[clang] [llvm] Make sanitizer special case list slash-agnostic (PR #149886)
Devon Loehr via llvm-commits
llvm-commits at lists.llvm.org
Tue Oct 7 09:10:15 PDT 2025
https://github.com/DKLoehr updated https://github.com/llvm/llvm-project/pull/149886
>From c613019535f3251d3201b0f1408988366f5732c2 Mon Sep 17 00:00:00 2001
From: Devon Loehr <dloehr at google.com>
Date: Mon, 21 Jul 2025 19:07:23 +0000
Subject: [PATCH 1/6] Make special case matcher slash-agnostic
---
clang/docs/SanitizerSpecialCaseList.rst | 1 +
clang/unittests/Basic/DiagnosticTest.cpp | 23 +++++++++++++++++++++++
llvm/docs/ReleaseNotes.md | 4 ++++
llvm/include/llvm/Support/GlobPattern.h | 1 +
llvm/lib/Support/GlobPattern.cpp | 4 ++++
5 files changed, 33 insertions(+)
diff --git a/clang/docs/SanitizerSpecialCaseList.rst b/clang/docs/SanitizerSpecialCaseList.rst
index 307c001664fba..f2a04dc9adcf1 100644
--- a/clang/docs/SanitizerSpecialCaseList.rst
+++ b/clang/docs/SanitizerSpecialCaseList.rst
@@ -174,6 +174,7 @@ tool-specific docs.
# Lines starting with # are ignored.
# Turn off checks for the source file
# Entries without sections are placed into [*] and apply to all sanitizers
+ # "/" matches both windows and unix path separators ("/" and "\")
src:path/to/source/file.c
src:*/source/file.c
# Turn off checks for this main file, including files included by it.
diff --git a/clang/unittests/Basic/DiagnosticTest.cpp b/clang/unittests/Basic/DiagnosticTest.cpp
index 4b3af00c3b0ce..a6557b1e35c4b 100644
--- a/clang/unittests/Basic/DiagnosticTest.cpp
+++ b/clang/unittests/Basic/DiagnosticTest.cpp
@@ -360,4 +360,27 @@ TEST_F(SuppressionMappingTest, ParsingRespectsOtherWarningOpts) {
clang::ProcessWarningOptions(Diags, Diags.getDiagnosticOptions(), *FS);
EXPECT_THAT(diags(), IsEmpty());
}
+
+TEST_F(SuppressionMappingTest, ForwardSlashMatchesBothDirections) {
+ llvm::StringLiteral SuppressionMappingFile = R"(
+ [unused]
+ src:*clang/*
+ src:*clang/lib/Sema/*=emit
+ src:*clang/lib\\Sema/foo*)";
+ Diags.getDiagnosticOptions().DiagnosticSuppressionMappingsFile = "foo.txt";
+ FS->addFile("foo.txt", /*ModificationTime=*/{},
+ llvm::MemoryBuffer::getMemBuffer(SuppressionMappingFile));
+ clang::ProcessWarningOptions(Diags, Diags.getDiagnosticOptions(), *FS);
+ EXPECT_THAT(diags(), IsEmpty());
+
+ EXPECT_TRUE(Diags.isSuppressedViaMapping(
+ diag::warn_unused_function, locForFile(R"(clang/lib/Basic/foo.h)")));
+ EXPECT_FALSE(Diags.isSuppressedViaMapping(
+ diag::warn_unused_function, locForFile(R"(clang/lib/Sema\bar.h)")));
+ EXPECT_TRUE(Diags.isSuppressedViaMapping(
+ diag::warn_unused_function, locForFile(R"(clang\lib\Sema/foo.h)")));
+ // The third pattern requires a literal backslash before Sema
+ EXPECT_FALSE(Diags.isSuppressedViaMapping(
+ diag::warn_unused_function, locForFile(R"(clang/lib/Sema/foo.h)")));
+}
} // namespace
diff --git a/llvm/docs/ReleaseNotes.md b/llvm/docs/ReleaseNotes.md
index 85c16b9c33f10..6ff8d18b07e84 100644
--- a/llvm/docs/ReleaseNotes.md
+++ b/llvm/docs/ReleaseNotes.md
@@ -174,6 +174,10 @@ Changes to BOLT
Changes to Sanitizers
---------------------
+* The [sanitizer special case list format](https://clang.llvm.org/docs/SanitizerSpecialCaseList.html#format)
+ now treats forward slashes as either a forward or a backslash, to handle
+ paths with mixed unix and window styles.
+
Other Changes
-------------
diff --git a/llvm/include/llvm/Support/GlobPattern.h b/llvm/include/llvm/Support/GlobPattern.h
index 62ed4a0f23fd9..af92c63331282 100644
--- a/llvm/include/llvm/Support/GlobPattern.h
+++ b/llvm/include/llvm/Support/GlobPattern.h
@@ -35,6 +35,7 @@ namespace llvm {
/// expansions are not supported. If \p MaxSubPatterns is empty then
/// brace expansions are not supported and characters `{,}` are treated as
/// literals.
+/// * `/` matches both unix and windows path separators: `/` and `\`.
/// * `\` escapes the next character so it is treated as a literal.
///
/// Some known edge cases are:
diff --git a/llvm/lib/Support/GlobPattern.cpp b/llvm/lib/Support/GlobPattern.cpp
index 7004adf461a0c..26b3724863ee8 100644
--- a/llvm/lib/Support/GlobPattern.cpp
+++ b/llvm/lib/Support/GlobPattern.cpp
@@ -231,6 +231,10 @@ bool GlobPattern::SubGlobPattern::match(StringRef Str) const {
++S;
continue;
}
+ } else if (*P == '/' && (*S == '/' || *S == '\\')) {
+ ++P;
+ ++S;
+ continue;
} else if (*P == *S || *P == '?') {
++P;
++S;
>From a81b5509f3d9633eee2276c2242c595378d1cfdc Mon Sep 17 00:00:00 2001
From: Devon Loehr <dloehr at google.com>
Date: Tue, 22 Jul 2025 14:57:34 +0000
Subject: [PATCH 2/6] Enable only for special case list
---
llvm/include/llvm/Support/GlobPattern.h | 9 +++++++--
llvm/lib/Support/GlobPattern.cpp | 12 +++++++-----
llvm/lib/Support/SpecialCaseList.cpp | 3 ++-
3 files changed, 16 insertions(+), 8 deletions(-)
diff --git a/llvm/include/llvm/Support/GlobPattern.h b/llvm/include/llvm/Support/GlobPattern.h
index af92c63331282..2729ba9a56649 100644
--- a/llvm/include/llvm/Support/GlobPattern.h
+++ b/llvm/include/llvm/Support/GlobPattern.h
@@ -56,8 +56,10 @@ class GlobPattern {
/// \param MaxSubPatterns if provided limit the number of allowed subpatterns
/// created from expanding braces otherwise disable
/// brace expansion
+ /// \param IsSlashAgnostic whether to treat '/' as matching '\\' as well
LLVM_ABI static Expected<GlobPattern>
- create(StringRef Pat, std::optional<size_t> MaxSubPatterns = {});
+ create(StringRef Pat, std::optional<size_t> MaxSubPatterns = {},
+ bool IsSlashAgnostic = false);
/// \returns \p true if \p S matches this glob pattern
LLVM_ABI bool match(StringRef S) const;
@@ -76,7 +78,9 @@ class GlobPattern {
struct SubGlobPattern {
/// \param Pat the pattern to match against
- LLVM_ABI static Expected<SubGlobPattern> create(StringRef Pat);
+ /// \param SlashAgnostic whether to treat '/' as matching '\\' as well
+ LLVM_ABI static Expected<SubGlobPattern> create(StringRef Pat,
+ bool SlashAgnostic);
/// \returns \p true if \p S matches this glob pattern
LLVM_ABI bool match(StringRef S) const;
StringRef getPat() const { return StringRef(Pat.data(), Pat.size()); }
@@ -88,6 +92,7 @@ class GlobPattern {
};
SmallVector<Bracket, 0> Brackets;
SmallVector<char, 0> Pat;
+ bool IsSlashAgnostic;
};
SmallVector<SubGlobPattern, 1> SubGlobs;
};
diff --git a/llvm/lib/Support/GlobPattern.cpp b/llvm/lib/Support/GlobPattern.cpp
index 26b3724863ee8..4aa30a81c3fbf 100644
--- a/llvm/lib/Support/GlobPattern.cpp
+++ b/llvm/lib/Support/GlobPattern.cpp
@@ -132,8 +132,9 @@ parseBraceExpansions(StringRef S, std::optional<size_t> MaxSubPatterns) {
return std::move(SubPatterns);
}
-Expected<GlobPattern>
-GlobPattern::create(StringRef S, std::optional<size_t> MaxSubPatterns) {
+Expected<GlobPattern> GlobPattern::create(StringRef S,
+ std::optional<size_t> MaxSubPatterns,
+ bool IsSlashAgnostic) {
GlobPattern Pat;
// Store the prefix that does not contain any metacharacter.
