[PATCH] D102339: [clang] On Windows, ignore case and separators when discarding duplicate dependency file paths.

[Sylvain Audi via Phabricator via cfe-commits]

saudi created this revision.
saudi added reviewers: rnk, thakis, dexonsmith, Bigcheese.
saudi added a project: clang.
saudi requested review of this revision.
Herald added a subscriber: cfe-commits.

We noticed that the dependency collectors of `clang` may report duplicate paths under Windows, when the differences are only in casing or separators (`/` vs `\`)
This problem affects the correctness of the information output by `clang-scan-deps`, which in turn, affects our integration with the `Fastbuild` cache.

For separators, it can happen as clang internally computes the full path of a header from an `#include` directive using native separator.

Example with argument `-I C:\includes`:
 `#include <subdir/header.h>`  -> internally represented in the `FileManager` as `C:\includes\subdir/header.h`
 `#include "header.h"` from another header in `subdir` folder  -> `C:\includes\subdir\header.h`

The same applies with casing discrepancies that can be found sometimes in headers from external SDKs.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D102339

Files:
  clang/lib/Frontend/DependencyFile.cpp
  clang/test/Frontend/dependency-gen-windows-duplicates.c


Index: clang/test/Frontend/dependency-gen-windows-duplicates.c
===================================================================
--- /dev/null
+++ clang/test/Frontend/dependency-gen-windows-duplicates.c
@@ -0,0 +1,27 @@
+// REQUIRES: system-windows
+
+// RUN: rm -rf %t.dir
+// RUN: mkdir -p %t.dir/subdir
+// RUN: echo > %t.dir/subdir/x.h
+// RUN: cp %s %t.dir/test.c
+// RUN: cd %t.dir
+
+// RUN: %clang -MD -MF - %t.dir/test.c -fsyntax-only -I %t.dir/subdir | FileCheck --ignore-case %s
+// CHECK: test.o:
+// CHECK-NEXT: \test.c
+// CHECK-NEXT: \subdir\x.h
+// File x.h must appear only once (case insensitive check).
+// CHECK-NOT: x.h
+
+// Include x.h several times, with different casing and separators.
+// Since all paths are passed to clang as absolute, all dependencies are absolute paths.
+// We expect the output dependencies to contain only one line for file x.h
+
+// Test case sensitivity.
+#include "subdir/x.h"
+#include "SubDir/X.h"
+
+// Test separator sensitivity:
+// clang will internally concatenates x.h paths using the Windows native separator.
+#include <x.h>
+
Index: clang/lib/Frontend/DependencyFile.cpp
===================================================================
--- clang/lib/Frontend/DependencyFile.cpp
+++ clang/lib/Frontend/DependencyFile.cpp
@@ -141,7 +141,18 @@
 }
 
 bool DependencyCollector::addDependency(StringRef Filename) {
-  if (Seen.insert(Filename).second) {
+  StringRef SearchPath;
+#ifdef _WIN32
+  // Make the search insensitive to case and separators.
+  llvm::SmallString<256> TmpPath = Filename;
+  std::replace(TmpPath.begin(), TmpPath.end(), '/', '\\');
+  std::transform(TmpPath.begin(), TmpPath.end(), TmpPath.begin(), ::tolower);
+  SearchPath = TmpPath.str();
+#else
+  SearchPath = Filename;
+#endif
+
+  if (Seen.insert(SearchPath).second) {
     Dependencies.push_back(std::string(Filename));
     return true;
   }


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D102339.344824.patch
Type: text/x-patch
Size: 1890 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20210512/0bfd78b7/attachment.bin>