CNA-BUG-271: known_bugs.md still lists the FNA3D dangling-device use-after-free as OPEN and a lost repeated SpriteBatch Begin/End as a live symptom, although the source has fixed both

CNA snapshot 009d40f5  ·  Known Issues › Current bugs  ·  source links pinned to 009d40f5

CNA's own known-bug list keeps two entries whose fixes are in the source and pinned by registered tests: FNA3D resource renderers now hold a shared device state that the renderer clears (Fna3d_Device_Lifetime), and repeated SpriteBatch Begin()/End() cycles on VULKAN are pinned by Vulkan_SpriteBatch_MultiBeginEnd.

Identifier

CNA-BUG-271

Category

Bug

Subsystem

Documentation & release tooling

Status

Open

Verified against

CNA 009d40f5 (009d40f5dd085c4e674d3479675fac84b12b3e0a)

Severity

Low (a triage suggestion, not a project priority)

Evidence basis

Source-verified: read at TARGET, not executed

Tests touching this area

Yes: see Current tests

Affected contract

known_bugs.md, the entries for FNA3D resource renderers that outlive their device and for several SpriteBatch Begin/End cycles in one frame

Expected behaviour

A known-bug list states what still fails: an entry whose fix has landed and is pinned by a registered test is removed or marked fixed, so that a reader does not take it for open work.

Actual behaviour at TARGET

known_bugs.md says of the first entry “Status: OPEN”: an FNA3D resource renderer keeps a raw FNA3D_Device* guarded only by a null test, so a resource that outlives its GraphicsDevice frees through a dangling pointer. At the snapshot the destructors in Fna3dResources.cpp read the device through LiveDevice(deviceState_), a shared Fna3dDeviceState whose pointer the renderer's own destructor sets to null, and skip FNA3D_AddDisposeTexture and the render-buffer release when it is null; the test Fna3d_Device_Lifetime (fna3d_device_lifetime_test.cpp) destroys the renderer while every wrapper is alive.

The second entry, “Multiple SpriteBatch Begin/End in one frame discards all but the last” (backend VULKAN, dated 2026-06-27), carries no status line and reads as a live symptom, while the VULKAN examples CMake file registers Vulkan_SpriteBatch_MultiBeginEnd under the comment that several Begin()/End() cycles per frame must all render.

Source locations

• known_bugs.md — the two entries
• modules/renderers/fna3d/src/Fna3dResources.cpp — LiveDevice and the resource-renderer destructors
• modules/renderers/fna3d/examples/fna3d_device_lifetime_test.cpp — destroys the renderer while every wrapper is alive
• modules/renderers/vulkan/examples/CMakeLists.txt — Vulkan_SpriteBatch_MultiBeginEnd registration (Task 664)

Evidence

Read at 009d40f5; nothing was built or executed, so the two fixes are read from the source and the registered tests, not observed. The commit that introduced the shared device state and its test (d5c3460db, 2026-08-13) is an ancestor of the snapshot, while the FNA3D entry says it was found on 2026-08-14; the entry was not updated. The site's Contribute page had turned the first entry into an open task and was corrected during this audit.

Focused reproduction

No focused reproduction is known. Nothing has been invented here; the evidence above is what exists.

Current tests

The two fixes are pinned by Fna3d_Device_Lifetime and Vulkan_SpriteBatch_MultiBeginEnd; nothing checks that known_bugs.md agrees with them.

Regression test

Remove the two entries, naming the fixing tasks, or mark them fixed; a document check could compare each entry's status line with the plan that closes it.

Blast radius

Readers who take known_bugs.md for CNA's list of open defects; the site's own Contribute page had done so.

Workaround

Treat the source and the registered tests as authoritative.