Added a test

Implemented review comment to restrict __stdcall default calling convention (for WinMain, wWinMain, and DllMain) to 32 bit Windows.

LGTM Thanks!

In D87701#2280246, @dmajor wrote:

This broke Firefox builds too, in one of our helper binaries that uses a wWinMain, although I'm having trouble writing a minimal reproducer for it. Simply making a barebones wWinMain program doesn't hit the error.

If the patch re-lands, please cc me and I'll re-test.

In D87701#2274860, @rnk wrote:

lgtm, thanks.

In D72242#1873017, @smeenai wrote:

I'm not sure why Phabricator is still showing Needs Review, but @rnk's approval should make this count as accepted.

Actually I just noticed it still says 'Needs review' above. I'm not sure what the protocol is. Can I push the patch?

Thanks! I'll push it shortly. Just running a final ninja check-all after updating workspace.

Thanks for taking a look Richard. This patch adds the required value-dependency check and sets type-dependency accordingly.

In D72242#1820908, @rnk wrote:

I think I liked the first version of this patch better. I would say, if the AST after instantiation remains the same as it was before D69950, then the first one seems like the right fix.

Semantic analysis for value dependent conditions is now skipped as per Erich's comment. Patch adds an argument to CheckBooleanCondition to still do the required analysis for noexcept expressions.

Thanks for taking a look Erich!

In D69950#1771340, @mstorsjo wrote:

In D69950#1771133, @eandrews wrote:

In D69950#1770138, @mstorsjo wrote:

This (when reapplied in https://reviews.llvm.org/rG878a24ee244a24c39d1c57e9af2) broke compilation of code that earlier built fine. A reduced example:

namespace glslang {
class TPoolAllocator {
  void operator=(TPoolAllocator);
};
template <class> class a {
  TPoolAllocator *b;
  void c() { allocator = *b; }
  TPoolAllocator allocator;
};
} // namespace glslang

With this patch, some errors in templates are diagnosed earlier (i.e. does not wait till instantiation). Since 'allocator' and 'b' aren't dependent, I think this is a valid diagnosis. GCC throws an error on this code upon instantiation. https://godbolt.org/z/X9Y-Vy

The original non-reduced case does build fine with GCC though (I'm fairly sure). I can try to reduce one later that still builds with GCC but fails with current clang.

In D69950#1770138, @mstorsjo wrote:

This (when reapplied in https://reviews.llvm.org/rG878a24ee244a24c39d1c57e9af2) broke compilation of code that earlier built fine. A reduced example:

namespace glslang {
class TPoolAllocator {
  void operator=(TPoolAllocator);
};
template <class> class a {
  TPoolAllocator *b;
  void c() { allocator = *b; }
  TPoolAllocator allocator;
};
} // namespace glslang

In D69950#1752044, @gribozavr2 wrote:

In D69950#1751894, @eandrews wrote:

In D69950#1751859, @gribozavr2 wrote:

hence not throwing the warning on any platform?

The way I read the buildbot breakage, an existing ClangTidy test passed before and after this change, but broke on Windows. The breakage was that the warnings stopped being produced.

The existing test does not have this warning. I modified the test to add the check for this warning since it is generated on Linux after my patch. It is not generated on Windows because of delayed template parsing.

Is it just that warning that is not generated, or all warnings in that test file?

In D69950#1751859, @gribozavr2 wrote:

hence not throwing the warning on any platform?

The way I read the buildbot breakage, an existing ClangTidy test passed before and after this change, but broke on Windows. The breakage was that the warnings stopped being produced.

The patch changes type dependency of field 'std::string s' and sets it based on field type (i.e. not type-dependent).

Thank you for letting me know. The patch has been reverted.

