The perf suites were already running in CI, because they are ordinary CTest
tests -- but only in the coverage job, which is a Debug build with gcov
instrumentation. That is the one place their numbers cannot mean anything,
and it was spending full-scale iteration counts to prove it. They now run
there at AKGL_BENCH_SCALE=0.02, for path coverage alone, which takes the
whole coverage run to four seconds.
The `performance` job is where the timings are taken: RelWithDebInfo, full
scale, `ctest -L perf`, which is the only configuration in which
tests/benchutil.h enforces a budget. It keeps the tables as an artifact
whether it passes or fails -- a red run says which measurement moved, and a
green one is the next baseline for PERFORMANCE.md. The step uses --verbose
rather than --output-on-failure, because --output-on-failure prints nothing
at all for a passing benchmark, and sets pipefail, without which the status
reaching the runner is tee's and a blown budget reads as a pass.
The `memory_check` job runs scripts/memcheck.sh over every suite under
valgrind and keeps the logs. It carries continue-on-error, and that is a
deliberate, temporary lie: the six findings in TODO.md "Memory checking" are
real and all of them are libakgl's, so gating on it today would only teach
everyone to ignore a red build. The flag comes off with the commit that
closes the last item, and item 34 -- the missing json_decref in all four
asset loaders -- is four lines.
Both new jobs were run locally against a clean copy of the tree with the
exact commands the workflow uses.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
`cmake --build build --target memcheck` runs every registered CTest suite
under valgrind. There are no new test programs, and there should not be any:
tests/benchutil.h notices valgrind in LD_PRELOAD and drops the benchmark
scale to 0.0005, which turns the perf suites into the broadest path coverage
in the tree at a cost valgrind can survive. A benchmark walks one path a
hundred thousand times; a leak check wants every path walked once. Same
binaries, one flag apart, and the whole run is about thirty seconds.
scripts/memcheck.sh wraps `ctest -T memcheck` because that command records
defects and still exits 0, which cannot gate anything. It also forces the
headless drivers, so the vendor GPU stack is never loaded -- that removes
thousands of unfixable findings inside amdgpu_dri.so without suppressing
anything, and it is what the suites are written for anyway. Only definite
losses, invalid accesses and uninitialised reads count; "still reachable" is
what SDL and FreeType keep for the process lifetime and says nothing about
this library. The three suppressions in scripts/valgrind.supp are each a
decision that a finding belongs to somebody else.
Six defects, all filed in TODO.md under "Memory checking", the first of
which is the one that matters: not one of the four json_load_file calls in
src/ is ever matched by a json_decref, so every asset load abandons its
parsed document -- 1.5 KB per sprite, 2.1 KB per character, 9 KB per load of
the 2x2 fixture map, and on the order of a megabyte for a real one. A game
that reloads a level on death leaks a level's worth of JSON every time.
akgl_get_property also reads up to 4 KiB past the end of every property
value it copies, and the savegame name tables read past the end of every
registry key and write what they find into the file.
tests/util.c zeroes three fixtures it used to leave as stack garbage. The
library was never at fault there -- the tests handed it uninitialised floats
and then made one real call with them -- but sixteen findings of noise in a
new gate is how a gate gets ignored.
The unit suites' TIMEOUT goes from 30 to 300 because CTest applies the same
property to the checked run, where everything is twenty times slower.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
tests/perf.c and tests/perf_render.c drive the hot paths hard enough to
time them -- pool acquire and release at both ends of the scan, the
registry, the state-to-sprite lookup, the per-actor update and render, the
physics sweep, an all-pairs collision sweep, the JSON accessors, path
resolution, the drawing primitives, text, asset loading, a screenful of
tiles, and a whole frame through akgl_game_update. They are registered like
any other suite but carry the `perf` label, so `ctest -L perf` runs only
them and `-LE perf` leaves them out. Every measurement is held to a budget
at roughly ten times the recorded baseline, enforced only in an optimized
build at full scale.