@@ -147,7 +148,7 @@ GlobPattern::create(StringRef S, std::optional<size_t> MaxSubPatterns) {
if (auto Err = parseBraceExpansions(S, MaxSubPatterns).moveInto(SubPats))
return std::move(Err);
for (StringRef SubPat : SubPats) {
- auto SubGlobOrErr = SubGlobPattern::create(SubPat);
+ auto SubGlobOrErr = SubGlobPattern::create(SubPat, IsSlashAgnostic);
if (!SubGlobOrErr)
return SubGlobOrErr.takeError();
Pat.SubGlobs.push_back(*SubGlobOrErr);
@@ -157,8 +158,9 @@ GlobPattern::create(StringRef S, std::optional<size_t> MaxSubPatterns) {
}
Expected<GlobPattern::SubGlobPattern>
-GlobPattern::SubGlobPattern::create(StringRef S) {
+GlobPattern::SubGlobPattern::create(StringRef S, bool SlashAgnostic) {
SubGlobPattern Pat;
+ Pat.IsSlashAgnostic = SlashAgnostic;
// Parse brackets.
Pat.Pat.assign(S.begin(), S.end());
@@ -231,7 +233,7 @@ bool GlobPattern::SubGlobPattern::match(StringRef Str) const {
++S;
continue;
}
- } else if (*P == '/' && (*S == '/' || *S == '\\')) {
+ } else if (IsSlashAgnostic && *P == '/' && (*S == '/' || *S == '\\')) {
++P;
++S;
continue;
diff --git a/llvm/lib/Support/SpecialCaseList.cpp b/llvm/lib/Support/SpecialCaseList.cpp
index 8d4e043bc1c9f..c597f03188507 100644
--- a/llvm/lib/Support/SpecialCaseList.cpp
+++ b/llvm/lib/Support/SpecialCaseList.cpp
@@ -59,7 +59,8 @@ Error SpecialCaseList::Matcher::insert(StringRef Pattern, unsigned LineNumber,
Glob->LineNo = LineNumber;
// We must be sure to use the string in `Glob` rather than the provided
// reference which could be destroyed before match() is called
- if (auto Err = GlobPattern::create(Glob->Name, /*MaxSubPatterns=*/1024)
+ if (auto Err = GlobPattern::create(Glob->Name, /*MaxSubPatterns=*/1024,
+ /*IsSlashAgnostic=*/true)
.moveInto(Glob->Pattern))
return Err;
Globs.push_back(std::move(Glob));
>From 7bfc6ad6afbcafe9f230d3854b28b4406f83537b Mon Sep 17 00:00:00 2001
From: Devon Loehr <dloehr at google.com>
Date: Fri, 5 Sep 2025 16:48:17 +0000
Subject: [PATCH 3/6] Enable only on windows
---
clang/docs/SanitizerSpecialCaseList.rst | 2 +-
clang/unittests/Basic/DiagnosticTest.cpp | 3 +++
llvm/docs/ReleaseNotes.md | 4 ++--
llvm/include/llvm/Support/GlobPattern.h | 3 ++-
llvm/lib/Support/SpecialCaseList.cpp | 6 +++++-
5 files changed, 13 insertions(+), 5 deletions(-)
diff --git a/clang/docs/SanitizerSpecialCaseList.rst b/clang/docs/SanitizerSpecialCaseList.rst
index f2a04dc9adcf1..e14b654536b8a 100644
--- a/clang/docs/SanitizerSpecialCaseList.rst
+++ b/clang/docs/SanitizerSpecialCaseList.rst
@@ -174,7 +174,7 @@ tool-specific docs.
# Lines starting with # are ignored.
# Turn off checks for the source file
# Entries without sections are placed into [*] and apply to all sanitizers
- # "/" matches both windows and unix path separators ("/" and "\")
+ # On windows, "/" matches both styles of path separator ("/" and "\")
src:path/to/source/file.c
src:*/source/file.c
# Turn off checks for this main file, including files included by it.
diff --git a/clang/unittests/Basic/DiagnosticTest.cpp b/clang/unittests/Basic/DiagnosticTest.cpp
index a6557b1e35c4b..7e9653bdd3c7e 100644
--- a/clang/unittests/Basic/DiagnosticTest.cpp
+++ b/clang/unittests/Basic/DiagnosticTest.cpp
@@ -361,6 +361,8 @@ TEST_F(SuppressionMappingTest, ParsingRespectsOtherWarningOpts) {
EXPECT_THAT(diags(), IsEmpty());
}
+#ifdef _WIN32
+// We're only slash-agnostic on windows hosts
TEST_F(SuppressionMappingTest, ForwardSlashMatchesBothDirections) {
llvm::StringLiteral SuppressionMappingFile = R"(
[unused]
@@ -383,4 +385,5 @@ TEST_F(SuppressionMappingTest, ForwardSlashMatchesBothDirections) {
EXPECT_FALSE(Diags.isSuppressedViaMapping(
diag::warn_unused_function, locForFile(R"(clang/lib/Sema/foo.h)")));
}
+#endif
} // namespace
diff --git a/llvm/docs/ReleaseNotes.md b/llvm/docs/ReleaseNotes.md
index 6ff8d18b07e84..ae6222d2fa145 100644
--- a/llvm/docs/ReleaseNotes.md
+++ b/llvm/docs/ReleaseNotes.md
@@ -174,9 +174,9 @@ Changes to BOLT
Changes to Sanitizers
---------------------
-* The [sanitizer special case list format](https://clang.llvm.org/docs/SanitizerSpecialCaseList.html#format)
+* On windows hosts, the [sanitizer special case list format](https://clang.llvm.org/docs/SanitizerSpecialCaseList.html#format)
now treats forward slashes as either a forward or a backslash, to handle
- paths with mixed unix and window styles.
