+    try vbobind.attributeAdvanced(3, 2, gl.c.GL_INT, false, @sizeOf(Cell), offset);
+    offset += 2 * @sizeOf(i32);
+    try vbobind.attributeAdvanced(4, 4, gl.c.GL_UNSIGNED_BYTE, false, @sizeOf(Cell), offset);
+    offset += 4 * @sizeOf(u8);
+    try vbobind.attributeIAdvanced(5, 1, gl.c.GL_UNSIGNED_BYTE, @sizeOf(Cell), offset);
+    offset += 1 * @sizeOf(u8);
+    try vbobind.attributeIAdvanced(6, 1, gl.c.GL_UNSIGNED_BYTE, @sizeOf(Cell), offset);
+    try vbobind.enableAttribArray(0);
+    try vbobind.enableAttribArray(1);
+    try vbobind.enableAttribArray(2);
+    try vbobind.enableAttribArray(3);
+    try vbobind.enableAttribArray(4);
+    try vbobind.enableAttribArray(5);
+    try vbobind.enableAttribArray(6);
+    try vbobind.attributeDivisor(0, 1);
+    try vbobind.attributeDivisor(1, 1);
+    try vbobind.attributeDivisor(2, 1);
+    try vbobind.attributeDivisor(3, 1);
+    try vbobind.attributeDivisor(4, 1);
+    try vbobind.attributeDivisor(5, 1);
+    try vbobind.attributeDivisor(6, 1);
+
+    return .{
+        .program = program,
+        .vao = vao,
+        .ebo = ebo,
+        .vbo = vbo,
+    };
+}
+
+pub fn bind(self: CellProgram) !Binding {
+    const program = try self.program.use();
+    errdefer program.unbind();
+
+    const vao = try self.vao.bind();
+    errdefer vao.unbind();
+
+    const ebo = try self.ebo.bind(.element_array);
+    errdefer ebo.unbind();
+
+    const vbo = try self.vbo.bind(.array);
+    errdefer vbo.unbind();
+
+    return .{
+        .program = program,
+        .vao = vao,
+        .ebo = ebo,
+        .vbo = vbo,
+    };
+}
+
+pub fn deinit(self: CellProgram) void {
+    self.vbo.destroy();
+    self.ebo.destroy();
+    self.vao.destroy();
+    self.program.destroy();
+}
+
+pub const Binding = struct {
+    program: gl.Program.Binding,
+    vao: gl.VertexArray.Binding,
+    ebo: gl.Buffer.Binding,
+    vbo: gl.Buffer.Binding,
+
+    pub fn unbind(self: Binding) void {
+        self.vbo.unbind();
+        self.ebo.unbind();
+        self.vao.unbind();
+        self.program.unbind();
+    }
+};

