================
@@ -4753,15 +4755,12 @@ void ASTWriter::AddString(StringRef Str, RecordDataImpl
&Record) {
}
bool ASTWriter::PreparePathForOutput(SmallVectorImpl<char> &Path) {
- assert(Context && "should have context when outputting path");
-
// Leave special file names as they are.
StringRef PathStr(Path.data(), Path.size());
if (PathStr == "<built-in>" || PathStr == "<command line>")
return false;
- bool Changed =
- cleanPathForOutput(Context->getSourceManager().getFileManager(), Path);
+ bool Changed = cleanPathForOutput(PP->getFileManager(), Path);
----------------
jansvoboda11 wrote:
Yeah, I think `PP` has the same issues. Both `PP` and `Context` are null
outside of `WriteAST()`, which is a foot gun.
However, `PP` is guaranteed to be non-null during the `WriteAST()` call, while
`Context` may be null during that call (after
https://github.com/llvm/llvm-project/pull/115239). So `Context` will have more
complicated rules and they will only be exercised during scanning. I think that
would make it easy to make changes that will work for regular module compiles,
but break the scanner. That's why I see more value in doing this transformation
to `Context`.
To be clear I would like to do the same thing with `PP`, it's just that its
simpler rules don't make it super important in my opinion. And any uses of null
`PP` would be caught outside of the scanner, in normal module compiles.
https://github.com/llvm/llvm-project/pull/115235
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