+ paths with mixed unix and windows styles.
Other Changes
-------------
diff --git a/llvm/include/llvm/Support/GlobPattern.h b/llvm/include/llvm/Support/GlobPattern.h
index 2729ba9a56649..4abd6b1874593 100644
--- a/llvm/include/llvm/Support/GlobPattern.h
+++ b/llvm/include/llvm/Support/GlobPattern.h
@@ -35,7 +35,8 @@ namespace llvm {
/// expansions are not supported. If \p MaxSubPatterns is empty then
/// brace expansions are not supported and characters `{,}` are treated as
/// literals.
-/// * `/` matches both unix and windows path separators: `/` and `\`.
+/// * If IsSlashAgnostic is passed, `/` matches both unix and windows path
+/// separators: `/` and `\`.
/// * `\` escapes the next character so it is treated as a literal.
///
/// Some known edge cases are:
diff --git a/llvm/lib/Support/SpecialCaseList.cpp b/llvm/lib/Support/SpecialCaseList.cpp
index c597f03188507..89ec193e1991c 100644
--- a/llvm/lib/Support/SpecialCaseList.cpp
+++ b/llvm/lib/Support/SpecialCaseList.cpp
@@ -18,6 +18,8 @@
#include "llvm/Support/LineIterator.h"
#include "llvm/Support/MemoryBuffer.h"
#include "llvm/Support/VirtualFileSystem.h"
+#include "llvm/TargetParser/Host.h"
+#include "llvm/TargetParser/Triple.h"
#include <stdio.h>
#include <string>
#include <system_error>
@@ -57,10 +59,12 @@ Error SpecialCaseList::Matcher::insert(StringRef Pattern, unsigned LineNumber,
auto Glob = std::make_unique<Matcher::Glob>();
Glob->Name = Pattern.str();
Glob->LineNo = LineNumber;
+ // Backslashes are valid in posix-style filenames.
+ bool IsSlashAgnostic = Triple(sys::getDefaultTargetTriple()).isOSWindows();
// We must be sure to use the string in `Glob` rather than the provided
// reference which could be destroyed before match() is called
if (auto Err = GlobPattern::create(Glob->Name, /*MaxSubPatterns=*/1024,
- /*IsSlashAgnostic=*/true)
+ /*IsSlashAgnostic=*/IsSlashAgnostic)
.moveInto(Glob->Pattern))
return Err;
Globs.push_back(std::move(Glob));
>From a5eaf5880c295a6a50f0be504d3c421641f28932 Mon Sep 17 00:00:00 2001
From: Devon Loehr <dloehr at google.com>
Date: Mon, 8 Sep 2025 15:00:04 +0000
Subject: [PATCH 4/6] Incorporate review feedback
---
clang/docs/SanitizerSpecialCaseList.rst | 2 +-
clang/lib/Basic/Diagnostic.cpp | 5 ++++-
clang/lib/Basic/SanitizerSpecialCaseList.cpp | 2 +-
clang/unittests/Basic/DiagnosticTest.cpp | 12 ++++++-----
llvm/docs/ReleaseNotes.md | 4 ++--
llvm/include/llvm/Support/GlobPattern.h | 19 +++++++----------
llvm/include/llvm/Support/SpecialCaseList.h | 5 +++--
llvm/lib/Support/GlobPattern.cpp | 21 +++++++++----------
llvm/lib/Support/SpecialCaseList.cpp | 22 +++++++++++---------
9 files changed, 48 insertions(+), 44 deletions(-)
diff --git a/clang/docs/SanitizerSpecialCaseList.rst b/clang/docs/SanitizerSpecialCaseList.rst
index e14b654536b8a..752602c1b3093 100644
--- a/clang/docs/SanitizerSpecialCaseList.rst
+++ b/clang/docs/SanitizerSpecialCaseList.rst
@@ -174,7 +174,7 @@ tool-specific docs.
# Lines starting with # are ignored.
# Turn off checks for the source file
# Entries without sections are placed into [*] and apply to all sanitizers
- # On windows, "/" matches both styles of path separator ("/" and "\")
+ # On windows, "/" also matches "\" in filenames
src:path/to/source/file.c
src:*/source/file.c
# Turn off checks for this main file, including files included by it.
diff --git a/clang/lib/Basic/Diagnostic.cpp b/clang/lib/Basic/Diagnostic.cpp
index dc3778bbf339c..9dd133cb4c03e 100644
--- a/clang/lib/Basic/Diagnostic.cpp
+++ b/clang/lib/Basic/Diagnostic.cpp
@@ -622,6 +622,8 @@ bool WarningsSpecialCaseList::isDiagSuppressed(diag::kind DiagId,
bool WarningsSpecialCaseList::globsMatches(
const llvm::StringMap<Matcher> &CategoriesToMatchers,
StringRef FilePath) const {
+ static bool HaveWindowsPathStyle =
+ llvm::sys::path::is_style_windows(llvm::sys::path::Style::native);
StringRef LongestMatch;
bool LongestIsPositive = false;
for (const auto &Entry : CategoriesToMatchers) {
@@ -631,7 +633,8 @@ bool WarningsSpecialCaseList::globsMatches(
for (const auto &Glob : Matcher.Globs) {
if (Glob->Name.size() < LongestMatch.size())
continue;
- if (!Glob->Pattern.match(FilePath))
+ if (!Glob->Pattern.match(FilePath,
+ /*IsSlashAgnostic=*/HaveWindowsPathStyle))
continue;
LongestMatch = Glob->Name;
LongestIsPositive = IsPositive;
diff --git a/clang/lib/Basic/SanitizerSpecialCaseList.cpp b/clang/lib/Basic/SanitizerSpecialCaseList.cpp
index f7bc1d5545d75..4ad35d4d73fdd 100644
--- a/clang/lib/Basic/SanitizerSpecialCaseList.cpp
+++ b/clang/lib/Basic/SanitizerSpecialCaseList.cpp
@@ -42,7 +42,7 @@ void SanitizerSpecialCaseList::createSanitizerSections() {
SanitizerMask Mask;
#define SANITIZER(NAME, ID) \
- if (S.SectionMatcher->match(NAME)) \
+ if (S.SectionMatcher->match(NAME, /*IsFilename=*/false)) \
Mask |= SanitizerKind::ID;
#define SANITIZER_GROUP(NAME, ID, ALIAS) SANITIZER(NAME, ID)
diff --git a/clang/unittests/Basic/DiagnosticTest.cpp b/clang/unittests/Basic/DiagnosticTest.cpp
index 7e9653bdd3c7e..2af86b6a5ef38 100644
--- a/clang/unittests/Basic/DiagnosticTest.cpp
+++ b/clang/unittests/Basic/DiagnosticTest.cpp
@@ -363,12 +363,13 @@ TEST_F(SuppressionMappingTest, ParsingRespectsOtherWarningOpts) {
#ifdef _WIN32
// We're only slash-agnostic on windows hosts
-TEST_F(SuppressionMappingTest, ForwardSlashMatchesBothDirections) {
+TEST_F(SuppressionMappingTest, TreatsFilesAsSlashAgnosticOnWindows) {
llvm::StringLiteral SuppressionMappingFile = R"(
[unused]
src:*clang/*
src:*clang/lib/Sema/*=emit
- src:*clang/lib\\Sema/foo*)";
+ src:*clang/lib\\Sema/foo*
+ fun:suppress/me)";
