Use multibyte character path for bundle_probe

[am11]

Fixes #44098.

[am11]

cc @janvorli

Waiting on a clean CI run. Checkout leg is failing: #44472.

[ghost]

Tagging subscribers to this area: @vitek-karas, @agocke
See info in area-owners.md if you want to be subscribed.

Details

---

Issue meta data

Issue content:	Fixes #44098.
Issue author:	am11
Assignees:	-
Milestone:	-

[vitek-karas]

LGTM - let's wait for @janvorli to take a second look

[am11]

Internal to CoreCLR; SString has a concept of internal representation, so if the string was originally UTF8, it will be preserved and GetUTF8 will return it as is (without conversion). I was trying to keep the diff smaller, as it is a deep rabbit hole to update all instances; but there seems to be few places where we have both wide and narrow variants of the same string available at the call sites of bundle probes. If it deemed profitable, we can update those in a separate PR by using narrow variant when constructing SString for probe call.

[janvorli]

LGTM, thank you!

[janvorli]

The win x86 leg was failing due to something that looks like a bug in xunit (xunit code modifying a collection while it was being iterated). I am re-running this leg.

[elinor-fung]

We should also make sure the single-file doc in dotnet/designs reflects this change.

Edit: Actually, apparently the design doc does have char, but the implementation didn't match.

[am11]

When runtime repo itself is cloned in a path containing multibyte character, I found two issues.

Steps to repro (using three-bytes characters):

$ git clone https://github.com/dotnet/runtime runtime-ⒾⓇⓇⒺⒼⓊⓁⒶⓇ␣ⓉⒺⓍⓉ
$ cd runtime-ⒾⓇⓇⒺⒼⓊⓁⒶⓇ␣ⓉⒺⓍⓉ
$ ./build.sh -c Release
$ ./build.sh Host.Tests -c Release -test

1. First the build errors, that were trivial to fix:

   --- a/src/coreclr/src/scripts/genDummyProvider.py
   +++ b/src/coreclr/src/scripts/genDummyProvider.py
   @@ -111,7 +111,7 @@ def generateDummyFiles(etwmanifest, out_dirname, extern, dryRun):
           providerName = trimProvName(providerNode.getAttribute('name'))
           providerName_File = escapeProvFilename(providerName)
   
   -        dummyevntprov = os.path.join(out_dirname, dummyevntprovPre + providerName_File + ".cpp")
   +        dummyevntprov = os.path.join(out_dirname.decode('utf-8'), dummyevntprovPre + providerName_File + ".cpp").encode('utf-8')
   
           if dryRun:
               print(dummyevntprov)
   diff --git a/src/coreclr/src/scripts/genEventPipe.py b/src/coreclr/src/scripts/genEventPipe.py
   index d8c6cdca24f..54aa704148d 100644
   --- a/src/coreclr/src/scripts/genEventPipe.py
   +++ b/src/coreclr/src/scripts/genEventPipe.py
   @@ -384,7 +384,7 @@ def generateEventPipeImplFiles(
           providerName_File = providerPrettyName.replace('-', '')
           providerName_File = providerName_File.lower()
           providerPrettyName = providerPrettyName.replace('-', '_')
   -        eventpipefile = os.path.join(eventpipe_directory, providerName_File + ".cpp")
   +        eventpipefile = os.path.join(eventpipe_directory.decode('utf-8'), providerName_File + ".cpp").encode('utf-8')
           if dryRun:
               print(eventpipefile)
           else:

2. ToLower() in C# is using ICU transofrmation, so HostFxrPath.ToLower() in this case becomes: /users/am11/projects/regular_text-ⓘⓡⓡⓔⓖⓤⓛⓐⓡ␣ⓣⓔⓧⓣ/runtime_pr/artifacts/tests/release/ha/nativehosting/0/valid/dotnet/host/fxr/2.3.0/libhostfxr.dylib, while C++ transformation is ASCII only, so it becomes /users/am11/projects/regular_text-ⒾⓇⓇⒺⒼⓊⓁⒶⓇ␣ⓉⒺⓍⓉ/runtime_pr/artifacts/tests/release/ha/nativehosting/0/valid/dotnet/host/fxr/2.3.0/libhostfxr.dylib and fails the Assert.Contains instances in Nethost.cs. The fix for this was to match C++, ASCII only tolower transformation: http://sprunge.us/UPR16s.

With these two patches, full runtime repo build succeeded and all installer tests passed on macOS. It seems like there are other ToLower() cases in MSBuild scripts (which also uses .NET's ICU-backed transformation), but those are unrelated to PR changes and can be repro'd with the current master branch (and probably ToLower() transformation for hostfxr). Perhaps we should try to get rid of ToLower transformation entirely and rely on underlying filesystem's case-sensitivity choices; i.e. don't care about the casing explicitly in code like the other components of runtime repo.

[vitek-karas]

Thanks a lot @am11 !