[clang-tidy] Avoid matching nodes in system headers #151035
Open
+75
−12
Conversation
This file contains hidden or 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
llvmbot
added
clang
|
@llvm/pr-subscribers-clang-tools-extra @llvm/pr-subscribers-clang-tidy Author: Carlos Galvez (carlosgalvezp) ChangesThis commit is a re-do of e4a8969, which got reverted, with the same goal: dramatically speed-up clang-tidy by avoiding doing work in system headers (which is wasteful as warnings are later discarded). This proposal was already discussed here with favorable feedback: The novelty of this patch is:
As a benchmark, I ran clang-tidy with all checks activated, on a single .cpp file which #includes all the standard C++ headers, then measure the time as well as found warnings. On trunk: With this patch: With the original patch that got reverted: A lot of warnings remain and the runtime is sligthly higher, but we still got a dramatic reduction with no change in functionality. Further investigation has shown that all of the remaining warnings are due to However this may not be as straightforward or wanted, it may even need to be done on a per-check basis (there's about 10 checks or so that would need to explicitly ignore system headers). I will leave that for another patch, it's low priority and does not improve the runtime much (it just prints better statistics). Fixes #52959 Full diff: https://github.com/llvm/llvm-project/pull/151035.diff 7 Files Affected:
diff --git a/clang-tools-extra/clang-tidy/ClangTidy.cpp b/clang-tools-extra/clang-tidy/ClangTidy.cpp
index e84be0461f280..eb91484363d6f 100644
--- a/clang-tools-extra/clang-tidy/ClangTidy.cpp
+++ b/clang-tools-extra/clang-tidy/ClangTidy.cpp
@@ -425,6 +425,10 @@ ClangTidyASTConsumerFactory::createASTConsumer(
FinderOptions.CheckProfiling.emplace(Profiling->Records);
}
+ // Avoid processing system headers, unless the user explicitly requests it
+ if (!Context.getOptions().SystemHeaders.value_or(false))
+ FinderOptions.IgnoreSystemHeaders = true;
+
std::unique_ptr<ast_matchers::MatchFinder> Finder(
new ast_matchers::MatchFinder(std::move(FinderOptions)));
diff --git a/clang-tools-extra/docs/ReleaseNotes.rst b/clang-tools-extra/docs/ReleaseNotes.rst
index 3ea1c5104316f..f761665677f20 100644
--- a/clang-tools-extra/docs/ReleaseNotes.rst
+++ b/clang-tools-extra/docs/ReleaseNotes.rst
@@ -97,6 +97,10 @@ Improvements to clang-tidy
now run checks in parallel by default using all available hardware threads.
Both scripts display the number of threads being used in their output.
+- :program:`clang-tidy` no longer attemps to match AST nodes from system headers
+ by default, greatly improving performance. This behavior is disabled if the
+ `SystemHeaders` option is enabled.
+
New checks
^^^^^^^^^^
diff --git a/clang-tools-extra/test/clang-tidy/infrastructure/file-filter.cpp b/clang-tools-extra/test/clang-tidy/infrastructure/file-filter.cpp
index 448ef9ddf166c..d9ec1049963b0 100644
--- a/clang-tools-extra/test/clang-tidy/infrastructure/file-filter.cpp
+++ b/clang-tools-extra/test/clang-tidy/infrastructure/file-filter.cpp
@@ -66,19 +66,14 @@ class A { A(int); };
// CHECK4-NOT: warning:
// CHECK4-QUIET-NOT: warning:
-// CHECK: Suppressed 3 warnings (3 in non-user code)
// CHECK: Use -header-filter=.* to display errors from all non-system headers.