Diags.getDiagnosticOptions().DiagnosticSuppressionMappingsFile = "foo.txt";
FS->addFile("foo.txt", /*ModificationTime=*/{},
llvm::MemoryBuffer::getMemBuffer(SuppressionMappingFile));
@@ -376,12 +377,13 @@ TEST_F(SuppressionMappingTest, ForwardSlashMatchesBothDirections) {
EXPECT_THAT(diags(), IsEmpty());
EXPECT_TRUE(Diags.isSuppressedViaMapping(
- diag::warn_unused_function, locForFile(R"(clang/lib/Basic/foo.h)")));
+ diag::warn_unused_function, locForFile(R"(clang/lib/Basic/bar.h)")));
EXPECT_FALSE(Diags.isSuppressedViaMapping(
- diag::warn_unused_function, locForFile(R"(clang/lib/Sema\bar.h)")));
+ diag::warn_unused_function, locForFile(R"(clang/lib/Sema\baz.h)")));
+
+ // We require a literal backslash before "Sema"
EXPECT_TRUE(Diags.isSuppressedViaMapping(
diag::warn_unused_function, locForFile(R"(clang\lib\Sema/foo.h)")));
- // The third pattern requires a literal backslash before Sema
EXPECT_FALSE(Diags.isSuppressedViaMapping(
diag::warn_unused_function, locForFile(R"(clang/lib/Sema/foo.h)")));
}
diff --git a/llvm/docs/ReleaseNotes.md b/llvm/docs/ReleaseNotes.md
index ae6222d2fa145..39c2a82239fe0 100644
--- a/llvm/docs/ReleaseNotes.md
+++ b/llvm/docs/ReleaseNotes.md
@@ -175,8 +175,8 @@ Changes to Sanitizers
---------------------
* On windows hosts, the [sanitizer special case list format](https://clang.llvm.org/docs/SanitizerSpecialCaseList.html#format)
- now treats forward slashes as either a forward or a backslash, to handle
- paths with mixed unix and windows styles.
+ now treats forward slashes in filenames as matching either a forward or a
+ backslash, to accommodate paths with mixed unix and windows styles.
Other Changes
-------------
diff --git a/llvm/include/llvm/Support/GlobPattern.h b/llvm/include/llvm/Support/GlobPattern.h
index 4abd6b1874593..5fd1e0764cc7a 100644
--- a/llvm/include/llvm/Support/GlobPattern.h
+++ b/llvm/include/llvm/Support/GlobPattern.h
@@ -35,9 +35,9 @@ namespace llvm {
/// expansions are not supported. If \p MaxSubPatterns is empty then
/// brace expansions are not supported and characters `{,}` are treated as
/// literals.
-/// * If IsSlashAgnostic is passed, `/` matches both unix and windows path
-/// separators: `/` and `\`.
/// * `\` escapes the next character so it is treated as a literal.
+/// * If \p IsSlashAgnostic is passed to the match function, then forward
+/// slashes `/` also match backslashes `\`.
///
/// Some known edge cases are:
/// * The literal `]` is allowed as the first character in a character class,
@@ -57,12 +57,11 @@ class GlobPattern {
/// \param MaxSubPatterns if provided limit the number of allowed subpatterns
/// created from expanding braces otherwise disable
/// brace expansion
- /// \param IsSlashAgnostic whether to treat '/' as matching '\\' as well
LLVM_ABI static Expected<GlobPattern>
- create(StringRef Pat, std::optional<size_t> MaxSubPatterns = {},
- bool IsSlashAgnostic = false);
+ create(StringRef Pat, std::optional<size_t> MaxSubPatterns = {});
+ /// \param IsSlashAgnostic whether to treat '/' as also matching '\'
/// \returns \p true if \p S matches this glob pattern
- LLVM_ABI bool match(StringRef S) const;
+ LLVM_ABI bool match(StringRef S, bool IsSlashAgnostic = false) const;
// Returns true for glob pattern "*". Can be used to avoid expensive
// preparation/acquisition of the input for match().
@@ -79,11 +78,10 @@ class GlobPattern {
struct SubGlobPattern {
/// \param Pat the pattern to match against
- /// \param SlashAgnostic whether to treat '/' as matching '\\' as well
- LLVM_ABI static Expected<SubGlobPattern> create(StringRef Pat,
- bool SlashAgnostic);
+ LLVM_ABI static Expected<SubGlobPattern> create(StringRef Pat);
+ /// \param IsSlashAgnostic whether to treat '/' as also matching '\'
/// \returns \p true if \p S matches this glob pattern
- LLVM_ABI bool match(StringRef S) const;
+ LLVM_ABI bool match(StringRef S, bool IsSlashAgnostic) const;
StringRef getPat() const { return StringRef(Pat.data(), Pat.size()); }
// Brackets with their end position and matched bytes.
@@ -93,7 +91,6 @@ class GlobPattern {
};
SmallVector<Bracket, 0> Brackets;
SmallVector<char, 0> Pat;
- bool IsSlashAgnostic;
};
SmallVector<SubGlobPattern, 1> SubGlobs;
};
diff --git a/llvm/include/llvm/Support/SpecialCaseList.h b/llvm/include/llvm/Support/SpecialCaseList.h
index 22a62eac9e01a..951f27eed8ee8 100644
--- a/llvm/include/llvm/Support/SpecialCaseList.h
+++ b/llvm/include/llvm/Support/SpecialCaseList.h
@@ -124,8 +124,9 @@ class SpecialCaseList {
LLVM_ABI Error insert(StringRef Pattern, unsigned LineNumber,
bool UseRegex);
// Returns the line number in the source file that this query matches to.
- // Returns zero if no match is found.
- LLVM_ABI unsigned match(StringRef Query) const;
+ // On windows, treat '/' as also matching '\' in filenames when using globs.
+ // Returns zero if no match is found
+ LLVM_ABI unsigned match(StringRef Query, bool IsFilename) const;
struct Glob {
std::string Name;
diff --git a/llvm/lib/Support/GlobPattern.cpp b/llvm/lib/Support/GlobPattern.cpp
index 4aa30a81c3fbf..578c0dd0760d2 100644
--- a/llvm/lib/Support/GlobPattern.cpp
+++ b/llvm/lib/Support/GlobPattern.cpp
@@ -132,9 +132,8 @@ parseBraceExpansions(StringRef S, std::optional<size_t> MaxSubPatterns) {
return std::move(SubPatterns);
}
-Expected<GlobPattern> GlobPattern::create(StringRef S,
- std::optional<size_t> MaxSubPatterns,
- bool IsSlashAgnostic) {
+Expected<GlobPattern>
+GlobPattern::create(StringRef S, std::optional<size_t> MaxSubPatterns) {
GlobPattern Pat;
// Store the prefix that does not contain any metacharacter.