src/renderer/opengl/custom.zig

@@ -0,0 +1,289 @@
+const std = @import("std");
+const Allocator = std.mem.Allocator;
+const gl = @import("opengl");
+const ScreenSize = @import("../size.zig").ScreenSize;
+
+const log = std.log.scoped(.opengl_custom);
+
+/// The "INDEX" is the index into the global GL state and the
+/// "BINDING" is the binding location in the shader.
+const UNIFORM_INDEX: gl.c.GLuint = 0;
+const UNIFORM_BINDING: gl.c.GLuint = 0;
+
+/// Global uniforms for custom shaders.
+pub const Uniforms = extern struct {
+    resolution: [3]f32 align(16) = .{ 0, 0, 0 },
+    time: f32 align(4) = 1,
+    time_delta: f32 align(4) = 1,
+    frame_rate: f32 align(4) = 1,
+    frame: i32 align(4) = 1,
+    channel_time: [4][4]f32 align(16) = [1][4]f32{.{ 0, 0, 0, 0 }} ** 4,
+    channel_resolution: [4][4]f32 align(16) = [1][4]f32{.{ 0, 0, 0, 0 }} ** 4,
+    mouse: [4]f32 align(16) = .{ 0, 0, 0, 0 },
+    date: [4]f32 align(16) = .{ 0, 0, 0, 0 },
+    sample_rate: f32 align(4) = 1,
+};
+
+/// The state associated with custom shaders. This should only be initialized
+/// if there is at least one custom shader.
+///
+/// To use this, the main terminal shader should render to the framebuffer
+/// specified by "fbo". The resulting "fb_texture" will contain the color
+/// attachment. This is then used as the iChannel0 input to the custom
+/// shader.
+pub const State = struct {
+    /// The uniform data
+    uniforms: Uniforms,
+
+    /// The OpenGL buffers
+    fbo: gl.Framebuffer,
+    ubo: gl.Buffer,
+    vao: gl.VertexArray,
+    ebo: gl.Buffer,
+    fb_texture: gl.Texture,
+
+    /// The set of programs for the custom shaders.
+    programs: []const Program,
+
+    /// The last time the frame was drawn. This is used to update
+    /// the time uniform.
+    first_frame_time: std.time.Instant,
+
+    pub fn init(
+        alloc: Allocator,
+        srcs: []const [:0]const u8,
+    ) !State {
+        if (srcs.len == 0) return error.OneCustomShaderRequired;
+
+        // Create our programs
+        var programs = std.ArrayList(Program).init(alloc);
+        defer programs.deinit();
+        errdefer for (programs.items) |p| p.deinit();
+        for (srcs) |src| {
+            try programs.append(try Program.init(src));
+        }
+
+        // Create the texture for the framebuffer
+        const fb_tex = try gl.Texture.create();
+        errdefer fb_tex.destroy();
+        {
+            const texbind = try fb_tex.bind(.@"2D");
+            try texbind.parameter(.WrapS, gl.c.GL_CLAMP_TO_EDGE);
+            try texbind.parameter(.WrapT, gl.c.GL_CLAMP_TO_EDGE);
+            try texbind.parameter(.MinFilter, gl.c.GL_LINEAR);
+            try texbind.parameter(.MagFilter, gl.c.GL_LINEAR);
+            try texbind.image2D(
+                0,
+                .rgb,
+                1,
+                1,
+                0,
+                .rgb,
+                .UnsignedByte,
+                null,
+            );
+        }
+
+        // Create our framebuffer for rendering off screen.
+        // The shader prior to custom shaders should use this
+        // framebuffer.
+        const fbo = try gl.Framebuffer.create();
+        errdefer fbo.destroy();
+        const fbbind = try fbo.bind(.framebuffer);
+        defer fbbind.unbind();
+        try fbbind.texture2D(.color0, .@"2D", fb_tex, 0);
+        const fbstatus = fbbind.checkStatus();
+        if (fbstatus != .complete) {
+            log.warn(
+                "framebuffer is not complete state={}",
+                .{fbstatus},
+            );
+            return error.InvalidFramebuffer;
+        }
+
+        // Create our uniform buffer that is shared across all
+        // custom shaders
+        const ubo = try gl.Buffer.create();
+        errdefer ubo.destroy();
+        {
+            var ubobind = try ubo.bind(.uniform);
+            defer ubobind.unbind();
+            try ubobind.setDataNull(Uniforms, .static_draw);
+        }
+
+        // Setup our VAO for the custom shader.
+        const vao = try gl.VertexArray.create();
+        errdefer vao.destroy();
+        const vaobind = try vao.bind();
+        defer vaobind.unbind();
+
+        // Element buffer (EBO)
+        const ebo = try gl.Buffer.create();
+        errdefer ebo.destroy();
+        var ebobind = try ebo.bind(.element_array);
+        defer ebobind.unbind();
+        try ebobind.setData([6]u8{
+            0, 1, 3, // Top-left triangle
+            1, 2, 3, // Bottom-right triangle
+        }, .static_draw);
+
+        return .{
+            .programs = try programs.toOwnedSlice(),
+            .uniforms = .{},
+            .fbo = fbo,
+            .ubo = ubo,
+            .vao = vao,
+            .ebo = ebo,
+            .fb_texture = fb_tex,