// CHECK-QUIET-NOT: Suppressed
-// CHECK2: Suppressed 1 warnings (1 in non-user code)
-// CHECK2: Use -header-filter=.* {{.*}}
// CHECK2-QUIET-NOT: Suppressed
-// CHECK3: Suppressed 2 warnings (2 in non-user code)
// CHECK3: Use -header-filter=.* {{.*}}
// CHECK3-QUIET-NOT: Suppressed
// CHECK4-NOT: Suppressed {{.*}} warnings
// CHECK4-NOT: Use -header-filter=.* {{.*}}
// CHECK4-QUIET-NOT: Suppressed
-// CHECK6: Suppressed 2 warnings (2 in non-user code)
// CHECK6: Use -header-filter=.* {{.*}}
int x = 123;
diff --git a/clang-tools-extra/test/clang-tidy/infrastructure/system-headers.cpp b/clang-tools-extra/test/clang-tidy/infrastructure/system-headers.cpp
index 9fa990b6aac8c..a25480e9aa39c 100644
--- a/clang-tools-extra/test/clang-tidy/infrastructure/system-headers.cpp
+++ b/clang-tools-extra/test/clang-tidy/infrastructure/system-headers.cpp
@@ -11,9 +11,9 @@
// RUN: clang-tidy -help | FileCheck -check-prefix=CHECK-OPT-PRESENT %s
// RUN: clang-tidy -checks='-*,google-explicit-constructor' -header-filter='.*' -system-headers=true %s -- -isystem %S/Inputs/system-headers 2>&1 | FileCheck -check-prefix=CHECK-SYSTEM-HEADERS %s
-// RUN: clang-tidy -checks='-*,google-explicit-constructor' -header-filter='.*' -system-headers=false %s -- -isystem %S/Inputs/system-headers 2>&1 | FileCheck -check-prefix=CHECK-NO-SYSTEM-HEADERS %s
+// RUN: clang-tidy -checks='-*,google-explicit-constructor' -header-filter='.*' -system-headers=false %s -- -isystem %S/Inputs/system-headers 2>&1 | FileCheck -check-prefix=CHECK-NO-SYSTEM-HEADERS --allow-empty %s
// RUN: clang-tidy -checks='-*,google-explicit-constructor' -header-filter='.*' -config='SystemHeaders: true' %s -- -isystem %S/Inputs/system-headers 2>&1 | FileCheck -check-prefix=CHECK-SYSTEM-HEADERS %s
-// RUN: clang-tidy -checks='-*,google-explicit-constructor' -header-filter='.*' -config='SystemHeaders: false' %s -- -isystem %S/Inputs/system-headers 2>&1 | FileCheck -check-prefix=CHECK-NO-SYSTEM-HEADERS %s
+// RUN: clang-tidy -checks='-*,google-explicit-constructor' -header-filter='.*' -config='SystemHeaders: false' %s -- -isystem %S/Inputs/system-headers 2>&1 | FileCheck -check-prefix=CHECK-NO-SYSTEM-HEADERS --allow-empty %s
#include <system_header.h>
// CHECK-SYSTEM-HEADERS: system_header.h:1:13: warning: single-argument constructors must be marked explicit
diff --git a/clang/docs/ReleaseNotes.rst b/clang/docs/ReleaseNotes.rst
index ec51ffddce1af..7a0b833dd7bce 100644
--- a/clang/docs/ReleaseNotes.rst
+++ b/clang/docs/ReleaseNotes.rst
@@ -207,6 +207,9 @@ AST Matchers
- Ensure ``hasBitWidth`` doesn't crash on bit widths that are dependent on template
parameters.
+- Add a boolean member ``IgnoreSystemHeaders`` to ``MatchFinderOptions``. This
+ allows it to ignore nodes in system headers when traversing the AST.
+
clang-format
------------
diff --git a/clang/include/clang/ASTMatchers/ASTMatchFinder.h b/clang/include/clang/ASTMatchers/ASTMatchFinder.h
index 73cbcf1f25025..2d36e8c4fae1c 100644
--- a/clang/include/clang/ASTMatchers/ASTMatchFinder.h
+++ b/clang/include/clang/ASTMatchers/ASTMatchFinder.h
@@ -135,10 +135,15 @@ class MatchFinder {
llvm::StringMap<llvm::TimeRecord> &Records;
};
+ MatchFinderOptions() {}
+
/// Enables per-check timers.
///
/// It prints a report after match.
std::optional<Profiling> CheckProfiling;
+
+ /// Avoids matching declarations in system headers.