@@ -148,7 +147,7 @@ Expected<GlobPattern> GlobPattern::create(StringRef S,
if (auto Err = parseBraceExpansions(S, MaxSubPatterns).moveInto(SubPats))
return std::move(Err);
for (StringRef SubPat : SubPats) {
- auto SubGlobOrErr = SubGlobPattern::create(SubPat, IsSlashAgnostic);
+ auto SubGlobOrErr = SubGlobPattern::create(SubPat);
if (!SubGlobOrErr)
return SubGlobOrErr.takeError();
Pat.SubGlobs.push_back(*SubGlobOrErr);
@@ -158,9 +157,8 @@ Expected<GlobPattern> GlobPattern::create(StringRef S,
}
Expected<GlobPattern::SubGlobPattern>
-GlobPattern::SubGlobPattern::create(StringRef S, bool SlashAgnostic) {
+GlobPattern::SubGlobPattern::create(StringRef S) {
SubGlobPattern Pat;
- Pat.IsSlashAgnostic = SlashAgnostic;
// Parse brackets.
Pat.Pat.assign(S.begin(), S.end());
@@ -192,21 +190,22 @@ GlobPattern::SubGlobPattern::create(StringRef S, bool SlashAgnostic) {
return Pat;
}
-bool GlobPattern::match(StringRef S) const {
+bool GlobPattern::match(StringRef S, bool IsSlashAgnostic) const {
if (!S.consume_front(Prefix))
return false;
if (SubGlobs.empty() && S.empty())
return true;
for (auto &Glob : SubGlobs)
- if (Glob.match(S))
+ if (Glob.match(S, IsSlashAgnostic))
return true;
return false;
}
// Factor the pattern into segments split by '*'. The segment is matched
-// sequentianlly by finding the first occurrence past the end of the previous
+// sequentially by finding the first occurrence past the end of the previous
// match.
-bool GlobPattern::SubGlobPattern::match(StringRef Str) const {
+bool GlobPattern::SubGlobPattern::match(StringRef Str,
+ bool IsSlashAgnostic) const {
const char *P = Pat.data(), *SegmentBegin = nullptr, *S = Str.data(),
*SavedS = S;
const char *const PEnd = P + Pat.size(), *const End = S + Str.size();
@@ -233,7 +232,7 @@ bool GlobPattern::SubGlobPattern::match(StringRef Str) const {
++S;
continue;
}
- } else if (IsSlashAgnostic && *P == '/' && (*S == '/' || *S == '\\')) {
+ } else if (IsSlashAgnostic && *P == '/' && *S == '\\') {
++P;
++S;
continue;
diff --git a/llvm/lib/Support/SpecialCaseList.cpp b/llvm/lib/Support/SpecialCaseList.cpp
index 89ec193e1991c..c65cb977c005c 100644
--- a/llvm/lib/Support/SpecialCaseList.cpp
+++ b/llvm/lib/Support/SpecialCaseList.cpp
@@ -17,9 +17,8 @@
#include "llvm/ADT/STLExtras.h"
#include "llvm/Support/LineIterator.h"
#include "llvm/Support/MemoryBuffer.h"
+#include "llvm/Support/Path.h"
#include "llvm/Support/VirtualFileSystem.h"
-#include "llvm/TargetParser/Host.h"
-#include "llvm/TargetParser/Triple.h"
#include <stdio.h>
#include <string>
#include <system_error>
@@ -59,21 +58,22 @@ Error SpecialCaseList::Matcher::insert(StringRef Pattern, unsigned LineNumber,
auto Glob = std::make_unique<Matcher::Glob>();
Glob->Name = Pattern.str();
Glob->LineNo = LineNumber;
- // Backslashes are valid in posix-style filenames.
- bool IsSlashAgnostic = Triple(sys::getDefaultTargetTriple()).isOSWindows();
// We must be sure to use the string in `Glob` rather than the provided
// reference which could be destroyed before match() is called
- if (auto Err = GlobPattern::create(Glob->Name, /*MaxSubPatterns=*/1024,
- /*IsSlashAgnostic=*/IsSlashAgnostic)
+ if (auto Err = GlobPattern::create(Glob->Name, /*MaxSubPatterns=*/1024)
.moveInto(Glob->Pattern))
return Err;
Globs.push_back(std::move(Glob));
return Error::success();
}
-unsigned SpecialCaseList::Matcher::match(StringRef Query) const {
+unsigned SpecialCaseList::Matcher::match(StringRef Query,
+ bool IsFilename) const {
+ static bool HaveWindowsPathStyle =
+ llvm::sys::path::is_style_windows(llvm::sys::path::Style::native);
for (const auto &Glob : reverse(Globs))
- if (Glob->Pattern.match(Query))
+ if (Glob->Pattern.match(
+ Query, /*IsSlashAgnostic=*/(HaveWindowsPathStyle && IsFilename)))
return Glob->LineNo;
for (const auto &[Regex, LineNumber] : reverse(RegExes))
if (Regex->match(Query))
@@ -223,7 +223,8 @@ std::pair<unsigned, unsigned>
SpecialCaseList::inSectionBlame(StringRef Section, StringRef Prefix,
StringRef Query, StringRef Category) const {
for (const auto &S : reverse(Sections)) {
- if (S.SectionMatcher->match(Section)) {
+ bool IsFilename = Prefix == "src" || Prefix == "mainfile";
+ if (S.SectionMatcher->match(Section, IsFilename)) {
unsigned Blame = inSectionBlame(S.Entries, Prefix, Query, Category);
if (Blame)
return {S.FileIdx, Blame};
@@ -242,7 +243,8 @@ unsigned SpecialCaseList::inSectionBlame(const SectionEntries &Entries,
if (II == I->second.end())
return 0;
- return II->getValue().match(Query);
+ bool IsFilename = Prefix == "src" || Prefix == "mainfile";
+ return II->getValue().match(Query, IsFilename);
}
} // namespace llvm
>From 47236f1866050ac23126a64a3d340b755fa89918 Mon Sep 17 00:00:00 2001
From: Devon Loehr <dloehr at google.com>
Date: Thu, 2 Oct 2025 14:56:49 +0000
Subject: [PATCH 5/6] Add glob pattern unittest
---
llvm/lib/Support/GlobPattern.cpp | 2 +-
llvm/unittests/Support/GlobPatternTest.cpp | 9 +++++++++
2 files changed, 10 insertions(+), 1 deletion(-)
diff --git a/llvm/lib/Support/GlobPattern.cpp b/llvm/lib/Support/GlobPattern.cpp
index 578c0dd0760d2..e19c8c84d19a9 100644
--- a/llvm/lib/Support/GlobPattern.cpp
+++ b/llvm/lib/Support/GlobPattern.cpp
@@ -137,7 +137,7 @@ GlobPattern::create(StringRef S, std::optional<size_t> MaxSubPatterns) {
GlobPattern Pat;
// Store the prefix that does not contain any metacharacter.