+            .first_frame_time = try std.time.Instant.now(),
+        };
+    }
+
+    pub fn deinit(self: *const State, alloc: Allocator) void {
+        for (self.programs) |p| p.deinit();
+        alloc.free(self.programs);
+        self.ubo.destroy();
+        self.ebo.destroy();
+        self.vao.destroy();
+        self.fb_texture.destroy();
+        self.fbo.destroy();
+    }
+
+    pub fn setScreenSize(self: *State, size: ScreenSize) !void {
+        // Update our uniforms
+        self.uniforms.resolution = .{
+            @floatFromInt(size.width),
+            @floatFromInt(size.height),
+            1,
+        };
+        try self.syncUniforms();
+
+        // Update our texture
+        const texbind = try self.fb_texture.bind(.@"2D");
+        try texbind.image2D(
+            0,
+            .rgb,
+            @intCast(size.width),
+            @intCast(size.height),
+            0,
+            .rgb,
+            .UnsignedByte,
+            null,
+        );
+    }
+
+    /// Call this prior to drawing a frame to update the time
+    /// and synchronize the uniforms. This synchronizes uniforms
+    /// so you should make changes to uniforms prior to calling
+    /// this.
+    pub fn newFrame(self: *State) !void {
+        // Update our frame time
+        const now = std.time.Instant.now() catch self.first_frame_time;
+        const since_ns: f32 = @floatFromInt(now.since(self.first_frame_time));
+        self.uniforms.time = since_ns / std.time.ns_per_s;
+        self.uniforms.time_delta = since_ns / std.time.ns_per_s;
+
+        // Sync our uniform changes
+        try self.syncUniforms();
+    }
+
+    fn syncUniforms(self: *State) !void {
+        var ubobind = try self.ubo.bind(.uniform);
+        defer ubobind.unbind();
+        try ubobind.setData(self.uniforms, .static_draw);
+    }
+
+    /// Call this to bind all the necessary OpenGL resources for
+    /// all custom shaders. Each individual shader needs to be bound
+    /// one at a time too.
+    pub fn bind(self: *const State) !Binding {
+        // Move our uniform buffer into proper global index. Note that
+        // in theory we can do this globally once and never worry about
+        // it again. I don't think we're high-performance enough at all
+        // to worry about that and this makes it so you can just move
+        // around CustomProgram usage without worrying about clobbering
+        // the global state.
+        try self.ubo.bindBase(.uniform, UNIFORM_INDEX);
+
+        // Bind our texture that is shared amongst all
+        try gl.Texture.active(gl.c.GL_TEXTURE0);
+        var texbind = try self.fb_texture.bind(.@"2D");
+        errdefer texbind.unbind();
+
+        const vao = try self.vao.bind();
+        errdefer vao.unbind();
+
+        const ebo = try self.ebo.bind(.element_array);
+        errdefer ebo.unbind();
+
+        return .{
+            .vao = vao,
+            .ebo = ebo,
+            .fb_texture = texbind,
+        };
+    }
+
+    pub const Binding = struct {
+        vao: gl.VertexArray.Binding,
+        ebo: gl.Buffer.Binding,
+        fb_texture: gl.Texture.Binding,
+
+        pub fn unbind(self: Binding) void {
+            self.ebo.unbind();
+            self.vao.unbind();
+            self.fb_texture.unbind();
+        }
+    };
+};
+
+/// A single OpenGL program (combined shaders) for custom shaders.
+pub const Program = struct {
+    program: gl.Program,
+
+    pub fn init(src: [:0]const u8) !Program {
+        const program = try gl.Program.createVF(
+            @embedFile("../shaders/custom.v.glsl"),
+            src,
+            //@embedFile("../shaders/temp.f.glsl"),
+        );
+        errdefer program.destroy();
+
+        // Map our uniform buffer to the global GL state
+        try program.uniformBlockBinding(UNIFORM_INDEX, UNIFORM_BINDING);
+
+        return .{ .program = program };
+    }
+
+    pub fn deinit(self: *const Program) void {
+        self.program.destroy();
+    }
+
+    /// Bind the program for use. This should be called so that draw can
+    /// be called.
+    pub fn bind(self: *const Program) !Binding {
+        const program = try self.program.use();
+        errdefer program.unbind();
+
+        return .{
+            .program = program,
+        };
+    }
+
+    pub const Binding = struct {
+        program: gl.Program.Binding,
+
+        pub fn unbind(self: Binding) void {
+            self.program.unbind();
+        }
+
+        pub fn draw(self: Binding) !void {
+            _ = self;
+            try gl.drawElementsInstanced(
+                gl.c.GL_TRIANGLES,
+                6,
+                gl.c.GL_UNSIGNED_BYTE,
+                1,
+            );
+        }
+    };
+};