+ bool IgnoreSystemHeaders{false};
};
MatchFinder(MatchFinderOptions Options = MatchFinderOptions());
diff --git a/clang/lib/ASTMatchers/ASTMatchFinder.cpp b/clang/lib/ASTMatchers/ASTMatchFinder.cpp
index 6d0ba0b7907a1..15d9183c9fddb 100644
--- a/clang/lib/ASTMatchers/ASTMatchFinder.cpp
+++ b/clang/lib/ASTMatchers/ASTMatchFinder.cpp
@@ -1336,6 +1336,45 @@ class MatchASTVisitor : public RecursiveASTVisitor<MatchASTVisitor>,
return false;
}
+ bool isInSystemHeader(const SourceLocation &Loc) {
+ const SourceManager &SM = getASTContext().getSourceManager();
+ return SM.isInSystemHeader(Loc);
+ }
+
+ template <typename T> SourceLocation getNodeLocation(T const &Node) {
+ return Node.getBeginLoc();
+ }
+
+ SourceLocation getNodeLocation(QualType const &Node) { return {}; }
+
+ SourceLocation getNodeLocation(NestedNameSpecifier const &Node) { return {}; }
+
+ SourceLocation getNodeLocation(CXXCtorInitializer const &Node) {
+ return Node.getSourceLocation();
+ }
+
+ SourceLocation getNodeLocation(TemplateArgumentLoc const &Node) {
+ return Node.getLocation();
+ }
+
+ SourceLocation getNodeLocation(Attr const &Node) {
+ return Node.getLocation();
+ }
+
+ template <typename T>
+ auto shouldSkipNode(T const &Node)
+ -> std::enable_if_t<std::is_pointer_v<T>, bool> {
+ return (Node == nullptr) || shouldSkipNode(*Node);
+ }
+
+ template <typename T>
+ auto shouldSkipNode(T const &Node)
+ -> std::enable_if_t<!std::is_pointer_v<T>, bool> {
+ if (Options.IgnoreSystemHeaders && isInSystemHeader(getNodeLocation(Node)))
+ return true;
+ return false;
+ }
+
/// Bucket to record map.
///
/// Used to get the appropriate bucket for each matcher.
@@ -1465,9 +1504,8 @@ bool MatchASTVisitor::objcClassIsDerivedFrom(
}
bool MatchASTVisitor::TraverseDecl(Decl *DeclNode) {
- if (!DeclNode) {
+ if (shouldSkipNode(DeclNode))
return true;
- }
bool ScopedTraversal =
TraversingASTNodeNotSpelledInSource || DeclNode->isImplicit();
@@ -1495,9 +1533,9 @@ bool MatchASTVisitor::TraverseDecl(Decl *DeclNode) {
}
bool MatchASTVisitor::TraverseStmt(Stmt *StmtNode, DataRecursionQueue *Queue) {
- if (!StmtNode) {
+ if (shouldSkipNode(StmtNode))
return true;
- }
+
bool ScopedTraversal = TraversingASTNodeNotSpelledInSource ||
TraversingASTChildrenNotSpelledInSource;
@@ -1507,11 +1545,17 @@ bool MatchASTVisitor::TraverseStmt(Stmt *StmtNode, DataRecursionQueue *Queue) {
}
bool MatchASTVisitor::TraverseType(QualType TypeNode) {
+ if (shouldSkipNode(TypeNode))
+ return true;
+
match(TypeNode);
return RecursiveASTVisitor<MatchASTVisitor>::TraverseType(TypeNode);
}
bool MatchASTVisitor::TraverseTypeLoc(TypeLoc TypeLocNode) {
+ if (shouldSkipNode(TypeLocNode))
+ return true;
+
// The RecursiveASTVisitor only visits types if they're not within TypeLocs.
