[Diagnostics] Add InProc CrashReport Logging by mdh1418 · Pull Request #128105 · dotnet/runtime

mdh1418 · 2026-05-12T23:23:26Z

This adds a compact text view of the in-proc crash report for mobile, while preserving the existing JSON crash report file output when DOTNET_DbgMiniDumpName is configured.

The compact output is intentionally tombstone-style: it prints process/signal metadata, per-thread frame stacks, a module table, and a footer. It omits data that platform crash reports already provide well, such as registers, memory-near dumps, and memory maps.

On Android, the compact report is emitted to logcat under a dedicated DOTNET_CRASH tag for easy filtering. On iOS, the same compact report is emitted through the existing stderr/minipal log path.

Details

Adds signal-safe formatting helpers shared by the JSON writer and compact log writer.
Adds a compact crash log writer: Android uses DOTNET_CRASH, while non-Android targets write newline-delimited compact report lines through the existing error log path.
Emits compact per-thread managed frame stacks alongside the JSON file writer.
Deduplicates managed modules into a fixed-size module table and references frames by module index.
Adds DOTNET_CrashReportFrameLimitPerThread; default is 32 frames per thread, and 0 disables the compact-log cap.
Keeps the JSON crash report file as the authoritative detailed artifact; the frame cap only limits compact log output.
Enables the in-proc crash reporter for Android and Apple app-model targets (iOS, tvOS, and MacCatalyst), including Apple ucontext_t register extraction.

Android Emulator Validation

Validated on Android x64 emulator with the same crash scenarios as #126916, comparing the console output with the json file.

Unhandled managed exception: compact output and JSON emitted; top frames matched.
Managed FailFast: compact output and JSON emitted; top frames matched.
P/Invoke abort: compact output and JSON emitted; top frames matched.
SIGSEGV: compact output and JSON emitted; top frames matched.
Interleaved managed/native frames via pthread_once: compact output and JSON emitted; top frames matched.

iOS simulator validation

Validated compact console crash-report emission on iossimulator-arm64

All scenarios emitted both:

the on-disk .crashreport.json
the compact .NET Crash Report v1.0.0 console report in the app log

nullref, sigsegv, abort,failfast, unhandled, interleaved, stackoverflow

For every scenario, the console report included the compact header, signal, per-thread blocks, managed frames, and a modules: table. Console thread counts matched the corresponding JSON crash report.

… namespace Moves the async-signal-safe integer formatting helpers and buffer-size constants out of SignalSafeJsonWriter into a shared SignalSafeFormat namespace. The helpers are JSON-agnostic and are needed by both the JSON writer and later compact console output without introducing a sibling dependency. This is intended as a behavior-preserving refactor: JSON writer call sites continue to use the same bounded fixed-buffer formatting logic, just through the shared helper namespace. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Adds SignalSafeConsoleWriter, a bounded line-oriented sink that writes DOTNET_CRASH entries through __android_log_write on Android and newline-terminated stderr lines elsewhere. CreateReport now emits a compact tombstone-style header/footer alongside the existing JSON report path. The JSON header/footer emission is split into helpers, and DbgMiniDumpName becomes optional: when no JSON path is configured, the JSON writer uses a no-op sink while the compact log still runs. PROCGetSignalNameAscii exposes the existing signal-name table in the narrow form used by the compact log. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Adds shared frame-sink plumbing so each walked frame can feed both the JSON writer and the compact console writer. The compact log now emits per-thread headers, managed exception info, managed frame lines with IL offset/token, native frame lines with module offsets, and a marker when no managed frames were reported. Both normal thread enumeration and the synthesized crash-thread fallback use the same console block helpers, keeping per-thread compact-log structure in one place. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

…dule-index references Adds ModuleTable, a 64-entry fixed-capacity table keyed by MVID for one crash report. Compact-log frames can refer to modules by short [N] indices, and the footer emits a modules block that maps each index back to the module filename and MVID. If a managed frame's module cannot be stored because the table is full or the GUID is missing, the frame renders the module name inline as (in <name>) instead of using a lossy placeholder. JSON output is unchanged; module indices are only a compact-log representation detail. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Adds DOTNET_CrashReportFrameLimitPerThread, parsed as base 10 with default 32, to cap the number of frames written per thread to the compact log. Setting the value to 0 disables the limit. Frames past the cap are still emitted to the JSON report. The compact log skips only the console frame line, tracks how many frames were omitted for the current thread, and emits an "... +N more frames" summary in the thread footer. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

dotnet-policy-service · 2026-05-12T23:25:07Z

Tagging subscribers to this area: @steveisok, @tommcdon, @dotnet/dotnet-diag
See info in area-owners.md if you want to be subscribed.