src/renderer/shaders/cell.metal

@@ -253,3 +253,20 @@ fragment float4 image_fragment(
   uint4 rgba = image.sample(textureSampler, in.tex_coord);
   return float4(rgba) / 255.0f;
 }
+
+//-------------------------------------------------------------------
+// Post Shader
+//-------------------------------------------------------------------
+#pragma mark - Post Shader
+
+struct PostVertexOut {
+    float4 position [[ position ]];
+};
+
+constant float2 post_pos[4] = { {-1,-1}, {1,-1}, {-1,1}, {1,1 } };
+
+vertex PostVertexOut post_vertex(uint id [[ vertex_id ]]) {
+    PostVertexOut out;
+    out.position = float4(post_pos[id], 0, 1);
+    return out;
+}

src/renderer/shaders/cell.v.glsl

@@ -21,20 +21,18 @@ layout (location = 2) in vec2 glyph_size;
 // Offset of the top-left corner of the glyph when rendered in a rect.
 layout (location = 3) in vec2 glyph_offset;
 
-// The background color for this cell in RGBA (0 to 1.0)
-layout (location = 4) in vec4 fg_color_in;
-
-// The background color for this cell in RGBA (0 to 1.0)
-layout (location = 5) in vec4 bg_color_in;
+// The color for this cell in RGBA (0 to 1.0). Background or foreground
+// depends on mode.
+layout (location = 4) in vec4 color_in;
 
 // The mode of this shader. The mode determines what fields are used,
 // what the output will be, etc. This shader is capable of executing in
 // multiple "modes" so that we can share some logic and so that we can draw
 // the entire terminal grid in a single GPU pass.
-layout (location = 6) in uint mode_in;
+layout (location = 5) in uint mode_in;
 
 // The width in cells of this item.
-layout (location = 7) in uint grid_width;
+layout (location = 6) in uint grid_width;
 
 // The background or foreground color for the fragment, depending on
 // whether this is a background or foreground pass.
@@ -117,7 +115,7 @@ void main() {
         cell_pos = cell_pos + cell_size_scaled * position;
 
         gl_Position = projection * vec4(cell_pos, cell_z, 1.0);
-        color = bg_color_in / 255.0;
+        color = color_in / 255.0;
         break;
 
     case MODE_FG:
@@ -150,7 +148,7 @@ void main() {
         glyph_tex_coords = glyph_tex_pos + glyph_tex_size * position;
 
         // Set our foreground color output
-        color = fg_color_in / 255.;
+        color = color_in / 255.;
         break;
 
     case MODE_STRIKETHROUGH:
@@ -166,7 +164,7 @@ void main() {
         cell_pos = cell_pos + strikethrough_offset - (strikethrough_size * position);
 
         gl_Position = projection * vec4(cell_pos, cell_z, 1.0);
-        color = fg_color_in / 255.0;
+        color = color_in / 255.0;
         break;
     }
 }

src/renderer/shaders/custom.v.glsl

@@ -0,0 +1,8 @@
+#version 330 core
+
+void main(){
+    vec2 position;
+    position.x = (gl_VertexID == 0 || gl_VertexID == 1) ? -1. : 1.;
+    position.y = (gl_VertexID == 0 || gl_VertexID == 3) ? 1. : -1.;
+    gl_Position = vec4(position.xy, 0.0f, 1.0f);
+}

src/renderer/shaders/shadertoy_prefix.glsl

@@ -0,0 +1,29 @@
+#version 430 core
+
+layout(binding = 0) uniform Globals {
+    uniform vec3	iResolution;
+    uniform float	iTime;
+    uniform float	iTimeDelta;
+    uniform float	iFrameRate;
+    uniform int	    iFrame;
+    uniform float	iChannelTime[4];
+    uniform vec3	iChannelResolution[4];
+    uniform vec4	iMouse;
+    uniform vec4	iDate;
+    uniform float	iSampleRate;
+};
+
+layout(binding = 0) uniform sampler2D	iChannel0;
+
+// These are unused currently by Ghostty:
+// layout(binding = 1) uniform sampler2D	iChannel1;
+// layout(binding = 2) uniform sampler2D	iChannel2;
+// layout(binding = 3) uniform sampler2D	iChannel3;
+
+layout(location = 0) in vec4 gl_FragCoord;
+layout(location = 0) out vec4 _fragColor;
+
+#define texture2D texture
+
+void mainImage( out vec4 fragColor, in vec2 fragCoord );
+void main() { mainImage (_fragColor, gl_FragCoord.xy); }