- size_t PrefixSize = S.find_first_of("?*[{\\");
+ size_t PrefixSize = S.find_first_of("?*[{\\/");
Pat.Prefix = S.substr(0, PrefixSize);
if (PrefixSize == std::string::npos)
return Pat;
diff --git a/llvm/unittests/Support/GlobPatternTest.cpp b/llvm/unittests/Support/GlobPatternTest.cpp
index e4f1025b00956..ba33e233c70bc 100644
--- a/llvm/unittests/Support/GlobPatternTest.cpp
+++ b/llvm/unittests/Support/GlobPatternTest.cpp
@@ -271,4 +271,13 @@ TEST_F(GlobPatternTest, Pathological) {
EXPECT_FALSE(Pat->match(S));
EXPECT_TRUE(Pat->match(S + 'b'));
}
+
+TEST_F(GlobPatternTest, SlashAgnostic) {
+ auto Pat = GlobPattern::create("clang/*");
+ ASSERT_TRUE((bool)Pat);
+ EXPECT_TRUE(Pat->match("clang/foo"));
+ EXPECT_FALSE(Pat->match(R"(clang\foo)"));
+ EXPECT_TRUE(Pat->match("clang/foo", /*isSlashAgnostic=*/true));
+ EXPECT_TRUE(Pat->match(R"(clang\foo)", /*isSlashAgnostic=*/true));
+}
}
>From d46bce9b1ccf20cab73e0ec9ac44bd98809a6c10 Mon Sep 17 00:00:00 2001
From: Devon Loehr <dloehr at google.com>
Date: Mon, 6 Oct 2025 17:48:19 +0000
Subject: [PATCH 6/6] Canonicalize special case list filenames when loading
---
clang/docs/SanitizerSpecialCaseList.rst | 18 ++++++++++-
clang/lib/Basic/Diagnostic.cpp | 21 ++++++++-----
clang/lib/Basic/SanitizerSpecialCaseList.cpp | 2 +-
clang/unittests/Basic/DiagnosticTest.cpp | 16 +++++++---
llvm/docs/ReleaseNotes.md | 7 +++--
llvm/include/llvm/Support/GlobPattern.h | 8 ++---
llvm/include/llvm/Support/SpecialCaseList.h | 8 ++---
llvm/lib/Support/GlobPattern.cpp | 15 +++------
llvm/lib/Support/SpecialCaseList.cpp | 31 ++++++++++---------
llvm/unittests/Support/GlobPatternTest.cpp | 9 ------
.../unittests/Support/SpecialCaseListTest.cpp | 18 +++++++++++
11 files changed, 92 insertions(+), 61 deletions(-)
diff --git a/clang/docs/SanitizerSpecialCaseList.rst b/clang/docs/SanitizerSpecialCaseList.rst
index 752602c1b3093..f5c45c1b81df0 100644
--- a/clang/docs/SanitizerSpecialCaseList.rst
+++ b/clang/docs/SanitizerSpecialCaseList.rst
@@ -174,7 +174,6 @@ tool-specific docs.
# Lines starting with # are ignored.
# Turn off checks for the source file
# Entries without sections are placed into [*] and apply to all sanitizers
- # On windows, "/" also matches "\" in filenames
src:path/to/source/file.c
src:*/source/file.c
# Turn off checks for this main file, including files included by it.
@@ -197,6 +196,23 @@ tool-specific docs.
[{cfi-vcall,cfi-icall}]
fun:*BadCfiCall
+
+.. note::
+
+ By default, ``src`` and ``mainfile`` are matched against the filename as seen
+ by LLVM. On Windows, this might involve a mix of forward and backslashes as
+ file separators, and writing patterns to match both variants can be
+ inconvenient.
+
+ If the special case list file begins with ``#!canonical-paths``, then paths
+ will be canonicalized before patterns are matched against them. This involves
+ stripping any leading dots and slashes, and (on Windows only) converting all
+ backslashes to forward slashes.
+
+ If the file uses both ``#!special-case-list-v1`` and ``#!canonical-paths``,
+ then they should occupy the first two lines, and ``#!canonical-paths`` must
+ appear on the second line.
+
``mainfile`` is similar to applying ``-fno-sanitize=`` to a set of files but
does not need plumbing into the build system. This works well for internal
linkage functions but has a caveat for C++ vague linkage functions.
diff --git a/clang/lib/Basic/Diagnostic.cpp b/clang/lib/Basic/Diagnostic.cpp
index 9dd133cb4c03e..71762d10aefa6 100644
--- a/clang/lib/Basic/Diagnostic.cpp
+++ b/clang/lib/Basic/Diagnostic.cpp
@@ -612,18 +612,24 @@ bool WarningsSpecialCaseList::isDiagSuppressed(diag::kind DiagId,
SrcEntriesIt->getValue();
// We also use presumed locations here to improve reproducibility for
// preprocessed inputs.
- if (PresumedLoc PLoc = SM.getPresumedLoc(DiagLoc); PLoc.isValid())
- return globsMatches(
- CategoriesToMatchers,
- llvm::sys::path::remove_leading_dotslash(PLoc.getFilename()));
+ if (PresumedLoc PLoc = SM.getPresumedLoc(DiagLoc); PLoc.isValid()) {
+ if (CanonicalizePaths) {
+ return globsMatches(
+ CategoriesToMatchers,
+ llvm::sys::path::convert_to_slash(
+ llvm::sys::path::remove_leading_dotslash(PLoc.getFilename())));
+ } else {
+ return globsMatches(
+ CategoriesToMatchers,
+ llvm::sys::path::remove_leading_dotslash(PLoc.getFilename()));
+ }
+ }
return false;
}
bool WarningsSpecialCaseList::globsMatches(
const llvm::StringMap<Matcher> &CategoriesToMatchers,
StringRef FilePath) const {
- static bool HaveWindowsPathStyle =
- llvm::sys::path::is_style_windows(llvm::sys::path::Style::native);
StringRef LongestMatch;
bool LongestIsPositive = false;
for (const auto &Entry : CategoriesToMatchers) {
@@ -633,8 +639,7 @@ bool WarningsSpecialCaseList::globsMatches(
for (const auto &Glob : Matcher.Globs) {
if (Glob->Name.size() < LongestMatch.size())
continue;
- if (!Glob->Pattern.match(FilePath,
- /*IsSlashAgnostic=*/HaveWindowsPathStyle))
+ if (!Glob->Pattern.match(FilePath))
continue;
LongestMatch = Glob->Name;
LongestIsPositive = IsPositive;
diff --git a/clang/lib/Basic/SanitizerSpecialCaseList.cpp b/clang/lib/Basic/SanitizerSpecialCaseList.cpp
index 4ad35d4d73fdd..f7bc1d5545d75 100644
--- a/clang/lib/Basic/SanitizerSpecialCaseList.cpp
+++ b/clang/lib/Basic/SanitizerSpecialCaseList.cpp
@@ -42,7 +42,7 @@ void SanitizerSpecialCaseList::createSanitizerSections() {
SanitizerMask Mask;
#define SANITIZER(NAME, ID) \
- if (S.SectionMatcher->match(NAME, /*IsFilename=*/false)) \
+ if (S.SectionMatcher->match(NAME)) \
Mask |= SanitizerKind::ID;
#define SANITIZER_GROUP(NAME, ID, ALIAS) SANITIZER(NAME, ID)
diff --git a/clang/unittests/Basic/DiagnosticTest.cpp b/clang/unittests/Basic/DiagnosticTest.cpp
index 2af86b6a5ef38..2e052b9e5eaf3 100644
--- a/clang/unittests/Basic/DiagnosticTest.cpp
+++ b/clang/unittests/Basic/DiagnosticTest.cpp
@@ -362,9 +362,8 @@ TEST_F(SuppressionMappingTest, ParsingRespectsOtherWarningOpts) {
}
#ifdef _WIN32
-// We're only slash-agnostic on windows hosts
-TEST_F(SuppressionMappingTest, TreatsFilesAsSlashAgnosticOnWindows) {
- llvm::StringLiteral SuppressionMappingFile = R"(
