[PATCH] D116593: Fix `performance-unnecessary-value-param` for template specialization

[Felix Berger via Phabricator via cfe-commits]

flx added inline comments.


================
Comment at: clang-tools-extra/clang-tidy/performance/UnnecessaryValueParamCheck.cpp:177-178
     const auto &CurrentParam = *FunctionDecl->getParamDecl(Index);
+    if (IsExplicitTemplateSpecialization && Function != FunctionDecl)
+      continue;
     Diag << utils::fixit::changeVarDeclToReference(CurrentParam,
----------------
Sockke wrote:
> flx wrote:
> > Could you add a comment here why we're skipping the fix here?
> > Could you add a comment here why we're skipping the fix here?
> 
> Specialization template may match the primary template again by `getPreviousDecl`. Skipping the fix to avoid repeated fixes for the primary template.
Thanks for the explanation. Would you mind adding this as code comment?


================
Comment at: clang-tools-extra/test/clang-tidy/checkers/performance-unnecessary-value-param.cpp:388
+  // CHECK-MESSAGES: [[@LINE-1]]:54: warning: the parameter 'E' is copied
+  // CHECK-FIXES: T templateSpecializationFunction(const ExpensiveToCopyType& E) {
+  return T();
----------------
Sockke wrote:
> flx wrote:
> > Should we apply the fixes or just issue the warning? For virtual methods we suppress the fix since we can't necessarily update all overrides of the method. Are template specializations always guaranteed to be in the same translation unit which  would make this safe?
> > Should we apply the fixes or just issue the warning? For virtual methods we suppress the fix since we can't necessarily update all overrides of the method. Are template specializations always guaranteed to be in the same translation unit which  would make this safe?
> 
> Do you mean that specialization templates are defined in different translation units?  If fixing one by one translation unit does have the problem, the `readability-const-return-type` also has such a problem. clang-tidy can not analyze across translation units, but the diagnosis and fix of it are separate, we can specify the complete `compile_commands.json` to avoid it. 
> I'm not sure whether this is reasonable, we may make a choice between clang-tidy's fault tolerance and advantages. What's your suggestion?
> > Should we apply the fixes or just issue the warning? For virtual methods we suppress the fix since we can't necessarily update all overrides of the method. Are template specializations always guaranteed to be in the same translation unit which  would make this safe?
> 
> Do you mean that specialization templates are defined in different translation units?  If fixing one by one translation unit does have the problem, the `readability-const-return-type` also has such a problem. clang-tidy can not analyze across translation units, but the diagnosis and fix of it are separate, we can specify the complete `compile_commands.json` to avoid it. 

I'm not sure how I understand how the compile command json file would help. I think std::hash (https://en.cppreference.com/w/cpp/utility/hash) is an example where you would specialize a type across translation units. If std::hash itself had the issue you wouldn't be able to modify it since it is an external library.

> I'm not sure whether this is reasonable, we may make a choice between clang-tidy's fault tolerance and advantages. What's your suggestion?

Yeah, this is not an easy balance to strike. ClangTidy is used in many different contexts, in some where you would like to see aggressive fixes, in others where only the safest fixes should be applied.

>From previous reviews and my own experience we should by default only provide safe fixes with low false positive rate. If you think its' not enough to only notify the user of an inefficiency in template specializations you could add an option to the check for fixing these occurrences.

This was done for example here to make a check find issues (including false positives) and report them:  https://github.com/llvm/llvm-project/blob/765dd8b8a44cd9689c87c0433739f421b9871061/clang-tools-extra/clang-tidy/performance/ForRangeCopyCheck.cpp#L26 

In summary, I would make this a warning only, but not issue the fix here.




Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D116593/new/

https://reviews.llvm.org/D116593