src/renderer/shaders/test_shadertoy_crt.glsl

@@ -0,0 +1,59 @@
+// This shader is NOT BUILD INTO Ghostty. This is only here for unit tests.
+
+// Loosely based on postprocessing shader by inigo quilez, License Creative
+// Commons Attribution-NonCommercial-ShareAlike 3.0 Unported License.
+
+vec2 curve(vec2 uv)
+{
+	uv = (uv - 0.5) * 2.0;
+	uv *= 1.1;
+	uv.x *= 1.0 + pow((abs(uv.y) / 5.0), 2.0);
+	uv.y *= 1.0 + pow((abs(uv.x) / 4.0), 2.0);
+	uv  = (uv / 2.0) + 0.5;
+	uv =  uv *0.92 + 0.04;
+	return uv;
+}
+void mainImage( out vec4 fragColor, in vec2 fragCoord )
+{
+    vec2 q = fragCoord.xy / iResolution.xy;
+    vec2 uv = q;
+    uv = curve( uv );
+    vec3 oricol = texture( iChannel0, vec2(q.x,q.y) ).xyz;
+    vec3 col;
+	float x =  sin(0.3*iTime+uv.y*21.0)*sin(0.7*iTime+uv.y*29.0)*sin(0.3+0.33*iTime+uv.y*31.0)*0.0017;
+
+    col.r = texture(iChannel0,vec2(x+uv.x+0.001,uv.y+0.001)).x+0.05;
+    col.g = texture(iChannel0,vec2(x+uv.x+0.000,uv.y-0.002)).y+0.05;
+    col.b = texture(iChannel0,vec2(x+uv.x-0.002,uv.y+0.000)).z+0.05;
+    col.r += 0.08*texture(iChannel0,0.75*vec2(x+0.025, -0.027)+vec2(uv.x+0.001,uv.y+0.001)).x;
+    col.g += 0.05*texture(iChannel0,0.75*vec2(x+-0.022, -0.02)+vec2(uv.x+0.000,uv.y-0.002)).y;
+    col.b += 0.08*texture(iChannel0,0.75*vec2(x+-0.02, -0.018)+vec2(uv.x-0.002,uv.y+0.000)).z;
+
+    col = clamp(col*0.6+0.4*col*col*1.0,0.0,1.0);
+
+    float vig = (0.0 + 1.0*16.0*uv.x*uv.y*(1.0-uv.x)*(1.0-uv.y));
+	col *= vec3(pow(vig,0.3));
+
+    col *= vec3(0.95,1.05,0.95);
+	col *= 2.8;
+
+	float scans = clamp( 0.35+0.35*sin(3.5*iTime+uv.y*iResolution.y*1.5), 0.0, 1.0);
+
+	float s = pow(scans,1.7);
+	col = col*vec3( 0.4+0.7*s) ;
+
+    col *= 1.0+0.01*sin(110.0*iTime);
+	if (uv.x < 0.0 || uv.x > 1.0)
+		col *= 0.0;
+	if (uv.y < 0.0 || uv.y > 1.0)
+		col *= 0.0;
+
+	col*=1.0-0.65*vec3(clamp((mod(fragCoord.x, 2.0)-1.0)*2.0,0.0,1.0));
+
+    float comp = smoothstep( 0.1, 0.9, sin(iTime) );
+
+	// Remove the next line to stop cross-fade between original and postprocess
+//	col = mix( col, oricol, comp );
+
+    fragColor = vec4(col,1.0);
+}

src/renderer/shaders/test_shadertoy_invalid.glsl

@@ -0,0 +1,12 @@
+vec2 curve(vec2 uv)
+{
+	uv = (uv - 0.5) * 2.0;
+	uv *= 1.1;
+	uv.x *= 1.0 + pow((abs(uv.y) / 5.0), 2.0);
+	uv.y *= 1.0 + pow((abs(uv.x) / 4.0), 2.0);
+	uv  = (uv / 2.0) + 0.5;
+	uv =  uv *0.92 + 0.04;
+	return uv;
+}
+
+// Missing mainImage!