+TEST_F(SuppressionMappingTest, CanonicalizesSlashesOnWindows) {
+ llvm::StringLiteral SuppressionMappingFile = R"(#!canonical-paths
[unused]
src:*clang/*
src:*clang/lib/Sema/*=emit
@@ -378,14 +377,21 @@ TEST_F(SuppressionMappingTest, TreatsFilesAsSlashAgnosticOnWindows) {
EXPECT_TRUE(Diags.isSuppressedViaMapping(
diag::warn_unused_function, locForFile(R"(clang/lib/Basic/bar.h)")));
+ EXPECT_TRUE(Diags.isSuppressedViaMapping(
+ diag::warn_unused_function, locForFile(R"(clang/lib/Basic\bar.h)")));
+ EXPECT_TRUE(Diags.isSuppressedViaMapping(
+ diag::warn_unused_function, locForFile(R"(clang\lib/Basic/bar.h)")));
+ EXPECT_FALSE(Diags.isSuppressedViaMapping(
+ diag::warn_unused_function, locForFile(R"(clang/lib/Sema/baz.h)")));
EXPECT_FALSE(Diags.isSuppressedViaMapping(
diag::warn_unused_function, locForFile(R"(clang/lib/Sema\baz.h)")));
- // We require a literal backslash before "Sema"
- EXPECT_TRUE(Diags.isSuppressedViaMapping(
+ // The backslash gets canonicalized so we never match the third pattern
+ EXPECT_FALSE(Diags.isSuppressedViaMapping(
diag::warn_unused_function, locForFile(R"(clang\lib\Sema/foo.h)")));
EXPECT_FALSE(Diags.isSuppressedViaMapping(
diag::warn_unused_function, locForFile(R"(clang/lib/Sema/foo.h)")));
}
#endif
+
} // namespace
diff --git a/llvm/docs/ReleaseNotes.md b/llvm/docs/ReleaseNotes.md
index 39c2a82239fe0..5a7f41ef3f0dd 100644
--- a/llvm/docs/ReleaseNotes.md
+++ b/llvm/docs/ReleaseNotes.md
@@ -174,9 +174,10 @@ Changes to BOLT
Changes to Sanitizers
---------------------
-* On windows hosts, the [sanitizer special case list format](https://clang.llvm.org/docs/SanitizerSpecialCaseList.html#format)
- now treats forward slashes in filenames as matching either a forward or a
- backslash, to accommodate paths with mixed unix and windows styles.
+* (Sanitizer Special Case Lists)[https://clang.llvm.org/docs/SanitizerSpecialCaseList.html]
+ may now be prefixed with ``#!canonical-paths`` to specify that filename patterns
+ should be matched against canonicalized paths, without leading dots or slashes
+ and (on Windows only) without any backslashes.
Other Changes
-------------
diff --git a/llvm/include/llvm/Support/GlobPattern.h b/llvm/include/llvm/Support/GlobPattern.h
index 5fd1e0764cc7a..62ed4a0f23fd9 100644
--- a/llvm/include/llvm/Support/GlobPattern.h
+++ b/llvm/include/llvm/Support/GlobPattern.h
@@ -36,8 +36,6 @@ namespace llvm {
/// brace expansions are not supported and characters `{,}` are treated as
/// literals.
/// * `\` escapes the next character so it is treated as a literal.
-/// * If \p IsSlashAgnostic is passed to the match function, then forward
-/// slashes `/` also match backslashes `\`.
///
/// Some known edge cases are:
/// * The literal `]` is allowed as the first character in a character class,
@@ -59,9 +57,8 @@ class GlobPattern {
/// brace expansion
LLVM_ABI static Expected<GlobPattern>
create(StringRef Pat, std::optional<size_t> MaxSubPatterns = {});
- /// \param IsSlashAgnostic whether to treat '/' as also matching '\'
/// \returns \p true if \p S matches this glob pattern
- LLVM_ABI bool match(StringRef S, bool IsSlashAgnostic = false) const;
+ LLVM_ABI bool match(StringRef S) const;
// Returns true for glob pattern "*". Can be used to avoid expensive
// preparation/acquisition of the input for match().
@@ -79,9 +76,8 @@ class GlobPattern {
struct SubGlobPattern {
/// \param Pat the pattern to match against
LLVM_ABI static Expected<SubGlobPattern> create(StringRef Pat);
- /// \param IsSlashAgnostic whether to treat '/' as also matching '\'
/// \returns \p true if \p S matches this glob pattern
- LLVM_ABI bool match(StringRef S, bool IsSlashAgnostic) const;
+ LLVM_ABI bool match(StringRef S) const;
StringRef getPat() const { return StringRef(Pat.data(), Pat.size()); }
// Brackets with their end position and matched bytes.
diff --git a/llvm/include/llvm/Support/SpecialCaseList.h b/llvm/include/llvm/Support/SpecialCaseList.h
index 951f27eed8ee8..5c5df23a8623d 100644
--- a/llvm/include/llvm/Support/SpecialCaseList.h
+++ b/llvm/include/llvm/Support/SpecialCaseList.h
@@ -122,11 +122,10 @@ class SpecialCaseList {
class Matcher {
public:
LLVM_ABI Error insert(StringRef Pattern, unsigned LineNumber,
- bool UseRegex);
+ bool UseGlobs);
// Returns the line number in the source file that this query matches to.
- // On windows, treat '/' as also matching '\' in filenames when using globs.
- // Returns zero if no match is found
- LLVM_ABI unsigned match(StringRef Query, bool IsFilename) const;
+ // Returns zero if no match is found.
+ LLVM_ABI unsigned match(StringRef Query) const;
struct Glob {
std::string Name;
@@ -155,6 +154,7 @@ class SpecialCaseList {
};
std::vector<Section> Sections;
+ bool CanonicalizePaths = false;
LLVM_ABI Expected<Section *> addSection(StringRef SectionStr,
unsigned FileIdx, unsigned LineNo,
diff --git a/llvm/lib/Support/GlobPattern.cpp b/llvm/lib/Support/GlobPattern.cpp
index e19c8c84d19a9..7004adf461a0c 100644
--- a/llvm/lib/Support/GlobPattern.cpp
+++ b/llvm/lib/Support/GlobPattern.cpp
@@ -137,7 +137,7 @@ GlobPattern::create(StringRef S, std::optional<size_t> MaxSubPatterns) {
GlobPattern Pat;
// Store the prefix that does not contain any metacharacter.
- size_t PrefixSize = S.find_first_of("?*[{\\/");
+ size_t PrefixSize = S.find_first_of("?*[{\\");
Pat.Prefix = S.substr(0, PrefixSize);
if (PrefixSize == std::string::npos)
return Pat;
@@ -190,22 +190,21 @@ GlobPattern::SubGlobPattern::create(StringRef S) {
return Pat;
}
-bool GlobPattern::match(StringRef S, bool IsSlashAgnostic) const {
+bool GlobPattern::match(StringRef S) const {
if (!S.consume_front(Prefix))
return false;
if (SubGlobs.empty() && S.empty())
return true;
for (auto &Glob : SubGlobs)
- if (Glob.match(S, IsSlashAgnostic))
+ if (Glob.match(S))
return true;
return false;
}
// Factor the pattern into segments split by '*'. The segment is matched
-// sequentially by finding the first occurrence past the end of the previous
+// sequentianlly by finding the first occurrence past the end of the previous
// match.