Three things had to be got right before the numbers meant anything, and
each was wrong first:
- No error checking inside the clock. PASS and CATCH call
akerr_valid_error_address, which walks AKERR_ARRAY_ERROR -- more work than
several of the calls being measured.
- Flush the renderer before stopping it. SDL batches, so the first version
measured queueing, reported a tilemap frame 250 times faster than it is,
and paid the real cost at teardown inside SDL_DestroyTexture.
- A drawing benchmark needs a raw-SDL control doing the same pixel work
with the same access pattern. With one, akgl_tilemap_draw turns out to
cost 0.2% of the frame it appeared to own: 16.26 ms against a control's
16.23 ms for the same 1200 blits. The rasterizer is the frame.
PERFORMANCE.md records the baseline, the frame budget it adds up to, and
what the numbers say -- including that the string pool's acquire is 64x
slower full than empty because it is a megabyte of PATH_MAX buffers, that a
handled missing-sprite condition costs nine times the update it replaces,
that text rasterizes and throws away a texture on every call, and that 28 MB
of the library's BSS is one akgl_Tilemap.
TODO.md gains a Performance section: five defects the stress tests found
that the unit suites do not reach (a tilemap load leaking five pooled
strings, two JSON accessors that turn pool exhaustion into a segfault
rather than AKGL_ERR_HEAP, akgl_game_update crashing without
akgl_game_init, and its actor update sweep running once per tilemap layer),
and eighteen targets with today's number and whether it is met.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
AGENTS.md described style in one paragraph that said little more than
"follow the surrounding code", which was not actionable once files had
drifted apart. Replace it with an explicit specification.
The canonical style is Emacs cc-mode "stroustrup" with tabs enabled: a
4-column offset with 8-column tabs, so depth 1 is four spaces, depth 2 is
one tab, depth 3 is a tab plus four spaces. Most of src/ already followed
this; what looked like randomly mixed tabs and spaces was simply cc-mode
output. Document the byte ladder explicitly, since editors that expand
tabs or assume a 4-column tab silently corrupt it.
Rules beyond indentation are derived from counted majorities in the
existing code rather than invented: errctx over e (92 to 45), padded
control parens (140 to 7), pointer binding to the identifier (486 to 5),
and always-brace (only four unbraced bodies exist). Brace and else
placement are called out as house conventions that cc-mode does not
enforce, so "run the indenter" and "follow the guide" cannot conflict.
Also record the naming, error-handling, and API-surface rules that a
consistency sweep showed were being broken silently -- in particular that
a *_RETURN macro inside an ATTEMPT block skips CLEANUP.
Add the tooling to apply it:
.dir-locals.el applies the style to every c-mode buffer
scripts/reindent.el batch reindent via Emacs
scripts/reindent.sh reindent or --check the tree
scripts/hooks/pre-commit reindents staged sources
reindent.el deliberately avoids Emacs' tabify: its tabify-regexp is
" [ \t]+", which is not anchored to the start of a line, so it rewrites
runs of spaces anywhere and destroys the hand-aligned value columns in
the bit-flag tables in actor.h and iterator.h. Only leading whitespace is
ever rewritten, and lines starting inside a string literal are skipped.
The hook checks staged content rather than the working tree, so what is
committed is what was verified. It re-stages a fixed file only when the
index and working tree agree, otherwise a partial `git add -p` would
sweep unstaged work into the commit.
Enable with: git config core.hooksPath scripts/hooks
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Instrument libakgl in opt-in coverage builds, use CTest fixtures to reset counters and emit Cobertura XML and detailed HTML reports, and upload those reports from Gitea CI. Document the local coverage workflow for contributors.
Co-authored-by: Codex (GPT-5) <noreply@openai.com>
Port the scratch-copy mutation runner used by libakerror and libakstdlib, covering libakgl source files with sampling, thresholds, JUnit reports, and test timeouts. Add a CMake target and documentation, preserve the intentionally failing character test exclusion, and migrate the sprite test to the renderer backend so mutation baselines are green.
Co-authored-by: Codex (GPT-5) <noreply@openai.com>
