Linux: Update to DXC's null-invariant BSTR #11
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
MarijnS95
added a commit
that referenced
this pull request
Dec 29, 2020
DXC (all the Windows string types using WCHAR under the hood, really) require wide strings to be terminated with a NUL character. If not, they'd use a BSTR that has a sneaky length prefix (#11), otherwise the length or ending of the string is unknown. This has not caused any trouble before because into_vec performs the exact same conversion as into_vec_with_nul, *except* that it pops off the last character before returning the Vec. Only the size of it is reduced, the underlying memory is not reallocated and the final NUL byte is left in place. However, in the event that a vector is shuffled around or modified we cannot permit ourselves to loose that trailing NUL.
MarijnS95
added a commit
that referenced
this pull request
Dec 29, 2020
DXC (all the Windows string types using WCHAR under the hood, really) require wide strings to be terminated with a NUL character. If not, they'd use a BSTR that has a sneaky length prefix (#11), otherwise the length or ending of the string is unknown. This has not caused any trouble before because into_vec performs the exact same conversion as into_vec_with_nul, *except* that it pops off the last character before returning the Vec. Only the size of it is reduced, the underlying memory is not reallocated and the final NUL byte is left in place. However, in the event that a vector is shuffled around or modified we cannot permit ourselves to loose that trailing NUL.
MarijnS95
force-pushed
the
dxc-prefix-sized-BSTR
branch
from
b3d4ae5 to
e6a72df
Compare
Jasper-Bekkers
pushed a commit
that referenced
this pull request
Dec 29, 2020
DXC (all the Windows string types using WCHAR under the hood, really) require wide strings to be terminated with a NUL character. If not, they'd use a BSTR that has a sneaky length prefix (#11), otherwise the length or ending of the string is unknown. This has not caused any trouble before because into_vec performs the exact same conversion as into_vec_with_nul, *except* that it pops off the last character before returning the Vec. Only the size of it is reduced, the underlying memory is not reallocated and the final NUL byte is left in place. However, in the event that a vector is shuffled around or modified we cannot permit ourselves to loose that trailing NUL.
|
It's probably time to merge this in now. Aforementioned patch has been merged some time ago and is available under the The suggestion to provide these helpers from DXC however has gone unnoticed... |
MarijnS95
force-pushed
the
dxc-prefix-sized-BSTR
branch
from
e6a72df to
5deb75c
Compare
MarijnS95
force-pushed
the
dxc-prefix-sized-BSTR
branch
from
5deb75c to
81d91d7
Compare
We [recently] fixed memory leaks when calling into the intellisense part of DXC and made the decision shortly after merging to [implement null-invariant `BSTR`s properly] in DXC's shimmed type for non-Windows, which was promptly accepted. Currently our implementation of `SysFreeString` together with that PR segfaults because it frees the pointer at offset `+4` of the allocation. That is corrected together with `SysStringLen` now reading the size of the string from this prefix instead of looking for a (wrong!) null termination character. At least our Windows and Linux implementation `utils.rs` is the same again. [recently]: #10 [implement null-invariant `BSTR`s properly]: microsoft/DirectXShaderCompiler#3250
MarijnS95
force-pushed
the
dxc-prefix-sized-BSTR
branch
from
81d91d7 to
8edb988
Compare
MarijnS95
added a commit
that referenced
this pull request
Jan 12, 2022
…of a panic
Not all file loading should be fallible; DXC usually invokes the loader
with multiple paths depending on relative includes, and happily accepts
and expects us to return `-2_147_024_894 // ERROR_FILE_NOT_FOUND /
0x80070002` for some. As such the precedent should be to not panic but
let DXC decide if file loading failures should be critical or simply
result in a different path being tested.
