Six findings, all of them libakgl's, all closed. The memcheck run is clean.
Not one of the four json_load_file calls in src/ was ever matched by a
json_decref, so every asset load abandoned 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. Each loader now releases it in the
CLEANUP block it already had, on the success path as well as the failure
one. akgl_registry_load_properties needed its loop moved inside the ATTEMPT
block first: props is a borrowed reference into the document and is read
after the block ends, so every exit from that loop leaked the whole tree.
akgl_get_property copied a fixed AKGL_MAX_STRING_LENGTH bytes out of what
SDL handed back, which is SDL's strdup of the value -- four bytes for
"0.0". That read up to 4 KiB past the end of somebody else's allocation on
every property read, returned whatever was there past the terminator, and
would have faulted on a value that landed at the end of a page. It copies
the value and its terminator now, and refuses a value too long for an
akgl_String rather than truncating it into an unterminated buffer. The
header note that described the overread as a quirk describes correct
behaviour instead.
The four savegame name tables wrote a fixed-width field starting at the
registry key, and SDL sizes that allocation to the name. They read past it
on every entry and put what they found into the save file: up to half a
kilobyte of this process's heap per registered object, in a file a player
might send to somebody. They stage through a zeroed buffer now, and a
negative-array-size typedef fails the build if a table's width ever outgrows
it.
akgl_controller_list_keyboards never freed the array SDL_GetKeyboards
allocated for it.
A font could be opened and published and never handed back -- there was no
way to close one, so a game that changed fonts between scenes leaked ten
kilobytes each time, and loading over a live name leaked the font it
displaced. akgl_text_unloadfont is that missing half, and akgl_text_loadfont
calls it when it replaces a name, after the new font has opened so a failed
reload leaves the caller with the font they had. A new public symbol takes
the version to 0.4.0 and the soname with it: an 0.3 consumer cannot be
handed this library and told it is the same ABI.
tests/registry.c fills a destination with a sentinel and asserts the bytes
past the terminator survive a read, which fails against the old copy.
tests/text.c covers unload, double unload, unloading a name that was never
registered, and replacement closing the displaced font. The JSON releases
have no test of their own and cannot sensibly have one -- nothing in the
public API can observe a jansson refcount -- so the memcheck run is their
test, which is an argument for gating it rather than against.
The two remaining findings are in deps/semver's own unit test, which is
vendored. They are suppressed by function name, so a rewrite of those cases
comes back as a finding.
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>
All ten items under "API gaps blocking akbasic" are closed. Items 7, 9 and
10 add public symbols and item 9 adds three fields to akgl_AudioVoice, so
the version and the soname go with them -- an 0.2 consumer cannot be handed
this library and told it is compatible.
Also refreshes the coverage numbers against a fresh run (79.6% line, 87.2%
function, all 21 suites), records the mutation smoke runs over the two files
that changed most, and files one new defect the text tests turned up:
akgl_text_rendertextat refuses the empty string while akgl_text_measure
accepts it.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
akbasic built a line editor on akgl_controller_poll_key() and found the ring
carries a keycode and nothing else. No shifted character is reachable, so a
BASIC string literal cannot be typed at all.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Writing an honest @throws list means reading the implementation against the
contract its header claimed, and the two disagreed in more places than the
status codes. Recorded as items 14-25 under Defects, ordered by blast radius,
each with file, line, consequence and what closing it would touch. They are
also noted inline as @note or @warning on the function concerned, so a reader
of the generated documentation finds them without coming here first.
The three that matter most:
- akgl_path_relative returns from inside its ENOENT handler, skipping the
RELEASE_ERROR that FINISH carries, so it leaks one of libakerror's 128
error-context slots per call. That is not a rare path - it is the ordinary
one, taken whenever an asset names a neighbour relative to its own file.
- akgl_sprite_load_json writes json_array_size() entries into a 16-element
frameids array with no bound check, and does it through a uint32_t * cast of
a uint8_t *. The same unbounded shape appears twice more in the tilemap
loader, for objects per layer and tilesets per map.
- akgl_heap_release_character zeroes the character without walking its
state-to-sprite map, so every sprite reference it took is lost and the SDL
property set holding the map is leaked. Releasing characters between levels
exhausts the sprite pool. akgl_character_state_sprites_iterate exists to do
that walk and is not called from there.