-bool GlobPattern::SubGlobPattern::match(StringRef Str,
- bool IsSlashAgnostic) const {
+bool GlobPattern::SubGlobPattern::match(StringRef Str) const {
const char *P = Pat.data(), *SegmentBegin = nullptr, *S = Str.data(),
*SavedS = S;
const char *const PEnd = P + Pat.size(), *const End = S + Str.size();
@@ -232,10 +231,6 @@ bool GlobPattern::SubGlobPattern::match(StringRef Str,
++S;
continue;
}
- } else if (IsSlashAgnostic && *P == '/' && *S == '\\') {
- ++P;
- ++S;
- continue;
} else if (*P == *S || *P == '?') {
++P;
++S;
diff --git a/llvm/lib/Support/SpecialCaseList.cpp b/llvm/lib/Support/SpecialCaseList.cpp
index c65cb977c005c..4de7478aaf53b 100644
--- a/llvm/lib/Support/SpecialCaseList.cpp
+++ b/llvm/lib/Support/SpecialCaseList.cpp
@@ -17,7 +17,6 @@
#include "llvm/ADT/STLExtras.h"
#include "llvm/Support/LineIterator.h"
#include "llvm/Support/MemoryBuffer.h"
-#include "llvm/Support/Path.h"
#include "llvm/Support/VirtualFileSystem.h"
#include <stdio.h>
#include <string>
@@ -67,13 +66,9 @@ Error SpecialCaseList::Matcher::insert(StringRef Pattern, unsigned LineNumber,
return Error::success();
}
-unsigned SpecialCaseList::Matcher::match(StringRef Query,
- bool IsFilename) const {
- static bool HaveWindowsPathStyle =
- llvm::sys::path::is_style_windows(llvm::sys::path::Style::native);
+unsigned SpecialCaseList::Matcher::match(StringRef Query) const {
for (const auto &Glob : reverse(Globs))
- if (Glob->Pattern.match(
- Query, /*IsSlashAgnostic=*/(HaveWindowsPathStyle && IsFilename)))
+ if (Glob->Pattern.match(Query))
return Glob->LineNo;
for (const auto &[Regex, LineNumber] : reverse(RegExes))
if (Regex->match(Query))
@@ -158,12 +153,17 @@ bool SpecialCaseList::parse(unsigned FileIdx, const MemoryBuffer *MB,
return false;
}
+ // Scan the start of the file for special comments. These don't appear when
+ // iterating below because comment lines are automatically skipped.
+ StringRef Buffer = MB->getBuffer();
// In https://reviews.llvm.org/D154014 we added glob support and planned to
// remove regex support in patterns. We temporarily support the original
- // behavior using regexes if "#!special-case-list-v1" is the first line of the
- // file. For more details, see
+ // behavior using regexes if "#!special-case-list-v1" is the first line of
+ // the file. For more details, see
// https://discourse.llvm.org/t/use-glob-instead-of-regex-for-specialcaselists/71666
- bool UseGlobs = !MB->getBuffer().starts_with("#!special-case-list-v1\n");
+ bool UseGlobs = !Buffer.consume_front("#!special-case-list-v1\n");
+ // Specifies that patterns should be matched against canonicalized filepaths.
+ CanonicalizePaths = Buffer.consume_front("#!canonical-paths\n");
for (line_iterator LineIt(*MB, /*SkipBlanks=*/true, /*CommentMarker=*/'#');
!LineIt.is_at_eof(); LineIt++) {
@@ -223,8 +223,7 @@ std::pair<unsigned, unsigned>
SpecialCaseList::inSectionBlame(StringRef Section, StringRef Prefix,
StringRef Query, StringRef Category) const {
for (const auto &S : reverse(Sections)) {
- bool IsFilename = Prefix == "src" || Prefix == "mainfile";
- if (S.SectionMatcher->match(Section, IsFilename)) {
+ if (S.SectionMatcher->match(Section)) {
unsigned Blame = inSectionBlame(S.Entries, Prefix, Query, Category);
if (Blame)
return {S.FileIdx, Blame};
@@ -243,8 +242,12 @@ unsigned SpecialCaseList::inSectionBlame(const SectionEntries &Entries,
if (II == I->second.end())
return 0;
- bool IsFilename = Prefix == "src" || Prefix == "mainfile";
- return II->getValue().match(Query, IsFilename);
+ if (CanonicalizePaths && (Prefix == "src" || Prefix == "mainfile")) {
+ return II->getValue().match(llvm::sys::path::convert_to_slash(
+ llvm::sys::path::remove_leading_dotslash(Query)));
+ } else {
+ return II->getValue().match(Query);
+ }
}
} // namespace llvm
diff --git a/llvm/unittests/Support/GlobPatternTest.cpp b/llvm/unittests/Support/GlobPatternTest.cpp
index ba33e233c70bc..e4f1025b00956 100644
--- a/llvm/unittests/Support/GlobPatternTest.cpp
+++ b/llvm/unittests/Support/GlobPatternTest.cpp
@@ -271,13 +271,4 @@ TEST_F(GlobPatternTest, Pathological) {
EXPECT_FALSE(Pat->match(S));
EXPECT_TRUE(Pat->match(S + 'b'));
}
-
-TEST_F(GlobPatternTest, SlashAgnostic) {
- auto Pat = GlobPattern::create("clang/*");
- ASSERT_TRUE((bool)Pat);
- EXPECT_TRUE(Pat->match("clang/foo"));
- EXPECT_FALSE(Pat->match(R"(clang\foo)"));
- EXPECT_TRUE(Pat->match("clang/foo", /*isSlashAgnostic=*/true));
- EXPECT_TRUE(Pat->match(R"(clang\foo)", /*isSlashAgnostic=*/true));
-}
}
diff --git a/llvm/unittests/Support/SpecialCaseListTest.cpp b/llvm/unittests/Support/SpecialCaseListTest.cpp
index 5be2b9e3a7a5d..00100beff6e94 100644
--- a/llvm/unittests/Support/SpecialCaseListTest.cpp
+++ b/llvm/unittests/Support/SpecialCaseListTest.cpp
@@ -372,4 +372,22 @@ TEST_F(SpecialCaseListTest, FileIdx) {
sys::fs::remove(Path);
}
+#ifdef _WIN32
+TEST_F(SpecialCaseListTest, CanonicalizePathsOnWindows) {
+ std::unique_ptr<SpecialCaseList> SCL =
+ makeSpecialCaseList("#!canonical-paths\n"
+ "\n"
+ "src:*foo/bar*\n"
+ "src:*foo\\\\baz\n"
+ "fun:hi\\\\bye=category\n");
+ EXPECT_TRUE(SCL->inSection("", "src", "foo/bar"));
+ EXPECT_TRUE(SCL->inSection("", "src", "foo\\bar"));
+ // The baz pattern doesn't match because paths are canonicalized first
+ EXPECT_FALSE(SCL->inSection("", "src", "foo/baz"));
+ EXPECT_FALSE(SCL->inSection("", "src", "foo\\baz"));
+ // The canonicalization only applies to files
+ EXPECT_TRUE(SCL->inSection("", "src", "hi\\bye", "category"));
+}
+#endif
+
} // namespace
More information about the llvm-commits
mailing list