@rnk @gribozavr2 clang-tidy test "bugprone-string-integer-assignment.cpp" caused a bot fail on windows. From the logs it looks like the 'CHECK-FIXES' I added in that test is failing because the fix isn't applied. I am not sure why. I don't have a windows build set up and I assume a new build is going to take some time. I am not sure what the protocol for this is. Should I go ahead and revert my patch while I investigate the issue on Windows or should I just push a new patch deleting the 'CHECK-FIXES' while I look into the issue?

Thank you for taking a look Reid and Dmitri. There were no fails with check-all.

Thank you for letting me know. I did try reproducing the issue with check-all yesterday but was unable to. I do not think I build with 'clang-tools-extra' project enabled though. I'll take another look today.

In D61027#1627655, @gribozavr wrote:

Sorry, but this change broke ClangTidy tests: http://lab.llvm.org:8011/builders/clang-x86_64-debian-fast/builds/16398. I reverted it.

Thanks for taking a look Reid! I rebased the patch and had a conflict with one of Richard's patches and so ran the lit tests again and noticed there are several new failures. I am taking a look at those now.

ping

I just realized I modified the summary when I uploaded a new patch, but did not add a comment about the changes I made. Based on feedback from the bug report (https://bugs.llvm.org/show_bug.cgi?id=40982), I changed my approach for this crash. This patch changes the type dependency of the member which causes the crash.

I ran the test you provided and it does throw errors without instantiation

I apologize for the confusion. I was working on an incorrect branch earlier, and so abandoned that patch and uploaded a new revision.

Thanks!

Upstreaming it first might be better, especially since the change seems to be trivial. Is this line addition the only change proposed for the benchmark library?

Yes this line is the only change.

Do you want me to submit the patch to the benchmark library? It seems that it would help to speedup the process.

That would be helpful. Thank you!

@kbobyrev I apologize if I was unclear in the comments. I was asking if the changes proposed in the comments are alright with you since they would involve modifying benchmark/CMakelists.txt (instead of llvm/CMakeLists.txt as discussed in mailing list). As Zachary mentioned in comments, _HAS_EXCEPTIONS should be set to 0 only when exceptions are disabled. Since exception handling for benchmarks is handled in benchmark/CMakeLists.txt, I think it makes most sense to add the definition there. I have now uploaded the proposed change for review.

In D52998#1259602, @lebedev.ri wrote:

In D52998#1258962, @eandrews wrote:

Yes. I understand. At the moment, exception handling flags are generated based on BENCHMARK_ENABLE_EXCEPTIONS in utils/benchmark/CMakeLists.txt . However, _HAS_EXCEPTIONS is not defined based on this (code below). The warnings are a result of this mismatch.

if (NOT BENCHMARK_ENABLE_EXCEPTIONS)
    add_cxx_compiler_flag(-EHs-)
    add_cxx_compiler_flag(-EHa-)
  endif()

I think it makes most sense to add definition for _HAS_EXCEPTIONS=0 here as opposed to modifying llvm/CMakeLists.txt.

Then i'd recommend/suggest to submit this upstream first.

Please correct me if I'm wrong. I'm not too familiar with CMake. @kbobyrev Please let me know what you think as well since you had suggested llvm/CMakeLists.txt.

I think it makes sense to include this case just to make the flag more complete. Also, we don't really match GCC behavior for this flag. GCC emits all static constants by default (when O0). When optimized, this flag has no effect on GCC. The negation is used to not emit the constants.

Yes. I understand. At the moment, exception handling flags are generated based on BENCHMARK_ENABLE_EXCEPTIONS in utils/benchmark/CMakeLists.txt . However, _HAS_EXCEPTIONS is not defined based on this (code below). The warnings are a result of this mismatch.

Thanks for the information Zachary.

I do not think defining _HAS_EXCEPTIONS=0 affects C++ exceptions in the application. I thought it only affected the STL. I will verify this and update you.

Alright. Thanks for the review. I will abandon this patch

@lebedev.ri is there a specific reason -Wtautological-unsigned-zero-compare was removed? All the issues I am aware of talks about comparisons with min/max.