Copilot

Pull request overview

This PR extends CoreCLR’s in-process crash reporting path to emit a compact, signal-safe crash log for mobile scenarios while preserving JSON file output when DOTNET_DbgMiniDumpName is configured.

Changes:

Adds shared signal-safe formatting helpers and a line-oriented compact console/logcat writer.
Refactors in-proc crash report emission into JSON header/footer and compact log sections.
Adds frame limiting and module-table support for compact per-thread stack output.

Reviewed changes

Copilot reviewed 13 out of 13 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
`src/coreclr/vm/crashreportstackwalker.cpp`	Allows crash reporting without a dump path and wires frame-limit configuration.
`src/coreclr/pal/src/thread/process.cpp`	Adds ASCII POSIX signal-name helper.
`src/coreclr/pal/src/include/pal/process.h`	Declares the signal-name helper for crash reporting.
`src/coreclr/inc/clrconfigvalues.h`	Adds `CrashReportFrameLimitPerThread` configuration.
`src/coreclr/debug/crashreport/signalsafejsonwriter.h`	Removes formatter declarations now moved to shared helper.
`src/coreclr/debug/crashreport/signalsafejsonwriter.cpp`	Uses shared signal-safe formatting helpers.
`src/coreclr/debug/crashreport/signalsafeformat.h`	Adds shared bounded integer formatting declarations.
`src/coreclr/debug/crashreport/signalsafeformat.cpp`	Implements shared bounded integer formatting.
`src/coreclr/debug/crashreport/signalsafeconsolewriter.h`	Adds compact line-oriented crash log writer interface.
`src/coreclr/debug/crashreport/signalsafeconsolewriter.cpp`	Implements Android logcat / stderr compact crash log output.
`src/coreclr/debug/crashreport/inproccrashreporter.h`	Adds compact output helpers and frame-limit state.
`src/coreclr/debug/crashreport/inproccrashreporter.cpp`	Emits compact crash logs alongside JSON reporting.
`src/coreclr/debug/crashreport/CMakeLists.txt`	Includes new crash-report helper source files.

+        ? *sinks->currentThreadFrameCount
+        : 0;
+
+    int moduleIndex = s_moduleTable.GetOrAddIndex(moduleName, moduleGuid);


+    {
+        m_buffer[m_pos++] = '\n';
+        m_buffer[m_pos] = '\0';
+    }


lateralusX · 2026-05-13T07:46:53Z

+        char guid[MINIPAL_GUID_BUFFER_LEN];
+    };
+
+    Entry m_entries[MAX_MODULES_IN_TABLE];


This will consume a lot of static memory just to hold the strings. This table only lives while we dump the report, maybe we could just store the Module pointer and then do the name+guid lookups when formatting the table, that should dramatically reduce the needed size and you could probably event increase the preallocated table to support it since you only store a pointer per module in this table.

lateralusX · 2026-05-13T07:50:08Z

+    s_consoleWriter.AppendStr("signal ");
+    s_consoleWriter.AppendSignedDecimal(signal);
+    s_consoleWriter.AppendStr(" (");
+    s_consoleWriter.AppendStr(PROCGetSignalNameAscii(signal));


Only called here, so maybe we could just have it as a static function here instead of spreading it out through the codebase? If we ever have the need to share this in the future, we could move it later.

lateralusX · 2026-05-13T07:53:11Z

+// Single-instance because CreateReport is one-shot per process (guarded by
+// the ``s_generating`` InterlockedCompareExchange in CreateReport).
+
+static constexpr size_t MAX_MODULES_IN_TABLE = 64;


Just storing Module * in the table would probably increase this.

lateralusX · 2026-05-13T08:05:36Z

-    m_jsonWriter.WriteDecimalAsString("pid", static_cast<uint64_t>(GetCurrentProcessId()));
+    EmitJsonHeader();

    m_jsonWriter.OpenArray("threads");


The open/close array calls feels a little misplaced here, all other serialization happens in separate methods or in callbacks. Maybe this is part of the EnumerateThreads/EmitSynthesizedCrashThread implementations instead?