Also: the background music never loops, because MIX_PROP_PLAY_LOOPS_NUMBER is
set on property set 0 rather than one akgl_load_start_bgm owns; and
akgl_get_json_with_default defaults on AKERR_INDEX while the array accessors
report a short array as AKERR_OUTOFBOUNDS, so the optional-element form
silently does not work for arrays.
No code changes. Every src/*.c:NNN citation in the new section was checked
against the file after the preceding commit shifted the line numbers.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
akbasic has now consumed all four of the API gaps this file listed as blocking
it: its text sink, graphics backend, audio backend and input backend are written
and tested, and the BASIC 7.0 graphics, sound and console verbs work against
them. Doing that turned up six things worth fixing here -- five of which still
are.
Four are workarounds akbasic had to write to build at all, and each is commented
at its site over there with the words "filed upstream":
- An embedded libakgl requires SDL, SDL_image, SDL_mixer, SDL_ttf and jansson to
be installed, while vendoring all five. The dependency block is gated on being
top-level, so an add_subdirectory() consumer takes the find_package path and
fails with the submodules sitting right there in deps/.
- include/akgl/controller.h does not compile on its own: it declares handlers
taking an akgl_Actor * and includes nothing that declares the type. src/
never notices because it includes game.h first.
- akgl_render_init2d() fuses two separable jobs -- creating a window, and
installing the backend vtable -- so a caller who already owns an SDL_Renderer
has no way to get the second without the first. That caller is exactly the
embedding host this section exists to serve.
- akgl_text_rendertextat() dereferences renderer->draw_texture with no check,
which is the state item 7 leaves a host in. Same class of defect the draw
commit added a test for; the text path still has it.
The fifth is a real API gap: SOUND's frequency sweep has no equivalent, and it
cannot be faked in the interpreter without tying audible pitch to the host's
frame rate, so it has to be advanced on the mixer's own frame counter.
The sixth was that 42b60f7 shipped 22 public symbols under an unchanged VERSION
and soname. 1066ac7 fixed that independently while this was being written -- the
two crossed rather than one following the other -- so it is filed as resolved
rather than dropped. The reason is the part worth keeping: akbasic could not
feature-test for the new API and had to pin libakgl by submodule commit in its
README until that landed, which is the concrete cost of an additive release
under an unchanged soname.
All five outstanding items were re-verified against this tree after the rebase,
not just carried forward: controller.h still includes neither actor.h nor a
forward declaration, render_init2d still calls SDL_CreateWindowAndRenderer
before installing the vtable, text.c:62 still reaches through draw_texture
unchecked, the vendored-dependency block is still gated on being top-level, and
there is still no akgl_audio_sweep.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
A mutation smoke run scored src/audio.c at 50% and src/draw.c at 70% and named
three real holes:
- Nothing asserted that an unconfigured voice is audible, which is the entire
reason the voice table defaults to a square wave at full level instead of a
zeroed struct. Deleting the defaults survived.
- No envelope test used a non-zero attack and decay together, so the decay
measuring from the start of the note rather than from the end of the attack
survived.
- The circle was only checked at its four axis points, which a mis-signed
octant reflection survives. It now checks that every plotted pixel has a
mirror in the other three quadrants.
Also adds a backend that exists but was never given an SDL_Renderer, which is
the state a host is in between allocating one and initializing it; every draw
entry point has to report it rather than dereference it.
audio 50% -> 75%, draw 70% -> 90%. The survivors are recorded in TODO.md and
are honestly untestable from here.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
77.2% line and 83.1% function coverage across 2637 lines, up from 72.2% / 78.6%,
with the three new suites at 91% (audio), 95% (draw) and 58% (text). All 19
suites pass: the note recording character as an intentional failure was stale,
and neither tests/character.c nor src/character.c has changed since it was
written.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
There was no audio API at all: SDL3_mixer was vendored and AKGL_REGISTRY_MUSIC
was declared, but nothing opened a mixer, and a mixer plays recordings anyway.
BASIC 7.0's SOUND, PLAY, ENVELOPE and VOL describe notes, so this adds a tone
generator -- three voices, five waveforms, an ADSR envelope each, mixed to one
float stream and fed to an SDL_AudioStream.
