This seems to work just fine.

Hi Xiang,

Hi, this patch causes an issue with the CMake Xcode build configuration, if I try to use xcode as the generator with CMake, using the build command:

Hi Qiongsi,

Hi Qiongsi

Hi, this patch causes an issue with the CMake Xcode build configuration, if I try to use xcode as the generator with CMake, using the build command:

Made changes as pointed out by Adrian

Add newlines to the ends of DWARFRecordSectionSplitter.h/.cpp

I noticed that there is no test for this change. Can you please add one?

LGTM! I will take care of adding a new section to the reflection-dump.test for dsymutil. You will need to cherry-pick this change into apple/llvm-project stable/20211026 branch for it to propagate to swift.

Ah I see, weird because I just locally reverted this change and those errors vanished though there are other errors that popped up in its place, which I suspected were due to some other change because greendragon has been red for a while. Thanks @abrachet

This change breaks the lldb greendragon bot

Add newline to Swift.def at the end of file

Done :)

Closed with 957b24ca9f1e

Hi I think I know what to do and am trying to fix it myself. Thanks for trying!

Hi it seems like this patch caused an error in the lldb build machines:

Hi this commit broke the lldb greendragon bots.

Hi it seems like this change might have been responsible for the lldb incremental bot to have broken

I forgot to add the extra checks for the test to the diff

466329d047fc53ea4a1ba3843c27a5cfdacd17a5

Changed name to binaryformat from swiftbinaryformat

In DwarfLinkerForBinary.cpp Line 475, we need to make sure that copySwiftReflectionMetadata is not called when the -no-output option is passed, this is because the DwarfStreamer object is not created, and copySwiftReflectionMetadata calls through to a member method of the DwarfStreamer Object. I also moved reflection-metadata.test to the X86 folder because some non X86 tests failed due to target triple issues

Added early return and change else { if .. to and else if

Added the comments as Adrian suggested, and removed the check Adrian suggested as well

Changed the logic for binaryHasSwiftReflectionSections to remove the NoError bool flag, made LinkOpts const, passed DebugMapObject and DwarfStreamer Objects as pointers to copySwiftReflectionMetadata, removed auto where it isn't obvious, removed braces from single line if, used llvm style if statement for dyn_cast

Removed the argument Swift5ReflectionSectionName from initMachOMCObjectFileInfo function and get the name from the Ctx object present within the MCObjectFileInfo object instead

This change broke the test in ModuleDebugInfo.cpp, I have fixed that issue, and this also will act as a test for this change

Put StringRef Swift5ReflectionSegmentName in MCContext, made the SwiftReflectionSections Array the size of enum::Last instead of harcoding.

Removed StringRef pointer and replaced it with a default empty StringRef argument. Changed Unkown to -1 and removed distinct swift5 sections in MOFI to one std::array. Changed names from Swift5ReflectionSectionType to Swift5ReflectionSectionKind.

Fixed comments

Added a new file called Swift.def, it contains macros that define the various swift reflection section names and contents so we can use them throughout the code, and code changes that come with using the def file. Also added a change to have dwarflinker pass in the segment "__DWARF" to the streamer and the MC to create the reflection sections in the dwarf segment

Reduced the testcase yaml files, made the Swift5ReflectionSection creation confirgurable, also moved the copying of the reflection section metadata to its own function called copySwiftReflectionMetadata

Added the changes as mentioned by Adrian, based on my comments in response to him

@probinson Hi Paul, is it possible for you to take another look? I have updated the test files to be more strict with the prologue_ends.

Added a regex check for ensuring no prologue_end is there on no-non-zero-debug-loc-prologue.ll

Updated the test files make sure there are check's for where prologue_end's should not be present. Added a better comment to explain the change made to findPrologueEndLoc as suggested

Fixed the clang-tidy warnings

As suggested by Adrian and Paul, I have reduced the original testcase to be smaller by using a small C sample and compiling only with -gline-tables-only. I have also added another test case where if the only .loc is a line 0 loc, we want to have a prologue_end there, and only move the prologue_end to a non-zero .loc if it exists.