[PATCH] D82904: [clang-tidy] Warn pointer captured in async block

[Aaron Ballman via Phabricator via cfe-commits]

aaron.ballman added inline comments.


================
Comment at: clang-tools-extra/clang-tidy/bugprone/BugproneTidyModule.cpp:82
         "bugprone-bool-pointer-implicit-conversion");
-    CheckFactories.registerCheck<BranchCloneCheck>(
-        "bugprone-branch-clone");
+    CheckFactories.registerCheck<BranchCloneCheck>("bugprone-branch-clone");
     CheckFactories.registerCheck<CopyConstructorInitCheck>(
----------------
ellis wrote:
> aaron.ballman wrote:
> > It looks like unrelated formatting changes may have snuck in?
> The script `add_new_check.py` doesn't format the code that it generated so a few lines aren't linted. Would you like me to undo these extra formatted lines?
Yes, please. The way I usually do this, btw, is to run the patch through clang-format-diff so that only the lines I've touched get formatted: https://clang.llvm.org/docs/ClangFormat.html#script-for-patch-reformatting


================
Comment at: clang-tools-extra/clang-tidy/bugprone/BugproneTidyModule.cpp:121
         "bugprone-narrowing-conversions");
+    CheckFactories.registerCheck<NoEscapeCheck>("bugprone-no-escape");
     CheckFactories.registerCheck<NotNullTerminatedResultCheck>(
----------------
ellis wrote:
> njames93 wrote:
> > aaron.ballman wrote:
> > > Given that this is limited to Objective-C, why register this under `bugprone` as opposed to the `objc` module? Alternatively, why limit to Objective-C when blocks can be used in other modes like C or C++ with `-fblocks`?
> > Thats a good point, maybe restrict this to `LangOptions::ObjC || LangOptions::Blocks` Then it can be left in bugprone.
> Ok, I'll include `LangOptions::Blocks`.
I think the only option you need is `LangOptions::Blocks` (it should be enabled automatically in ObjC language mode).


================
Comment at: clang-tools-extra/clang-tidy/bugprone/NoEscapeCheck.cpp:37
+    if (Var && Var->hasAttr<NoEscapeAttr>()) {
+      diag(MatchedEscapingBlock->getBeginLoc(),
+           "pointer %0 with attribute 'noescape' is captured by an "
----------------
ellis wrote:
> aaron.ballman wrote:
> > Given that the capture is the issue (not the block), why not point to the use of the captured variable?
> I actually agree that pointing to the use of the captured variable would be easier to read, but honestly I couldn't figure out how to grab the location of that use. All I could get was the definition of the variable.
Ah, yeah, I just looked at the class and it looks like we don't track that information yet and it would be somewhat involved to get it. Can you add a FIXME comment above the call to `diag()` mentioning that we'd prefer to diagnose the use of the capture rather than the block?


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D82904/new/

https://reviews.llvm.org/D82904