Twelve fixes from lateralusX's review pass on 2026-04-29 plus a build-fix for the cross-components configuration: - Replace GetFilename's hand-rolled scan-for-last-separator loop with strrchr. Adds a nullptr guard for path. dotnet#126916 (comment) - Rename the 'copied' local in CopyString and GetVersionString to 'toCopy'. At the binding site the value is the count about to be copied, not bytes already copied. dotnet#126916 (comment) dotnet#126916 (comment) - Null-check 'writer' / 'ctx' in JsonFrameCallback, ThreadCallback, and FrameCallback. The void* -> typed-pointer cast at a callback boundary is a contract the type system can't enforce; defending it at the callback entry mirrors the existing pattern in ChunkCallback. dotnet#126916 (comment) dotnet#126916 (comment) (FrameCallback fix found via the R3160838988 audit; not separately flagged) - Add ThreadEnumerationContext::EnumerateThreads to wrap the enumerate callback dispatch + EndEnumeration into a single call site. Forgetting EndEnumeration would silently leave the JSON malformed; wrapping removes the obligation from the caller. dotnet#126916 (comment) - Move OnThread, OnFrame (ThreadEnumerationContext) and HandleChunk (CrashReportOutputContext) to the private: section. They're only called from their sibling static thunks; the thunks remain public for the function-pointer ABI. dotnet#126916 (comment) dotnet#126916 (comment) dotnet#126916 (comment) - Replace FormatHexValue's, FormatUnsignedDecimal's, and FormatSignedDecimal's upfront 'bufferSize < MAX_*_BUFFER_SIZE' check with a tighter check tied to the actual write footprint after the digit-extraction loop. Removes the dependency on the constants being correctly aligned with the loops' writes. dotnet#126916 (comment) dotnet#126916 (comment) - Null-check ctx in FrameCallbackAdapter. Same trust-boundary principle as the JsonFrameCallback / ThreadCallback / FrameCallback fixes above, applied to the JIT-stackwalker callback. dotnet#126916 (comment) - Document the GCX_COOP safety reasoning in CrashReportGetExceptionForThread. The transition is safe because the crashing thread holds the thread-store lock taken by CrashReportSuspendThreads, and RareDisablePreemptiveGC's fast-path skips the suspend-blocking check in that condition. dotnet#126916 (comment) - Make the exception-info capture explicit at the signal-handler entry point. InProcCrashReportSignalDispatcher now captures hasException/exTypeBuf/exHresult on the crashing thread and passes them to CreateReport as parameters, instead of CreateReport calling m_getExceptionCallback as its first step. CreateReport's signature grows three parameters; the dispatcher gains a friend declaration so it can read m_getExceptionCallback without exposing it via a public accessor. The behavior is unchanged (the capture was already happening early in CreateReport, on the crashing thread); the structure makes the capture-then-write contract visible at the architecture level rather than buried in CreateReport's body. dotnet#126916 (comment) - Gate the inproccrashreport library on `NOT CLR_CROSS_COMPONENTS_BUILD` in src/coreclr/debug/CMakeLists.txt and src/coreclr/dlls/mscoree/coreclr/CMakeLists.txt. The cross-components build (used to produce cross-arch JITs) does not build PAL, so attaching `inproccrashreport` to `coreclrpal` / `coreclr_static` fails configure with "target_sources called with non-existing target". Surfaced when running the full `build.cmd clr` flow on Android x64. The pre-allocated-scratch-buffer suggestion (dotnet#126916 (comment)) is captured as an explicit "Async-signal-safety hardening" follow-up in the PR description and is not addressed here per the reviewer's "not for this PR" note.

Commits

Commits on Apr 14, 2026

Commits on Apr 15, 2026

Commits on Apr 16, 2026

Commits on Apr 17, 2026

Commits on Apr 20, 2026

Commits on Apr 21, 2026

Commits on Apr 22, 2026

Commits on Apr 23, 2026

Commits on Apr 28, 2026

Commits on Apr 29, 2026

Commits on Apr 30, 2026

Commits on May 4, 2026

Commits on May 5, 2026