In my case though the error seems to originate from DXIL compilation
being largely untested when ran on Linux systems through the WinAadapter
codepath. In this particular stacktrace below DXC seems to "stat" (get
file information) for the current working directory, and somehow ends up
trying to open the directory as file through our include handler:
#11 dxcutil::DxcArgsFileSystemImpl::TryFindOrOpen (this=<optimized out>, this@entry=0x5570a3b0ae10, lpFileName=lpFileName@entry=0x5570a3b5d0d0 L".../hassle-rs", index=<optimized out>)
at .../DirectXShaderCompiler/tools/clang/tools/dxcompiler/dxcfilesystem.cpp:327
#12 0x00007f6e64a41a8d in dxcutil::DxcArgsFileSystemImpl::GetFileAttributesW (this=0x5570a3b0ae10, lpFileName=0x5570a3b5d0d0 L".../hassle-rs")
at .../DirectXShaderCompiler/tools/clang/tools/dxcompiler/dxcfilesystem.cpp:570
#13 0x00007f6e64a42048 in dxcutil::DxcArgsFileSystemImpl::Stat (this=0x5570a3b0ae10, lpFileName=<optimized out>, Status=0x7fff9faec1e8)
at .../DirectXShaderCompiler/tools/clang/tools/dxcompiler/dxcfilesystem.cpp:807
#14 0x00007f6e64debc91 in llvm::sys::fs::status (Path=..., Result=...) at .../DirectXShaderCompiler/lib/Support/Unix/Path.inc:385
#15 0x00007f6e64dec708 in llvm::sys::fs::current_path (result=...) at .../DirectXShaderCompiler/lib/Support/Unix/Path.inc:195
#16 0x00007f6e64e64535 in clang::CodeGen::CGDebugInfo::getCurrentDirname (this=this@entry=0x5570a3b598f0)
at .../DirectXShaderCompiler/tools/clang/lib/CodeGen/CGDebugInfo.cpp:301
#17 0x00007f6e64e6230e in clang::CodeGen::CGDebugInfo::CreateCompileUnit (this=this@entry=0x5570a3b598f0)
at .../DirectXShaderCompiler/tools/clang/lib/CodeGen/CGDebugInfo.cpp:369
This stacktrace originates from the following panic:
thread 'main' panicked at 'called `Result::unwrap()` on an `Err` value: Os { code: 21, kind: IsADirectory, message: "Is a directory" }', src/utils.rs:51:48
MarijnS95
added a commit
that referenced
this pull request
Jan 25, 2022
…of a panic (#27) Not all file loading should be fallible; DXC usually invokes the loader with multiple paths depending on relative includes, and happily accepts and expects us to return `-2_147_024_894 // ERROR_FILE_NOT_FOUND / 0x80070002` for some. As such the precedent should be to not panic but let DXC decide if file loading failures should be critical or simply result in a different path being tested. In my case though the error seems to originate from DXIL compilation being largely untested when ran on Linux systems through the WinAadapter codepath. In this particular stacktrace below DXC seems to "stat" (get file information) for the current working directory, and somehow ends up trying to open the directory as file through our include handler: #11 dxcutil::DxcArgsFileSystemImpl::TryFindOrOpen (this=<optimized out>, this@entry=0x5570a3b0ae10, lpFileName=lpFileName@entry=0x5570a3b5d0d0 L".../hassle-rs", index=<optimized out>) at .../DirectXShaderCompiler/tools/clang/tools/dxcompiler/dxcfilesystem.cpp:327 #12 0x00007f6e64a41a8d in dxcutil::DxcArgsFileSystemImpl::GetFileAttributesW (this=0x5570a3b0ae10, lpFileName=0x5570a3b5d0d0 L".../hassle-rs") at .../DirectXShaderCompiler/tools/clang/tools/dxcompiler/dxcfilesystem.cpp:570 #13 0x00007f6e64a42048 in dxcutil::DxcArgsFileSystemImpl::Stat (this=0x5570a3b0ae10, lpFileName=<optimized out>, Status=0x7fff9faec1e8) at .../DirectXShaderCompiler/tools/clang/tools/dxcompiler/dxcfilesystem.cpp:807 #14 0x00007f6e64debc91 in llvm::sys::fs::status (Path=..., Result=...) at .../DirectXShaderCompiler/lib/Support/Unix/Path.inc:385 #15 0x00007f6e64dec708 in llvm::sys::fs::current_path (result=...) at .../DirectXShaderCompiler/lib/Support/Unix/Path.inc:195 #16 0x00007f6e64e64535 in clang::CodeGen::CGDebugInfo::getCurrentDirname (this=this@entry=0x5570a3b598f0) at .../DirectXShaderCompiler/tools/clang/lib/CodeGen/CGDebugInfo.cpp:301 #17 0x00007f6e64e6230e in clang::CodeGen::CGDebugInfo::CreateCompileUnit (this=this@entry=0x5570a3b598f0) at .../DirectXShaderCompiler/tools/clang/lib/CodeGen/CGDebugInfo.cpp:369 This stacktrace originates from the following panic: thread 'main' panicked at 'called `Result::unwrap()` on an `Err` value: Os { code: 21, kind: IsADirectory, message: "Is a directory" }', src/utils.rs:51:48
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
I keep forgetting to submit this change, which would cause segfaults in the long run when newer
libdxcompiler.sobuilds are used (in conjunction with intellisense only, if that makes it any better...).We recently fixed memory leaks when calling into the intellisense part of DXC and made the decision shortly after merging to implement null-invariant
BSTRs properly in DXC's shimmed type for non-Windows, which was promptly accepted.Currently our implementation of
SysFreeStringtogether with that PR segfaults because it frees the pointer at offset+4of the allocation. That is corrected together withSysStringLennow reading the size of the string from this prefix instead of looking for a (wrong!) null character. At least our Windows and Linux implementationutils.rsis the same again.Meanwhile there's a suggestion open at DXC to export these functions (
CoTaskMemFree,SysFreeStringetc) because the implementation is uniquely defined by DXC. We might want to hold off merging this PR (hence draft) until having confirmation whether that's acceptable. Otherwise we should coordinate this release with a version bump in DXC?