src/renderer/shadertoy.zig

@@ -0,0 +1,365 @@
+const std = @import("std");
+const builtin = @import("builtin");
+const assert = std.debug.assert;
+const Allocator = std.mem.Allocator;
+const ArenaAllocator = std.heap.ArenaAllocator;
+const glslang = @import("glslang");
+const spvcross = @import("spirv_cross");
+
+const log = std.log.scoped(.shadertoy);
+
+/// The target to load shaders for.
+pub const Target = enum { glsl, msl };
+
+/// Load a set of shaders from files and convert them to the target
+/// format. The shader order is preserved.
+pub fn loadFromFiles(
+    alloc_gpa: Allocator,
+    paths: []const []const u8,
+    target: Target,
+) ![]const [:0]const u8 {
+    var list = std.ArrayList([:0]const u8).init(alloc_gpa);
+    defer list.deinit();
+    errdefer for (list.items) |shader| alloc_gpa.free(shader);
+
+    for (paths) |path| {
+        const shader = try loadFromFile(alloc_gpa, path, target);
+        log.info("loaded custom shader path={s}", .{path});
+        try list.append(shader);
+    }
+
+    return try list.toOwnedSlice();
+}
+
+/// Load a single shader from a file and convert it to the target language
+/// ready to be used with renderers.
+pub fn loadFromFile(
+    alloc_gpa: Allocator,
+    path: []const u8,
+    target: Target,
+) ![:0]const u8 {
+    var arena = ArenaAllocator.init(alloc_gpa);
+    defer arena.deinit();
+    const alloc = arena.allocator();
+
+    // Load the shader fiel
+    const cwd = std.fs.cwd();
+    const file = try cwd.openFile(path, .{});
+    defer file.close();
+
+    // Read it all into memory -- we don't expect shaders to be large.
+    var buf_reader = std.io.bufferedReader(file.reader());
+    const src = try buf_reader.reader().readAllAlloc(
+        alloc,
+        4 * 1024 * 1024, // 4MB
+    );
+
+    // Convert to full GLSL
+    const glsl: [:0]const u8 = glsl: {
+        var list = std.ArrayList(u8).init(alloc);
+        try glslFromShader(list.writer(), src);
+        try list.append(0);
+        break :glsl list.items[0 .. list.items.len - 1 :0];
+    };
+
+    // Convert to SPIR-V
+    const spirv: []const u8 = spirv: {
+        // SpirV pointer must be aligned to 4 bytes since we expect
+        // a slice of words.
+        var list = std.ArrayListAligned(u8, @alignOf(u32)).init(alloc);
+        var errlog: SpirvLog = .{ .alloc = alloc };
+        defer errlog.deinit();
+        spirvFromGlsl(list.writer(), &errlog, glsl) catch |err| {
+            if (errlog.info.len > 0 or errlog.debug.len > 0) {
+                log.warn("spirv error path={s} info={s} debug={s}", .{
+                    path,
+                    errlog.info,
+                    errlog.debug,
+                });
+            }
+
+            return err;
+        };
+        break :spirv list.items;
+    };
+
+    // Convert to MSL
+    return switch (target) {
+        // Important: using the alloc_gpa here on purpose because this
+        // is the final result that will be returned to the caller.
+        .glsl => try glslFromSpv(alloc_gpa, spirv),
+        .msl => try mslFromSpv(alloc_gpa, spirv),
+    };
+}
+
+/// Convert a ShaderToy shader into valid GLSL.
+///
+/// ShaderToy shaders aren't full shaders, they're just implementing a
+/// mainImage function and don't define any of the uniforms. This function
+/// will convert the ShaderToy shader into a valid GLSL shader that can be
+/// compiled and linked.
+pub fn glslFromShader(writer: anytype, src: []const u8) !void {
+    const prefix = @embedFile("shaders/shadertoy_prefix.glsl");
+    try writer.writeAll(prefix);
+    try writer.writeAll("\n\n");
+    try writer.writeAll(src);
+}
+
+/// Convert a GLSL shader into SPIR-V assembly.
+pub fn spirvFromGlsl(
+    writer: anytype,
+    errlog: ?*SpirvLog,
+    src: [:0]const u8,
+) !void {
+    // So we can run unit tests without fear.
+    if (builtin.is_test) try glslang.testing.ensureInit();
+
+    const c = glslang.c;
+    const input: c.glslang_input_t = .{
+        .language = c.GLSLANG_SOURCE_GLSL,
+        .stage = c.GLSLANG_STAGE_FRAGMENT,
+        .client = c.GLSLANG_CLIENT_VULKAN,