// We still want to find those types via matchers, so we match them here. Note
// that the TypeLocs are structurally a shadow-hierarchy to the expressed
@@ -1523,6 +1567,9 @@ bool MatchASTVisitor::TraverseTypeLoc(TypeLoc TypeLocNode) {
}
bool MatchASTVisitor::TraverseNestedNameSpecifier(NestedNameSpecifier *NNS) {
+ if (shouldSkipNode(NNS))
+ return true;
+
match(*NNS);
return RecursiveASTVisitor<MatchASTVisitor>::TraverseNestedNameSpecifier(NNS);
}
@@ -1532,6 +1579,9 @@ bool MatchASTVisitor::TraverseNestedNameSpecifierLoc(
if (!NNS)
return true;
+ if (shouldSkipNode(NNS))
+ return true;
+
match(NNS);
// We only match the nested name specifier here (as opposed to traversing it)
@@ -1544,7 +1594,7 @@ bool MatchASTVisitor::TraverseNestedNameSpecifierLoc(
bool MatchASTVisitor::TraverseConstructorInitializer(
CXXCtorInitializer *CtorInit) {
- if (!CtorInit)
+ if (shouldSkipNode(CtorInit))
return true;
bool ScopedTraversal = TraversingASTNodeNotSpelledInSource ||
@@ -1562,11 +1612,17 @@ bool MatchASTVisitor::TraverseConstructorInitializer(
}
bool MatchASTVisitor::TraverseTemplateArgumentLoc(TemplateArgumentLoc Loc) {
+ if (shouldSkipNode(Loc))
+ return true;
+
match(Loc);
return RecursiveASTVisitor<MatchASTVisitor>::TraverseTemplateArgumentLoc(Loc);
}
bool MatchASTVisitor::TraverseAttr(Attr *AttrNode) {
+ if (shouldSkipNode(AttrNode))
+ return true;
+
match(*AttrNode);
return RecursiveASTVisitor<MatchASTVisitor>::TraverseAttr(AttrNode);
}
|
carlosgalvezp
requested review from
Xazax-hun,
AaronBallman,
steakhal,
5chmidti and
vbvictor
carlosgalvezp
force-pushed
the
system_headers_new
branch
2 times, most recently
from
d52db8c to
cc9a245
Compare
vbvictor
reviewed
Jul 28, 2025
5chmidti
reviewed
Jul 28, 2025
There was a problem hiding this comment.
Awesome to see this continuing. The changes also look quite clean and extendable to, e.g., modules, like you mentioned. The performance improvement is still incredible to see. Especially when considering library layering in user-code from external or internal libraries, giving this a potentially much stronger impact on analysis time. I'm not sure if we could recover the last ~2.5s compared to the previous change without breaking something.
It's probably an LGTM from me, but I will think about the change for a bit and try it out a bit in a few days (e.g., w.r.t. macros).
| @@ -135,10 +135,15 @@ class MatchFinder { | |||
| llvm::StringMap<llvm::TimeRecord> &Records; | |||
| }; | |||
|
|
|||
| MatchFinderOptions() {} | |||
There was a problem hiding this comment.
MatchFinderOptions() = default; (why add it in the first place? Having no ctor should be enough)
There was a problem hiding this comment.
I know, it bothers me a lot as well. I am not sure if this is a compiler bug or expected, but the moment I add a non-static member initializer (NSDMI) to the struct, I get the following compiler error:
https://godbolt.org/z/Mb3bnxjjM
Using = default surprisingly does not work :(
The solutions are:
- Move
MatchFinderOptionsoutside theMatchFinderclass (like I did on the original patch). I didn't like this because it introduced unrelated noise to the patch, but I'm happy to do it on a separate patch if we feel it's the cleanest. - Not use NSDMI, and instead create a constructor with member initializer list.
- Add a non-default user-defined constructor (this patch).