lateralusX · 2026-05-13T08:09:10Z

    }

-    m_jsonWriter.WriteDecimalAsString("pid", static_cast<uint64_t>(GetCurrentProcessId()));
+    EmitJsonHeader();


Maybe this could be named OpenJsonReport, then we have OpenConsoleReport or StartJsonReport/StartConsoleReport, then we could have Close or Finish, less details, clear intent. We should also group them together in code, maybe put them into a static function, Open|StartReport.

lateralusX · 2026-05-13T08:09:44Z

-    m_jsonWriter.WriteString("SystemManufacturer", "apple");
-#endif
-    m_jsonWriter.CloseObject(); // parameters
+    EmitJsonFooter(signal);


Same comment as above.

lateralusX · 2026-05-13T08:10:33Z

+    EmitConsoleModulesAndFooter();

-    if (fd != -1)
+    if (jsonEnabled)


Guess this could go with the close/finish json report function instead of here.

lateralusX · 2026-05-13T08:28:56Z

+        consoleWriter->AppendStr(" (token=0x");
+        consoleWriter->AppendHex(static_cast<uint64_t>(token));
+        consoleWriter->AppendChar(')');
+    }


Should we have an option where methodName, className is nullptr but there is module + token + ilOffset, then we could output that data since it can be used to get back to the managed code. If there is no token + ilOffset, we could then fallback to native IP + offset.

lateralusX · 2026-05-13T08:32:28Z

+    }
+    else if (sinks->currentThreadDroppedCount != nullptr)
+    {
+        ++*sinks->currentThreadDroppedCount;


Post increment reads nicer.

lateralusX · 2026-05-13T08:32:34Z

+
+    if (sinks->currentThreadFrameCount != nullptr)
+    {
+        ++*sinks->currentThreadFrameCount;


lateralusX · 2026-05-13T08:36:46Z

+}
+
+void
+CrashReportHelpers::FrameSinkCallback(


WriteFrame?

lateralusX · 2026-05-13T08:42:54Z

    const char* moduleGuid)
 {
-    CrashReportHelpers::JsonFrameCallback(ip, stackPointer, methodName, className, moduleName, nativeOffset, token, ilOffset, moduleTimestamp, moduleSize, moduleGuid, m_writer);
+    CrashReportHelpers::FrameSinks sinks =


Do we need yet another context object for this? If so maybe it should be named similar to other using Context, like FrameContext or similar?

lateralusX · 2026-05-13T08:43:47Z

 }

+void
+ThreadEnumerationContext::FinishCurrentThreadCompactBlock()


This should be named better.

lateralusX · 2026-05-13T08:45:24Z

+}
+
+void
+CrashReportHelpers::WriteThreadBlockHeaderToConsole(


Could we come up with a better naming pair than Header/Closer?

lateralusX · 2026-05-13T08:48:48Z


+    void FinishCurrentThreadCompactBlock();
+
    SignalSafeJsonWriter* m_writer;


Should this be more specific now when there are two writers? Like m_jsonWriter?

lateralusX · 2026-05-13T08:55:32Z

+// --- InProcCrashReporter: console header and footer ------------------------
+
+void
+InProcCrashReporter::EmitConsoleHeader(int signal)


WriteConsoleHeader ?

mdh1418 and others added 5 commits May 12, 2026 01:53

mdh1418 requested review from Copilot and lateralusX May 12, 2026 23:23

Copilot started reviewing on behalf of mdh1418 May 12, 2026 23:23 View session

mdh1418 added area-Diagnostics-coreclr os-android os-ios Apple iOS labels May 12, 2026

dotnet-policy-service Bot assigned mdh1418 May 12, 2026

Copilot AI reviewed May 12, 2026

View reviewed changes

build-analysis Bot mentioned this pull request May 13, 2026

Unable to pull image from mcr.microsoft.com #117164

Open

lateralusX reviewed May 13, 2026

View reviewed changes


		void FinishCurrentThreadCompactBlock();

		SignalSafeJsonWriter* m_writer;

Conversation

mdh1418 commented May 12, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Details

Android Emulator Validation

iOS simulator validation

Uh oh!

dotnet-policy-service Bot commented May 12, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

lateralusX May 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

mdh1418 commented May 12, 2026 •

edited

Loading

lateralusX May 13, 2026 •

edited

Loading