+        .client_version = c.GLSLANG_TARGET_VULKAN_1_2,
+        .target_language = c.GLSLANG_TARGET_SPV,
+        .target_language_version = c.GLSLANG_TARGET_SPV_1_5,
+        .code = src.ptr,
+        .default_version = 100,
+        .default_profile = c.GLSLANG_NO_PROFILE,
+        .force_default_version_and_profile = 0,
+        .forward_compatible = 0,
+        .messages = c.GLSLANG_MSG_DEFAULT_BIT,
+        .resource = c.glslang_default_resource(),
+    };
+
+    const shader = try glslang.Shader.create(&input);
+    defer shader.delete();
+
+    shader.preprocess(&input) catch |err| {
+        if (errlog) |ptr| ptr.fromShader(shader) catch {};
+        return err;
+    };
+    shader.parse(&input) catch |err| {
+        if (errlog) |ptr| ptr.fromShader(shader) catch {};
+        return err;
+    };
+
+    const program = try glslang.Program.create();
+    defer program.delete();
+    program.addShader(shader);
+    program.link(
+        c.GLSLANG_MSG_SPV_RULES_BIT |
+            c.GLSLANG_MSG_VULKAN_RULES_BIT,
+    ) catch |err| {
+        if (errlog) |ptr| ptr.fromProgram(program) catch {};
+        return err;
+    };
+    program.spirvGenerate(c.GLSLANG_STAGE_FRAGMENT);
+    const size = program.spirvGetSize();
+    const ptr = try program.spirvGetPtr();
+    const ptr_u8: [*]u8 = @ptrCast(ptr);
+    const slice_u8: []u8 = ptr_u8[0 .. size * 4];
+    try writer.writeAll(slice_u8);
+}
+
+/// Retrieve errors from spirv compilation.
+pub const SpirvLog = struct {
+    alloc: Allocator,
+    info: [:0]const u8 = "",
+    debug: [:0]const u8 = "",
+
+    pub fn deinit(self: *const SpirvLog) void {
+        if (self.info.len > 0) self.alloc.free(self.info);
+        if (self.debug.len > 0) self.alloc.free(self.debug);
+    }
+
+    fn fromShader(self: *SpirvLog, shader: *glslang.Shader) !void {
+        const info = try shader.getInfoLog();
+        const debug = try shader.getDebugInfoLog();
+        self.info = "";
+        self.debug = "";
+        if (info.len > 0) self.info = try self.alloc.dupeZ(u8, info);
+        if (debug.len > 0) self.debug = try self.alloc.dupeZ(u8, debug);
+    }
+
+    fn fromProgram(self: *SpirvLog, program: *glslang.Program) !void {
+        const info = try program.getInfoLog();
+        const debug = try program.getDebugInfoLog();
+        self.info = "";
+        self.debug = "";
+        if (info.len > 0) self.info = try self.alloc.dupeZ(u8, info);
+        if (debug.len > 0) self.debug = try self.alloc.dupeZ(u8, debug);
+    }
+};
+
+/// Convert SPIR-V binary to MSL.
+pub fn mslFromSpv(alloc: Allocator, spv: []const u8) ![:0]const u8 {
+    return try spvCross(alloc, spvcross.c.SPVC_BACKEND_MSL, spv, null);
+}
+
+/// Convert SPIR-V binary to GLSL..
+pub fn glslFromSpv(alloc: Allocator, spv: []const u8) ![:0]const u8 {
+    // Our minimum version for shadertoy shaders is OpenGL 4.2 because
+    // Spirv-Cross generates binding locations for uniforms which is
+    // only supported in OpenGL 4.2 and above.
+    //
+    // If we can figure out a way to NOT do this then we can lower this
+    // version.
+    const GLSL_VERSION = 420;
+
+    const c = spvcross.c;
+    return try spvCross(alloc, c.SPVC_BACKEND_GLSL, spv, (struct {
+        fn setOptions(options: c.spvc_compiler_options) error{SpvcFailed}!void {
+            if (c.spvc_compiler_options_set_uint(
+                options,
+                c.SPVC_COMPILER_OPTION_GLSL_VERSION,
+                GLSL_VERSION,
+            ) != c.SPVC_SUCCESS) {
+                return error.SpvcFailed;
+            }
+        }
+    }).setOptions);
+}
+
+fn spvCross(
+    alloc: Allocator,
+    backend: spvcross.c.spvc_backend,
+    spv: []const u8,
+    comptime optionsFn_: ?*const fn (c: spvcross.c.spvc_compiler_options) error{SpvcFailed}!void,
+) ![:0]const u8 {
+    // Spir-V is always a multiple of 4 because it is written as a series of words
+    if (@mod(spv.len, 4) != 0) return error.SpirvInvalid;
+
+    // Compiler context
+    const c = spvcross.c;
+    var ctx: c.spvc_context = undefined;
+    if (c.spvc_context_create(&ctx) != c.SPVC_SUCCESS) return error.SpvcFailed;
