If #undef is only inserted into Procfs.h or conditioned for ANDROID, that will only fix one include path or use case.
These .h files won't compile for other include paths or with any other system files that define FS or CS.
An alternative is not to use FS or CS in these .h files, but #undef before use seems less intrusive.

fix windows test failure

fix windows test failure

Factor out LocationFilter; improve more tests.

Factor out LocationFilter; improve more test cases.

This is enhanced implementation over D98710, to handle the new found failed test cases.

sync to the latest source; apply clang-format; add new skip-headers-4.cpp test

sync to the latest source; apply clang-format; add new skip-headers-4.cpp test;
skip modernize-use-nullptr warnings in header files in UseNullptrCheck.cpp

apply clang-format

Add new test cases into misc-unused-using-decls.cpp.

• Use clang-format layout in ClangTidy.cpp and ClangTidyModule.h.
• Fix misc-unused-using-decls false-positive warning from --skip-headers.
  • UnusedUsingDeclsCheck is now registered as an AllFileCheck.
  • Add new test cases into misc-unused-using-decls.cpp.

add skip-headers-3.cpp test to show that checking only Decl locations is not enough to skip some warnings in header files

add skip-headers-3.cpp test to show that checking only Decl locations is not enough to skip some warnings in header files

Add an AllFileFinder for checks that need to match all source files.
ForwardDeclarationNamespaceCheck is registered as an AllFileCheck,

Some Android developers and legacy code care less about clang-tidy warnings.
Newer developers spend a lot of time to get lint-free new code.
So Android source tree has a lot of clang-tidy flags like
header-filter, checks, and warnings-as-errors to select checks
for different modules. People are even asking for new features like
making header-filter to accept a list of regexp like the checks flag.

Add one more skip-headers-2.cpp test, which shows desired check at a top-level Decl but not at nested Decl. Checking locations of only top-level Decls will miss the opportunity to skip nested Decls.

Add one more skip-headers-2.cpp test.

Add bugprone-forward-declaration-namespace-header.cpp test to
show that MatchFinder-based checks can also depend on header Decls.

Add bugprone-forward-declaration-namespace-header.cpp test; fix some coding style warnings.

The latest tested change contained one of your suggestions:

using setTraversalScope for all consumers in the
ClangTidyASTConsumer::HandleTranslationUnit

It is good to pass all existing tests,
although not a proof of correctness yet.

--skip-headers is set to false by default;
many tidy tests runs are augmented with explicit --skip-headers or --skip-headers=0
to test this feature as on or off.

--skip-headers is set to false by default;
many tidy tests runs are augmented with explicit --skip-headers or --skip-headers=0
to test this feature as on or off.

to test skip-headers as default, do not submit;
show that setTraversalScope failed some static analyzer checks,
see #if 0 comments in ClangTidy.cpp.

only coding style changes in clang-tidy files, to match other alternative implementation

sync with latest source, update more tests to test skip-headers as default

Sam, the revision summary is updated. Could you review it again?

sync up latest clang/llvm source; more format/style changes

No more change to or derivation from MatchFinder, but override two member functions in MultiplexConsumer.

fix clang-format coding style issue

fix clang-format name style issue

Last build tests were green. Let's see if the new updated diff still passes all the tests.

with clang-format suggested change

I reproduced the ClangdTests::TargetDeclTest.Concept failure locally without this change,
using llvm git commit 0c400e8953. Will wait for upstream fix of the failed tests.

Move almost all changes to MatchFinder into ClangTidy.cpp.

update some comments and sync again with the latest clang/llvm source

Sync up; moved DeclFilter to ClangTidy.cpp.

Friendly ping reviewers. Now this change can pass all clang-tidy tests.
Please feel free to comment and/or add other reviewers. Thanks.

Please review https://reviews.llvm.org/D98710, which uses get/setTraversalScope instead of adding a TopLevelDecl bit.

sync with latest clang/llvm

sync with latest clang/llvm

sync with latest clang/llvm

hide the --show-all-warnings flag from --help

hide the --show-all-warnings flag from --help

hide the --show-all-warnings flag from --help

fix windows test failures

Here are some important sections copied from my design docs.
This patch presents only one of the simplest working implementation I have tried.
I hope it would serve as a good baseline to compare any alternatives.

In D98618#2625737, @sammccall wrote:

In D98618#2625675, @njames93 wrote:

Can I ask what is the purpose of --skip-headers?
What does it offer that --header-filter doesn't already offer?

TL;DR: it avoids traversing a bunch of nodes where diagnostics will be filtered out anyway, thus running significantly faster.

@chh You're written quite a lot about the design of this, can you make this public please?

I applied this patch to current Android clang/llvm toolchain to build clang compiler.
Not all toolchain was built due to dependencies on other clang/llvm changes not picked into Android yet.
We will known only later when Android toolchain is updated with this and all dependent changes
whether f128 type still works on Android x86_64 devices.
As long as we keep this and future dependent changes small, we should be able to manage any regression found later.

Craig, I tested your patch with some AOSP libm code.

I haven't read all code details yet; just some quick questions and comments.

I probably won't have enough time to test this today.
This looks good to me. If you submit it now, I can cherry pick it and test later.
Thanks.

Make it a positive unit test. Create library to test from .s file.

Add a unit test case.

Rui, could you help me creating a test?
All I have now is Android's link command for dex2oatds.
I don't know how to reduce that to an lld unit test.

I tested with clang-tidy test.cpp -- -c -Iinc -flto,
and found that in Driver.cpp:getTool Inputs has one action equal to
BackendJobClass (without -flto) or CompileJobClass (with -flto).