In D40925#1206416, @rnk wrote:

lgtm!

Based on Reid's feedback, I changed option to CodeGenOption

This patch fell through the cracks earlier. I apologize. Based on Reid's and Erich's feedback, I am now adding the variable to @llvm.used. Additionally I modified the if statements to ensure only static variables are considered.

I think we can use CodeGenModule::addUsedGlobal for this purpose. I noticed this is what attribute 'used' calls. I need to look into it though.

That makes sense. Is it not possible to implement the required functionality using a flag vs an attribute? In an earlier comment you mentioned adding the global to @llvm.used to prevent GlobalDCE from removing it. Can you not do that when using a flag?

Thanks for the review Reid. Sorry for the delay in my response. I was OOO.

*ping*

No problem! Thanks!

*ping*

Updated patch to set default calling convention for main() in CheckMain()

I've modified the patch to emit a warning for re-declarations only. I also removed the isUsed check.

I've updated the patch based on review comments. The changes include setting section unconditionally for extern variables and emitting a warning if section attribute is specified on a redeclaration.

*ping*

LangRef has been updated and accepted. You can find it here - https://reviews.llvm.org/rL311459

Updated the documentation to explicitly state that section information for a variable can be explicit or inferred.

In D36712#843414, @kparzysz wrote:

The problem is that the mismatch between sections does not have to lead to any undesirable behavior. Whether it does or not depends on a particular case. I think we should be consistent though---if we decide that a mismatch between the section for a declaration and the section for a definition leads to an undefined behavior, then it should be so regardless of whether the sections were given explicitly or inferred.

Updated the patch to include Krzysztof's comment about explicitly stating undefined behavior for section information mismatch in global variable declaration and definition. This should cover the case where section is explicitly specified in definition but not declaration.

In D36712#842477, @kparzysz wrote:

In the cases when the section is explicitly given on a definition, it was likely imposed by something like the "section" attribute in the source. I don't think it's unreasonable to expect that the declarations (in the original source as well as in the generated IR) should carry that information as well. However, since clang has apparently been ignoring that attribute on declarations, it's been generating IR where declarations may not have sections, but the corresponding definitions do.

Corrected spelling error.

I updated the documentation to include Simon's comments. I wasn't sure whether to include the line about section information being retained in LLVM IR, since Eli mentioned it could vary for different front-ends. I have included it in this revision.

As per review comments, removed extra RUNS in lit test.

Regcall-specific checks for Lin64 now occur only if CXXABI returns false. An existing test has also been modified to verify behavior with non trivial destructors.

As per revision comments, I moved the condition for extended precision floating type to isX86VectorTypeForVectorCall. This update will now alter behavior for complex long double type under vectorcall calling convention as well. Returns/parameters will be passed in memory.

In D35259#805415, @erichkeane wrote:

In D35259#805409, @rnk wrote:

In D35259#805284, @erichkeane wrote:

Oren discovered this miss to the original implementation. I'd reviewed this internally quite a bit.

The reason for the Win32ABI test is that MSVC 'long double' is actually small enough for SSE registers in this case. I DO now (looking again, sorry Elizabeth) wonder if there is a better way to exclude extended-length LongDouble type? Could we us the 'length' of it instead? @rnk : opinion?

Yeah, you can ask clang::TargetInfo for the format of most basic FP types. The code to do that looks like:

&TI.getLongDoubleFormat() == &llvm::APFloat::x87DoubleExtended()

Any reason you can't just add that condition to isX86VectorTypeForVectorCall? I assume we don't want to pass x86_fp80s in SSE registers for vectorcall either, right? That would eliminate the need for the isRegCallReturnableHA helper and the IsWin32StructABI parameter, which is a poorly named variable.

It actually WOULD make sense to apply this to vectorcall as well, wouldn't it? I presumed we didnt want to change its behavior, however vectorcall is MSVC-only (other than us), so the long-double issue isn't a thing over there.@eandrews?