The voice table works whether or not a device is open. akgl_audio_init connects
it to one; a host that owns its own audio pipeline can call akgl_audio_mix
instead. That is also what makes the tests deterministic: a device pulls samples
on SDL's audio thread whenever it likes, so tests/audio.c mixes by hand and only
opens a device in its last test.
Phase is computed from the frame counter rather than accumulated, because a
float increment of hz/44100 added 44100 times a second walks a held note off
pitch. A voice that has never been configured defaults to a square wave at full
level, since a zeroed voice would have a sustain of 0.0 and make no sound with
no error to explain why. Voices summing past full scale are clamped rather than
scaled, so one voice plays at the level it was asked for.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
BASIC 7.0's graphics verbs -- DRAW, BOX, CIRCLE, PAINT, SSHAPE and GSHAPE -- all
plot against the current screen right now rather than adding to a scene, and
draw.h had exactly one function in it. This adds akgl_draw_point, _line, _rect,
_filled_rect, _circle, _flood_fill, _copy_region and _paste_region, each taking
the akgl_RenderBackend the host already initialized.
Color is an argument rather than a current-color global, so there is no second
copy of state to disagree with the caller's own, and each call restores the
renderer's draw color when it is done. SDL3 has no circle and no flood fill: the
circle is a midpoint circle, and the fill reads the target back, walks the
region with a fixed 4096-entry span stack, and blits back only the bounding box
of what changed. Exhausting that stack reports AKERR_OUTOFBOUNDS and says in the
header that the region is left partially filled.
tests/draw.c draws into a 64x64 software renderer under the dummy video driver
and reads the pixels back, so it needs no display and no offscreen harness.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
akbasic's GET and GETKEY ask "is there a keystroke waiting, yes or no" and must
not require the interpreter to pump SDL events itself. akgl_controller_poll_key
drains one key per call from a fixed 32-entry ring that
akgl_controller_handle_event fills, so the host keeps pumping and the embedded
interpreter reads at its own pace. akgl_controller_flush_keys discards a
backlog.
The recording happens before the control-map scan, not after: that scan returns
as soon as a binding claims the event, so recording afterwards would have
dropped exactly the keys a game also acts on. A full buffer refuses the newest
keystroke rather than overwriting the oldest, so a caller reading a line gets
what was typed first.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
akbasic needs the advance width of one character cell to build a terminal-style
text surface, and the wrapped size to know where a string crosses the right
margin. akgl_text_measure and akgl_text_measure_wrapped report both over
TTF_GetStringSize and TTF_GetStringSizeWrapped; neither needs a renderer, so
this half of the text subsystem is testable without the offscreen harness.
A negative wrap length is refused with AKERR_OUTOFBOUNDS: SDL_ttf reads it as a
very large unsigned width and quietly stops wrapping, which would return a
measurement that is wrong rather than an error the caller can see.
Also fixes akgl_text_loadfont, which checked name twice and passed an unchecked
filepath to TTF_OpenFont (TODO item 39).
The fixture font is a 10 KB monospaced ASCII subset of Liberation Mono, renamed
because "Liberation" is a Reserved Font Name; see
tests/assets/akgl_test_mono.LICENSE.txt. Monospaced so the suite can assert
width("AAAA") == 4 * width("A") rather than hardcode glyph metrics.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
akbasic is a C port of the BASIC interpreter being built to link into libakgl as
a scripting engine for game authors. Its interpreter core is complete and passes
its whole acceptance corpus against a stdio text sink; four gaps here block the
libakgl-backed sink and the graphics, sound and console verbs of Commodore
BASIC 7.0.
Filed rather than worked around downstream, because growing libakgl to serve a
consumer is the wanted outcome. Each entry says what the BASIC verb needs, what
the akgl_* entry point should look like, and what would cover it.
Only the first blocks work already designed and waiting. text.h can load a font
and render a string and cannot say how large that string will be, so there is no
way to build a terminal-style surface on it -- a cursor needs the advance width
of a cell and wrapping needs to know where a string crosses the margin. The
reference interpreter derived its whole character grid from SDL2_ttf's
SizeUTF8("A"). akgl_text_measure() over TTF_GetStringSize closes it.
