fix(gamefont): Ceil font glyph buffer size to the actual glyph size to prevent a buffer write overflow - #3268
Conversation
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
PR Summary by QodoPrevent large glyphs from overflowing font buffers
AI Description
Diagram
High-Level Assessment
Files changed (3)
|
Code Review by Qodo
1. Oversized interface text disappears
|
|
| Filename | Overview |
|---|---|
| Core/GameEngine/Source/GameClient/GUI/GameFont.cpp | Clamps positive font requests to 512 points at the common font-creation boundary. |
| Core/Libraries/Source/WWVegas/WW3D2/render2dsentence.cpp | Allocates glyph buffers according to required capacity, checks their recorded lengths, and releases their pixel arrays correctly. |
| Core/Libraries/Source/WWVegas/WW3D2/render2dsentence.h | Replaces the fixed pixel array with a value descriptor containing allocated length and pixel pointer. |
Flowchart
%%{init: {'theme': 'neutral'}}%%
flowchart LR
Request[Font request] --> Clamp[Clamp point size to 512]
Clamp --> Glyph[Rasterize glyph]
Glyph --> Size[Compute width × height]
Size --> Capacity{Current buffer has capacity?}
Capacity -->|Yes| Store[Store glyph pixels]
Capacity -->|No| Allocate[Allocate max of default capacity and glyph size]
Allocate --> Store
Store --> Destroy[Font destruction]
Destroy --> Free[Delete each pixel array]
Reviews (6): Last reviewed commit: "refactor(gamefont): Store glyph buffer d..." | Re-trigger Greptile
dbebfc4 to
33b6309
Compare
|
Can't we use At what font size did it overflow the original buffer? Was this leading to crashes? |
Theoretically somewhere around 130 to 150 point, can also be triggered by something like a custom map with
Each glyph caches a raw pointer into its slab |
|
Do this effect PR #3231 ? |
eae556c to
cb3ed94
Compare
1ce2f06 to
099cb78
Compare
* bugfix(gamewindow): Remove destroyed windows from the modal stack and prevent duplicate modals for the same window (TheSuperHackers#3224) * feat(commandline): Add working directory command line options (TheSuperHackers#3149) Append -useCwd to apply the startup working directory, -setCwd "path" to apply a custom working directory, otherwise it falls back to the default executable working directory * bugfix(neutronmissile): Fix and improve Nuke Missile damage for large objects inside the outer blast radius (TheSuperHackers#3161) * bugfix(dozeraiupdate): Fix issue where builders could resume completed tasks after being disabled (TheSuperHackers#2793) * refactor(milesaudiomanager): Use consistent variable names for PlayingAudio in MilesAudioManager (TheSuperHackers#3254) * refactor(milesaudiomanager): Simplify MilesAudioManager::notifyOfAudioCompletion() (TheSuperHackers#3254) * refactor(milesaudiomanager): Simplify MilesAudioManager::findLowestPrioritySound() (TheSuperHackers#3254) * bugfix(milesaudiomanager): Fix premature 2d and 3d sound cancellations from MilesAudioManager::stopAudioEvent() (TheSuperHackers#3254) * bugfix(milesaudiomanager): No longer use stopped audio in queries and updates (TheSuperHackers#3254) * refactor(bink): Replace the Bink SDK stub with a Bink runtime loader (TheSuperHackers#3272) The Bink SDK stub was linked as an import library, so binkw32.dll had to be resolvable while the process image was still loading, long before WinMain and therefore long before the command line was parsed. That is why -setCwd could not point a build at a retail installation: the working directory it selects is set far too late to influence how the library is found. BinkLoader loads binkw32.dll explicitly once BinkVideoPlayer is initialized, at which point the working directory is final. The Bink functions declared in bink.h are now ordinary functions that forward to the matching export of the loaded module, so no call site changes. An unresolved function returns the same neutral value the stub library returned, which means a missing binkw32.dll disables video playback instead of preventing the game from starting. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> * refactor(miles): Replace the Miles SDK stub with a Miles runtime loader (TheSuperHackers#3272) The Miles SDK stub was linked as an import library, so mss32.dll had to be resolvable while the process image was still loading, long before WinMain and therefore long before the command line was parsed. That is why -setCwd could not point a build at a retail installation: the working directory it selects is set far too late to influence how the library is found. MilesLoader loads mss32.dll explicitly once the audio device is opened, at which point the working directory is final. The Miles functions declared in mss/mss.h are now ordinary functions that forward to the matching export of the loaded module, so no call site changes. An unresolved function returns the same neutral value the stub library returned, which means a missing mss32.dll turns audio off instead of preventing the game from starting. Nine declarations were dropped along the way, because the retail mss32.dll does not export them and nothing has called them since they were replaced by their volume_pan counterparts: AIL_sample_volume, AIL_set_sample_volume, AIL_sample_pan, AIL_set_sample_pan and the four stream equivalents, plus AIL_open_stream_by_sample. The MSS_auto_cleanup hook was dropped as well, because its atexit handler would have called AIL_shutdown after the module was already freed. All 92 remaining exports were verified to resolve against the retail mss32.dll. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> * refactor(miles): Fix the written primitive types in mss.h and all its call sites; no ABI changes (TheSuperHackers#3272) * perf(productionupdate): Simplify and correct implementations of cancel functions in ProductionUpdate (TheSuperHackers#3270) * chore(gamememory): Compile out the memory link tester in Release (TheSuperHackers#3266) * perf(gamememory): Early exit delete and free functions on null pointer (TheSuperHackers#3266) * perf(gamememory): Inline preMainInitMemoryManager (TheSuperHackers#3266) * perf(gamememory): Add overloads for the deletes with size_t argument (TheSuperHackers#3266) * chore(gamememory): Remove superfluous extern keywords from operator overloads (TheSuperHackers#3266) * perf(gamememory): Remove unnecessary calls to preMainInitMemoryManager from delete and free functions and make freeBytes noexcept to get rid of EH frame (TheSuperHackers#3266) * fix(gamefont): Ceil font glyph buffer size to the actual glyph size to prevent a buffer write overflow (TheSuperHackers#3268) * refactor(particlesys): Parse IsGroundAligned as an enum instead of a boolean (TheSuperHackers#3265) * ci(release): Stop requesting permissions from the reusable workflow (TheSuperHackers#3276) * refactor(basetype): Add utility functions to Region and Coord types (TheSuperHackers#3271) New functions are: intersectWith, uniteWith for IRegion3D, IRegion2D, Region3D, Region2D updateMin, updateMax for ICoord3D, ICoord2D, Coord3D, Coord2D asICoord2D, asCoord2D for ICoord3D, Coord3D * build(cmake): Add retail compatibility option in CMake config (TheSuperHackers#2379) RTS_BUILD_OPTION_RETAIL_COMPATIBLE_GAME=DEFAULT/ON/OFF * bugfix(meshmatdesc): Fix mesh material color processing (TheSuperHackers#3246) * ci: Restore CI workflow permission compatibility (TheSuperHackers#3286) * chore: Remove trailing commas in braced initializers that break clang-format's compact layout (TheSuperHackers#3274) Scoped to comment-free array/struct literals (BorderColors, TeamGeneric, BezierSegment, GameMemoryInitPools, BFISH, Properties, Scripts) where clang-format explodes each element onto its own line without this. * chore(license): Add SPDX-License-Identifier to LICENSE.md (TheSuperHackers#3290) Helps github detect the license version * bugfix(pathfinder): Restore Generals retail compatibility after crash fix changes to Pathfinder::findAttackPath (TheSuperHackers#3289) * feat(recorder): Play a replay file from the command line (TheSuperHackers#3227) Use -loadreplay <file> as a command line argument to load the replay with full game context * fix(audio): Copy SoundSceneObjClass state safely (TheSuperHackers#3247) * fix(hash): Fix initialization of HashTableIteratorClass and make it work with an empty HashTableClass (TheSuperHackers#3284) * chore(pathfinder): Remove superfluous CPOP_STARTS_FROM_PREV_SEG macro (TheSuperHackers#3295) * fix(milesaudiomanager): Prevent heap-buffer-overflow read in MilesAudioManager::selectProvider() (TheSuperHackers#3281) * bugfix(filesystem): Preserve write paths with missing directories (TheSuperHackers#3104) * perf(pathfinder): Optimize appending node to end of the path (TheSuperHackers#3198) PathNode::appendToList() walks the entire list from the head to find the tail on every call, making repeated appendNode() calls O(n^2) in path length. Path already tracks m_pathTail, so append directly onto it in O(1) instead. Removed PathNode::appendToList() as it is not used anywhere else. * perf(pathfinder): Take parents cell's position outside of for-loop for optimization (TheSuperHackers#3198) The parent cell's world position fromPos never changes across the neighbour loop, so compute it once instead. * perf(pathfinder): Remove redundant isCrusher recomputation for optimization (TheSuperHackers#3198) * ci(windows): make bink and miles runtime stubs optional in build artifacts * fix(platform): preserve POSIX startup working directory and set Flatpak asset paths * docs(worklog): document CI fixes and verification for upstream sync PR 304 --------- Co-authored-by: ArcticDolphin <5984296+tintinhamans@users.noreply.github.com> Co-authored-by: Jacob Lane Ledbetter <23038070+CryoTheRenegade@users.noreply.github.com> Co-authored-by: xezon <4720891+xezon@users.noreply.github.com> Co-authored-by: Stubbjax <11547761+Stubbjax@users.noreply.github.com> Co-authored-by: Claude Opus 5 <noreply@anthropic.com> Co-authored-by: stm <14291421+stephanmeesters@users.noreply.github.com> Co-authored-by: mirelle7 <115191165+mirelle7@users.noreply.github.com> Co-authored-by: Caball009 <82909616+Caball009@users.noreply.github.com> Co-authored-by: Bobby Battista <bobtista@gmail.com> Co-authored-by: SkyAero <21192585+Skyaero42@users.noreply.github.com>
The glyph buffer was a fixed
uint16[32768].Store_GDI_Charwriteswidth * heightpixels with no bound check, so a big glyph writes past the end.A
pointSize > 100cap used to hide this, but it was removed in #3051 so 4K UI scalingcan use bigger fonts. A scaled font can now hit the overflow.
FontCharsBufferallocates its pixels and grows to fit the glyph.getFontclamps requests to 512 instead of rejecting them, so an oversizedrequest still returns a usable font rather than
nullptr(callers likeW3DDisplayString::setFontignore null and would show no text).A font size of about 460 is the most any real screen needs (a 48pt heading blown up 9.6x on an 8K display) so I think 512 is a safe value for now.