+    defer c.spvc_context_destroy(ctx);
+
+    // It would be better to get this out into an output parameter to
+    // show users but for now we can just log it.
+    c.spvc_context_set_error_callback(ctx, @ptrCast(&(struct {
+        fn callback(_: ?*anyopaque, msg_ptr: [*c]const u8) callconv(.C) void {
+            const msg = std.mem.sliceTo(msg_ptr, 0);
+            std.log.warn("spirv-cross error message={s}", .{msg});
+        }
+    }).callback), null);
+
+    // Parse the Spir-V binary to an IR
+    var ir: c.spvc_parsed_ir = undefined;
+    if (c.spvc_context_parse_spirv(
+        ctx,
+        @ptrCast(@alignCast(spv.ptr)),
+        spv.len / 4,
+        &ir,
+    ) != c.SPVC_SUCCESS) {
+        return error.SpvcFailed;
+    }
+
+    // Build our compiler to GLSL
+    var compiler: c.spvc_compiler = undefined;
+    if (c.spvc_context_create_compiler(
+        ctx,
+        backend,
+        ir,
+        c.SPVC_CAPTURE_MODE_TAKE_OWNERSHIP,
+        &compiler,
+    ) != c.SPVC_SUCCESS) {
+        return error.SpvcFailed;
+    }
+
+    // Setup our options if we have any
+    if (optionsFn_) |optionsFn| {
+        var options: c.spvc_compiler_options = undefined;
+        if (c.spvc_compiler_create_compiler_options(compiler, &options) != c.SPVC_SUCCESS) {
+            return error.SpvcFailed;
+        }
+
+        try optionsFn(options);
+
+        if (c.spvc_compiler_install_compiler_options(compiler, options) != c.SPVC_SUCCESS) {
+            return error.SpvcFailed;
+        }
+    }
+
+    // Compile the resulting string. This string pointer is owned by the
+    // context so we don't need to free it.
+    var result: [*:0]const u8 = undefined;
+    if (c.spvc_compiler_compile(compiler, @ptrCast(&result)) != c.SPVC_SUCCESS) {
+        return error.SpvcFailed;
+    }
+
+    return try alloc.dupeZ(u8, std.mem.sliceTo(result, 0));
+}
+
+/// Convert ShaderToy shader to null-terminated glsl for testing.
+fn testGlslZ(alloc: Allocator, src: []const u8) ![:0]const u8 {
+    var list = std.ArrayList(u8).init(alloc);
+    defer list.deinit();
+    try glslFromShader(list.writer(), src);
+    return try list.toOwnedSliceSentinel(0);
+}
+
+test "spirv" {
+    const testing = std.testing;
+    const alloc = testing.allocator;
+
+    const src = try testGlslZ(alloc, test_crt);
+    defer alloc.free(src);
+
+    var buf: [4096 * 4]u8 = undefined;
+    var buf_stream = std.io.fixedBufferStream(&buf);
+    const writer = buf_stream.writer();
+    try spirvFromGlsl(writer, null, src);
+}
+
+test "spirv invalid" {
+    const testing = std.testing;
+    const alloc = testing.allocator;
+
+    const src = try testGlslZ(alloc, test_invalid);
+    defer alloc.free(src);
+
+    var buf: [4096 * 4]u8 = undefined;
+    var buf_stream = std.io.fixedBufferStream(&buf);
+    const writer = buf_stream.writer();
+
+    var errlog: SpirvLog = .{ .alloc = alloc };
+    defer errlog.deinit();
+    try testing.expectError(error.GlslangFailed, spirvFromGlsl(writer, &errlog, src));
+    try testing.expect(errlog.info.len > 0);
+}
+
+test "shadertoy to msl" {
+    const testing = std.testing;
+    const alloc = testing.allocator;
+
+    const src = try testGlslZ(alloc, test_crt);
+    defer alloc.free(src);
+
+    var spvlist = std.ArrayListAligned(u8, @alignOf(u32)).init(alloc);
+    defer spvlist.deinit();
+    try spirvFromGlsl(spvlist.writer(), null, src);
+
+    const msl = try mslFromSpv(alloc, spvlist.items);
+    defer alloc.free(msl);
+}
+
+test "shadertoy to glsl" {
+    const testing = std.testing;
+    const alloc = testing.allocator;
+
+    const src = try testGlslZ(alloc, test_crt);
+    defer alloc.free(src);
+
+    var spvlist = std.ArrayListAligned(u8, @alignOf(u32)).init(alloc);
+    defer spvlist.deinit();
+    try spirvFromGlsl(spvlist.writer(), null, src);
+
+    const glsl = try glslFromSpv(alloc, spvlist.items);
+    defer alloc.free(glsl);
+
+    // log.warn("glsl={s}", .{glsl});
+}
+
+const test_crt = @embedFile("shaders/test_shadertoy_crt.glsl");
+const test_invalid = @embedFile("shaders/test_shadertoy_invalid.glsl");