Add clang-tidy check for thread_local vars

[theuni]

Forbid thread_local vars with non-trivial destructors.

This is a follow-up from: #30095 (comment)

[DrahtBot]

The following sections might be updated with supplementary metadata relevant to reviewers and maintainers.

Code Coverage

For detailed information about the code coverage, see the test coverage report.

Reviews

See the guideline for information on the review process.

Type	Reviewers
ACK	TheCharlatan, maflcko
Concept ACK	laanwj, hebasto

If your review is incorrectly listed, please react with 👎 to this comment and the bot will ignore it on the next update.

[laanwj]

Concept ACK

[theuni]

This PR originally contained a commit which purposely added a violation in init.cpp:

thread_local std::string bad_var;

From the tidy c-i task:

init.cpp:723:5: error: Variable with non-trivial destructor cannot be thread_local. [bitcoin-nontrivial-threadlocal,-warnings-as-errors]
723 | thread_local std::string bad_var;

Since it was correctly detected, I'll now remove that commit.

[DrahtBot]

🚧 At least one of the CI tasks failed. Make sure to run all tests locally, according to the
documentation.

Possibly this is due to a silent merge conflict (the changes in this pull request being
incompatible with the current code in the target branch). If so, make sure to rebase on the latest
commit of the target branch.

Leave a comment here, if you need help tracking down a confusing failure.

_{Debug: https://github.com/bitcoin/bitcoin/runs/25236805741}

[hebasto]

Concept ACK.

[TheCharlatan]

How about:

diff --git a/contrib/devtools/bitcoin-tidy/example_nontrivial-threadlocal.cpp b/contrib/devtools/bitcoin-tidy/example_nontrivial-threadlocal.cpp
index 2b74df5d0e..b002712baf 100644
--- a/contrib/devtools/bitcoin-tidy/example_nontrivial-threadlocal.cpp
+++ b/contrib/devtools/bitcoin-tidy/example_nontrivial-threadlocal.cpp
@@ -2,0 +3,30 @@ thread_local std::string foo;
+thread_local char* bar;
+
+struct Test_Ok_1 {
+    int member;
+    Test_Ok_1() {
+        member = 1;
+    }
+    ~Test_Ok_1() {}
+};
+
+struct Test_Ok_2 {
+    int member;
+    Test_Ok_2() {
+        member = 1;
+    }
+};
+
+struct Test_Bad {
+    int member;
+    Test_Bad() {
+        member = 1;
+    }
+    ~Test_Bad() {
+        member = 0;
+    }
+};
+
+thread_local Test_Ok_1 test_1;
+thread_local Test_Ok_2 test_2;
+thread_local Test_Bad test_3;
diff --git a/contrib/devtools/bitcoin-tidy/nontrivial-threadlocal.cpp b/contrib/devtools/bitcoin-tidy/nontrivial-threadlocal.cpp
index bce1a99165..3f5a85e444 100644
--- a/contrib/devtools/bitcoin-tidy/nontrivial-threadlocal.cpp
+++ b/contrib/devtools/bitcoin-tidy/nontrivial-threadlocal.cpp
@@ -14,2 +14,19 @@ AST_MATCHER(clang::CXXRecordDecl, hasNonTrivialDestructor) {
-  // TODO: If the dtor is there but empty we don't want to warn either.
-  return Node.hasDefinition() && Node.hasNonTrivialDestructor();
+  if (!Node.hasDefinition()) {
+    return false;
+  }
+  if (Node.hasTrivialDestructor()) {
+    return false;
+  }
+  const clang::CXXDestructorDecl* dtor = Node.getDestructor();
+  if (!dtor) {
+    return false;
+  }
+
+  if (dtor->hasBody()) {
+    const clang::Stmt* body = dtor->getBody();
+    if (const clang::CompoundStmt* comp_stmt = clang::dyn_cast<clang::CompoundStmt>(body)) {
+      return !comp_stmt->body_empty();
+    }
+  }
+
+  return false;

[maflcko]

I don't have freeBSD to test this, but are you sure it doesn't crash when the dtor is there but empty?

In any case, what is the point of an empty dtor? Might as well remove it, or modernize-use-equals-default it?

[theuni]

@TheCharlatan Thanks for playing around with it :)

I don't have the internal clang knowledge to know if that's correct or not. Looks fine to me, but I agree with @maflcko that we could just remove the empty dtor in that case.

Also, to be clear, that TODO is an upstream comment, not one from me: https://github.com/llvm/llvm-project/blob/main/clang-tools-extra/clang-tidy/bugprone/UnusedRaiiCheck.cpp#L18

[TheCharlatan]

I don't have freeBSD to test this, but are you sure it doesn't crash when the dtor is there but empty?

It looks like it should work to me. If a method has a body it should also contain a compound statement, or at least this book seems to suggest as much.

Also, to be clear, that TODO is an upstream comment, not one from me: https://github.com/llvm/llvm-project/blob/main/clang-tools-extra/clang-tidy/bugprone/UnusedRaiiCheck.cpp#L18

...but if it is an upstream TODO I might well be missing something.

[maflcko]

I don't have freeBSD to test this, but are you sure it doesn't crash when the dtor is there but empty?

It looks like it should work to me. If a method has a body it should also contain a compound statement, or at least this book seems to suggest as much.

Oh, I didn't mean clang-tidy to crash, but the program that uses thread_local in reference to https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=278701

[maflcko]

modernize-use-equals-default

See #30406

[TheCharlatan]

ACK ace5b35

[theuni]

Updated to resolve @TheCharlatan's nit, and also specified where hasNonTrivialDestructor came from.

[TheCharlatan]

Re-ACK 34c9cee

[maflcko]

ACK 34c9cee