carlosgalvezp
force-pushed
the
system_headers_new
branch
2 times, most recently
from
38fb173 to
220d450
Compare
|
I'm happy to split the little refactor of |
I think this patch is good and small enough as a whole, so not need to worry about further reduction. |
@5chmidti Here goes 2 more seconds :) With both patches, the result is very close to the reverted patch, just 0.5 seconds diff :) Maybe the PPCallbacks aren't worth touching after all. |
Could you please save this file (or make a gist) for later. I think we could have some benchmarks of |
|
Something else to benchmark: #109450. |
|
@firewave Here's the results: On trunk: With this patch: |
|
Benchmark of whole LLVM with new option With patch: So we are having ~70% speedup overall. |
|
@Xazax-hun @steakhal @AaronBallman Do you have any feedback on the changes to |
Xazax-hun
reviewed
Aug 3, 2025
As it is right now, we do not perform any expansion of the If this is deemed too aggressive I can change it to perform the check on the expanded location instead. This would be consistent with the reverted patch as well as some checks that already do |
This is not always true, e.g., when it comes to macros with arguments. I am mostly interested in the scenario where someone has something like: In one of the files we check. Would we still get warnings for |
steakhal
reviewed
Aug 3, 2025
carlosgalvezp
force-pushed
the
system_headers_new
branch
from
220d450 to
89a2228
Compare
carlosgalvezp
force-pushed
the
system_headers_new
branch
from
89a2228 to
638b35e
Compare
Yes, we would. For example, a simple: Triggers It's also my understanding that clang-tidy already filters out all system header diagnostics, before doing macro expansion: |
This commit is a re-do of e4a8969, which got reverted, with the same goal: dramatically speed-up clang-tidy by avoiding doing work in system headers (which is wasteful as warnings are later discarded). This proposal was already discussed here with favorable feedback: llvm#132725 The novelty of this patch is: - It's less aggressive: it does not fiddle with AST traversal. This solves the issue with the previous patch, which impacted the ability to inspect parents of a given node. - Instead, what we optimize for is exitting early in each Traverse* function of MatchASTVisitor if the node is in a system header, thus avoiding calling the match() function with its corresponding callback (when there is a match). - It does not cause any failing tests. - It does not move MatchFinderOptions outside - instead we add a user-defined default constructor which solves the same problem. - It introduces a function "shouldSkipNode" which can be extended for adding more conditions. For example there's a PR open about skipping modules in clang-tidy where this could come handy: llvm#145630 As a benchmark, I ran clang-tidy with all checks activated, on a single .cpp file which #includes all the standard C++ headers, then measure the time as well as found warnings. On trunk: Suppressed 213314 warnings (213314 in non-user code). real 0m14.311s user 0m14.126s sys 0m0.185s With this patch: Suppressed 149399 warnings (149399 in non-user code). real 0m3.583s user 0m3.454s sys 0m0.128s With the original patch that got reverted: Suppressed 8050 warnings (8050 in non-user code). Runtime: around 1 second. A lot of warnings remain and the runtime is sligthly higher, but we still got a dramatic reduction with no change in functionality. Further investigation has shown that all of the remaining warnings are due to PPCallbacks - implementing a similar system-header exclusion mechanism there can lead to almost no warnings left in system headers. This does not bring the runtime down as much, though. However this may not be as straightforward or wanted, it may even need to be done on a per-check basis (there's about 10 checks or so that would need to explicitly ignore system headers). I will leave that for another patch, it's low priority and does not improve the runtime much (it just prints better statistics). Fixes llvm#52959
carlosgalvezp
force-pushed
the
system_headers_new
branch
from
638b35e to
ba0f486
Compare
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.
This commit is a re-do of e4a8969, which got reverted, with the same goal: dramatically speed-up clang-tidy by avoiding doing work in system headers (which is wasteful as warnings are later discarded). This proposal was already discussed here with favorable feedback: #132725
The novelty of this patch is:
It's less aggressive: it does not fiddle with AST traversal. This solves the issue with the previous patch, which impacted the ability to inspect parents of a given node.
Instead, what we optimize for is exitting early in each
Traverse*function ofMatchASTVisitorif the node is in a system header, thus avoiding calling thematch()function with its corresponding callback (when there is a match).It does not cause any failing tests.
It does not move
MatchFinderOptions- instead we add a user-defined default constructor which solves the same problem.It introduces a function
shouldSkipNodewhich can be extended for adding more conditions. For example there's a PR open about skipping modules in clang-tidy where this could come handy: [clang-tidy] [Modules] Skip checking decls in clang-tidy #145630As a benchmark, I ran clang-tidy with all checks activated, on a single .cpp file which #includes all the standard C++ headers, then measure the time as well as found warnings.
On trunk:
With this patch:
With the original patch that got reverted:
We therefore get a dramatic reduction in number of warnings and runtime, with no change in functionality.
The remaining warnings are due to
PPCallbacks- implementing a similar system-header exclusion mechanism there can lead to almost no warnings left in system headers. This does not bring the runtime down as much, though, so it's probably not worth the effort.Fixes #52959