The other three block verbs not yet started: draw.h declares one function and
src/draw.c is at 0%, so DRAW/BOX/CIRCLE/PAINT have nothing to plot with; there
is no audio API at all despite SDL3_mixer being vendored, and PLAY/ENVELOPE want
a synthesised voice SDL3_mixer does not provide; and controller.h is built
around event handlers the host pumps, which suits a game loop but not GET/GETKEY
-- those ask whether a keystroke is waiting and must not require the interpreter
to own the event loop, which goal 3 forbids anyway.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Defects #13: gcovr searches for gcov data under --root, which
CMakeLists.txt:208 and :223 set to the source directory, so any
instrumented build tree left there is merged into the report. The
--object-directory argument beside each does not narrow the search; it
only maps gcda files back to the compiler's working directory.
A leftover build-coverage/ therefore fails coverage_reset before any
test runs, with gcovr unable to reconcile a function reported on two
different lines. The build*/ entry in .gitignore keeps the offending
tree out of git status, so the state is easy to get into and hard to
spot.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Add an "Internal consistency" section to TODO.md: 41 numbered findings
from a sweep of src/ and include/, each cited by file:line and grouped by
blast radius. These are convention problems, not a re-audit of behavior,
but several have live functional consequences:
- AKGL_TIME_ONESEC_MS is misnamed. Its value is nanoseconds-per-
millisecond, but akgl_game_state_lock uses it as a one-second budget
against a 100ms delay loop, so the lock blocks ~16 minutes.
- SUCCEED_RETURN inside an ATTEMPT block at tilemap.c:52 bypasses the
CLEANUP two lines below, leaking two heap strings on every successful
tilemap property lookup, against a 256-entry pool.
- AKGL_ACTOR_STATE_STRING_NAMES is declared [33] and defined [32], and
names bits 11 and 12 as UNDEFINED where actor.h defines MOVING_IN and
MOVING_OUT, so no character JSON can bind a sprite to those states.
- 19 non-static functions are defined in src/ but declared in no header,
and akgl_game_init_screen is declared but never defined.
- AKGL_COLLIDE_RECTANGLES has unbalanced parentheses and cannot compile.
- text.c:19 checks `name` twice and never validates `filepath`.
Item 30 is marked resolved by the reindent in the preceding commit.
Also add Defects #12: a failed controller-DB fetch silently destroys the
tracked fallback. include/akgl/SDL_GameControllerDB.h is committed on
purpose so the library still builds if upstream disappears, but
mkcontrollermappings.sh has no `set -e` and never checks curl's exit
status -- on failure it overwrites the good header with
AKGL_SDL_GAMECONTROLLER_DB_LEN 0 and an empty initializer, and exits 0.
Verified against an unresolvable host. CMakeLists.txt:89-93 compounds
this: the custom command's OUTPUT is relative, so CMake resolves it
against the binary dir while the script writes to the source dir, leaving
the command permanently out of date and re-running on every build.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Cover control map capacity, the default binding set, keyboard and gamepad
dispatch including cross-device rejection, the dpad handlers, and device
add and remove against the dummy SDL drivers. Synthesize SDL events directly
rather than pumping the event queue, so the suite needs no real hardware.
Fix the three-way appstate check in the four gamepad handlers, which tested
appstate in place of both the event pointer and the looked-up player actor,
so a null event or a missing player was dereferenced instead of reported.
Record the new coverage status in TODO.md along with the seven defects the
suites uncovered and fixed, eleven that remain open, and the reason the
build tree has to precede LD_LIBRARY_PATH for the CTest run.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Record the 39.6 percent line and 44.3 percent function coverage baseline
measured with the AKGL_COVERAGE build, and plan twelve test suites to raise
it to roughly 70 percent lines and 80 percent functions. Tier the work by
what each suite needs: pure logic suites for physics, json_helpers, heap,
and savegame handling that require no renderer; renderer suites gated behind
a shared offscreen harness; and controller and frame-loop coverage deferred.
Note that vendored dependency shared objects are absent from the loader path
in a fresh out-of-tree build, which reports every test as zero coverage, and
record ten suspected defects found while reading the uncovered code so the
new tests assert correct behavior rather than pinning the current one.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
