This introduces a new package `src/synthetic` for generating synthetic
data, currently primarily for benchmarking but other use cases can
emerge.
The synthetic package exports a runtime-dispatched type `Generator` that
can generate data of various types. To start, we have a bytes, utf8,
and OSC generator. The goal of each generator is to expose knobs to tune the
probabilities of various outcomes. For example, the UTF-8 generator has
a knob to tune the probability of generating 1, 2, 3, or 4-byte UTF-8
sequences.
Ultimately, the goal is to be able to collect probability data
empirically that we can then use for benchmarks so we can optimize
various parts of the codebase on real-world data shape distributions.
This commit adds a few new mode flags to the `bench-stream` program
to generator synthetic OSC sequences. The new modes are `gen-osc`,
`gen-osc-valid`, and `gen-osc-invalid`. The `gen-osc` mode generates
equal parts valid and invalid OSC sequences, while the suffixed variants
are for generating only valid or invalid sequences, respectively.
This commit also fixes our build system to actually be able to build the
benchmarks. It turns out we were just rebuilding the main Ghostty binary
for `-Demit-bench`. And, our benchmarks didn't run under Zig 0.14, which
is now fixed.
An important new design I'm working towards in this commit is to split
out synthetic data generation to a dedicated package in
`src/bench/synth` although I'm tempted to move it to `src/synth` since
it may be useful outside of benchmarks.
The synth package is a work-in-progress, but it contains a hint of
what's to come. I ultimately want to able to generate all kinds of
synthetic data with a lot of knobs to control dimensionality (e.g. in
the case of OSC sequences: valid/invalid, length, operation types,
etc.).
Related to #3224
Previously, Ghostty used a static API for async event handling: io_uring
on Linux, kqueue on macOS. This commit changes the backend to be dynamic
on Linux so that epoll will be used if io_uring isn't available, or if
the user explicitly chooses it.
This introduces a new config `async-backend` (default "auto") which can
be set by the user to change the async backend in use. This is a
best-effort setting: if the user requests io_uring but it isn't
available, Ghostty will fall back to something that is and that choice
is up to us.
Basic benchmarking both in libxev and Ghostty (vtebench) show no
noticeable performance differences introducing the dynamic API, nor
choosing epoll over io_uring.
Instead of "polling" to see if a quit timer has expired, start a single
timer that expires after the confiugred delay when no more surfaces are
open. That timer can be cancelled if necessary.
