[Native] Add printf-style native logging

[simonrozsival]

Summary

Introduce printf-style native logging primitives and migrate the initial NativeAOT formatted logging call sites to them.

This is the first implementation step from #12139 toward removing the Android NativeAOT dependency on libc++.

Changes

• add log_writev(), log_writef(), log_debugf(), log_infof(), log_warnf(), and log_errorf() alongside the existing std::format APIs;
• implement the same printf helpers in the CLR and MonoVM shared logging backends so shared runtime code can use one logging path;
• add Helpers::abort_applicationf() for formatted fatal messages without truncating tombstone text;
• include <cstdio> explicitly for vasprintf();
• omit an unused log_fatalf() wrapper; formatted fatal termination uses abort_applicationf() instead;
• replace the NativeAOT GC-user-peer initialization error's std::format;
• replace the NativeAOT JNI on-load debug message's {} formatting;
• avoid std::format inside Helpers::abort_application().

The existing std::format APIs remain available so unrelated call sites can be migrated incrementally. The current stacked migrations are #12148 (shared runtime utilities), #12150 (JNI reference logging), #12153 (timing logging), and #12155 (configuration diagnostics).

Runtime behavior

• NativeAOT and CoreCLR use the CLR shared logging implementation.
• MonoVM uses its own shared logging implementation with the same printf helper surface.
• category filtering and Android log priorities remain unchanged.
• null format pointers are logged as <null> rather than passed to __android_log_vprint().

Validation

• git diff --check;
• Release build of src/native/native-nativeaot.csproj;
• Release build of src/native/native-clr.csproj;
• NDK Clang C++23 syntax compilation of the CLR helper changes across NativeAOT/CoreCLR configurations;
• NDK Clang C++23 syntax compilation of the MonoVM printf implementation in 8 Android configurations;
• focused review of va_list, category filtering, format checking, source-location behavior, and Android log-priority mapping;
• latest head: 09f0690da.

Part of #12139.

[Copilot]

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.

[simonrozsival]

CI failure analysis

The public dotnet-android build #1513800 completed with 40 of 42 jobs passing. The two failed jobs are unrelated to the logging changes in this PR:

• MSBuild+Emulator 9 failed only DotNetInstallAndRunMinorAPILevels(True, "net10.0-android36.1", NativeAOT). The linker found duplicate __unw_* symbols in .NET 10.0.9's libRuntime.WorkstationGC.a and NDK 29.0.14206865's libunwind.a. The same test fails with the same collision in the independent [NativeAOT] Reduce owning C++ standard library state #12142 build.

• MSBuild+Emulator 7 had 16 auto-retry failures. Sampled CoreCLR, NativeAOT, fast-deployment, and dotnet run logs converge on an Android emulator installation failure:

  java.lang.NullPointerException: Attempt to invoke virtual method
  'boolean android.app.usage.StorageStatsManager.isQuotaSupported(java.lang.String)'
  on a null object reference
  

  Bundletool and adb install then fail while creating the package-install session. This is an emulator/system-service failure rather than a product assertion.

The NativeAOT/libunwind collision is additional real-world evidence for #12139.

[simonrozsival]

/review

[github-actions]

✅ Android PR Reviewer completed successfully!

[github-actions]

🤖 Code Review — ⚠️ Minor suggestions (no blockers)

Reviewed the printf-style native logging primitives and the NativeAOT call-site migration. This is clean, well-scoped work.

What I verified

• printf format attributes are correct. Helpers::abort_applicationf uses format(printf, 3, 4); since it's a static member (no implicit this), the 3/4 indices are right. The free functions log_writef (3,4) and log_debugf/log_infof/... (2,3) are also indexed correctly.
• Category-guard semantics are consistent. The new (log_categories & category) == 0 early-return matches the modern CLR DO_LOG_FMT macro (!= 0 to log), so behavior is unchanged for the migrated log_debugf/log_infof paths. log_warnf/errorf/fatalf are unguarded, matching the existing log_warn/log_error/log_fatal.
• Null-format safety. log_writev and abort_applicationf both guard nullptr formats with "<null>"/safe_format, and abort_applicationf correctly falls back to safe_format when vasprintf returns < 0.
• abort_application no longer routes fatal text through std::format — good, this preserves full tombstone text and drops the libc++ dependency as intended.
• Leak of the vasprintf buffer in abort_applicationf is acceptable: it's a [[noreturn]] abort path, so the process dies immediately.
• Host-native regression test is a nice addition — it exercises %s/%p/%d forwarding, priority mapping, and disabled-category suppression against Android log stubs.

Suggestions (both inline, non-blocking)

• 💡 Add an explicit #include <cstdio> in helpers.cc for vasprintf rather than relying on a transitive include.
• 💡 log_warnf/log_errorf/log_fatalf are currently unused — confirm the follow-up migration PRs consume them so they don't rot.

CI

GitHub commit status is currently pending (checks not yet reported) at 4a27f68. Not treating this as a failure, but the PR isn't mergeable until the Azure DevOps pipeline is green.

Nice, focused first step toward dropping the NativeAOT libc++ dependency. 👍

Generated by Android PR Reviewer for #12140 · 104.3 AIC · ⌖ 13.1 AIC · ⊞ 6.8K
_{Comment /review to run again}

[simonrozsival]

/review

[github-actions]

✅ Android PR Reviewer completed successfully!

[github-actions]

✅ LGTM (with minor suggestions)

Reviewed the printf-style native logging primitives and the initial NativeAOT call-site migrations. This is a clean, well-scoped change and I verified the details that usually go wrong here:

• printf format attributes are correct — format(printf, 3, 4) for log_writef/abort_applicationf and format(printf, 2, 3) for log_{debug,info,warn,error}f all index the format string and first vararg correctly (source-location/level params counted).
• Category filtering is consistent — log_debugf/log_infof gate on log_categories, while log_warnf/log_errorf/log_writef do not, exactly mirroring the existing log_debug/log_info vs log_warn/log_error macro behavior.
• Null-format safety — log_writev and abort_applicationf substitute "<null>" before reaching __android_log_vprint/vasprintf.
• abort_application no longer routes the message through std::format (log_write(..., Fatal, message)), which also removes a latent %/{} interpretation hazard on the fatal path.
• optional_string() returns const char*, so the new %s usage in host.cc is safe (not a string_view).
• CI — all completed dotnet-android legs are green (build, MSBuild, emulator, package tests); overall status shows pending only from the fresh merge-commit re-trigger.

Suggestions (non-blocking)

• 💡 Document the intentional vasprintf buffer leak on the [[noreturn]] abort path (inline).
• 💡 Consider hoisting the duplicated log_*f bodies shared between the CLR and Mono backends into common code to prevent drift as later stacked PRs migrate more call sites (inline).

Verdict: ⚠️ nothing blocking — two optional 💡 suggestions. Nice, careful incremental step toward dropping the libc++ dependency.

Generated by Android PR Reviewer for #12140 · 94.2 AIC · ⌖ 13 AIC · ⊞ 6.8K
_{Comment /review to run again}