[DRAFT] try enabling regions on Mac again #115251
Conversation
src/coreclr/gc/gcpriv.h
Outdated
| @@ -144,7 +144,7 @@ inline void FATAL_GC_ERROR() | |||
| // For SunOS or illumos this is temporary, until we can add MAP_PRIVATE | |||
| // to the mmap() calls in unix/gcenv.unix.cpp More details here: | |||
| // https://github.com/dotnet/runtime/issues/104211 | |||
| #if defined (HOST_64BIT) && !defined (BUILD_AS_STANDALONE) && !defined(__APPLE__) && !defined(__sun) | |||
| #if defined (HOST_64BIT) && !defined (BUILD_AS_STANDALONE) && !defined(__sun) | |||
There was a problem hiding this comment.
The preprocessor condition no longer excludes APPLE, but the comment above still mentions APPLE. Please update the comment to accurately reflect that regions are now enabled on Mac.
Copilot uses AI. Check for mistakes.
|
Tagging subscribers to this area: @dotnet/gc |
|
It is ok to try and the CI may even pass if the run does not happen to generate crash dump along the way. We need a positive confirmation that enabling regions GC won't kill post-mortem diagnostic on mac. Have we done anything to resolve the problem with prohibitively large crash dumps on mac? |
|
the last time we looked we werent able to repro the large dumps locally. Believe @janvorli had tried it too, given lower priority we never investigated it completely. Hoping we get to root cause this time around. |
|
I have validated that with large memory reservation size with regions the dump generated on MacOS is about 100mb larger (so it's only about 2% more) -- not order of magnitude larger we thought it would be. Need to figure out failures in iOS/tvOS NativeAOT failures before we can enable. |
No description provided.