[libcxx][tests] Introduce 32-bit feature and use it for stringstream gcount test

[azat]

Summary

This will avoid hardcoding all unsupported targets, since even after one
more follow up fix [1], there is one more failure.

[1]: https://reviews.llvm.org/D150886

Plus, if you want to run it locally on some target that CI does not
covers, it could also false-positively fail, which is not good.

[azat]

Looks like CI failures is unrelated?

[mgorny]

Thanks. I can confirm that it works for me.

[Mordante]

Thanks for the followup patch!

libcxx/utils/libcxx/test/features.py

387

I'm not convinced we should rely on __SIZEOF_SIZE_T__ being available at least we should test here it's defined. (Alternatively we do a runtime test, for an example see has-unix-headers.)

[azat]

Fallback to 0 if SIZEOF_SIZE_T is not defined

[azat]

Rename feature to 32-bit-architecture

[ldionne]

libcxx/utils/libcxx/test/features.py

387

I would try using a runtime test that checks the result of sizeof(void*) == 4.

[azat]

libcxx/utils/libcxx/test/features.py

387

I'm not convinced we should rely on SIZEOF_SIZE_T being available at least we should test here it's defined. (Alternatively we do a runtime test, for an example see has-unix-headers.)

I guess a simple fallback to 0 will work, rebased.

[azat]

libcxx/utils/libcxx/test/features.py

387

@ldionne Ok, I could use this, but can you elaborate why int(compilerMacros(cfg).get('__SIZEOF_SIZE_T__', 0)) == 4 is not good, just to understand is there any difference? (since actually that test relies on the size_t size, but this feature flag is about 32-bit architecture so I'm fine both ways)

[mgorny]

libcxx/test/std/input.output/string.streams/stringstream.members/gcount.pass.cpp

9

Well, I don't want to be picky but if you want to use a long name anyway, than perhaps something like "32-bit-pointer" would be more precise. After all, I dare say n32 ABI is also using 32-bit pointers but it's works only on a 64-bit CPU architecture.

[ldionne]

libcxx/utils/libcxx/test/features.py

387

Using sizeof(void*) would work regardless of the compiler though.

[vvereschaka]

works for me, thanks.

[azat]

Use sizeof(void*)

[azat]

libcxx/utils/libcxx/test/features.py

387

I'm not convinced we should rely on SIZEOF_SIZE_T being available at least we should test here it's defined. (Alternatively we do a runtime test, for an example see has-unix-headers.)

@Mordante BTW it is defined in clang/lib/Frontend/InitPreprocessor.cpp, so it should always be defined.

[philnik]

libcxx/utils/libcxx/test/features.py

387

The tests aren't just for libc++. The test suite is designed to be as portable as possible, so __SIZEOF_SIZE_T__ might in fact not be defined. e.g. does MSVC define it? They actually use our test suite.

[azat]

libcxx/utils/libcxx/test/features.py

387

The tests aren't just for libc++. The test suite is designed to be as portable as possible, so __SIZEOF_SIZE_T__ might in fact not be defined. e.g. does MSVC define it?

They don't - https://godbolt.org/z/d4d5xG1ET

They actually use our test suite.

Got it thanks!

Anyway, now sizeof(void*) is used.

[Mordante]

I'll land this on your behalf.

libcxx/utils/libcxx/test/features.py

387

That @philnik said was my reason too. Thanks for fixing it.

libcxx/test/std/input.output/string.streams/stringstream.members/gcount.pass.cpp

 //===----------------------------------------------------------------------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.
 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 //
 //===----------------------------------------------------------------------===//
 
-// All 32 bit arches that CI has:
-//
-// UNSUPPORTED: target=powerpc-ibm-aix
-// UNSUPPORTED: target={{.*}}armv7l{{.*}}
-// UNSUPPORTED: target={{.*}}armv8l{{.*}}
-// UNSUPPORTED: target=i686-w64-windows-gnu
+// UNSUPPORTED: 32-bit-pointer
 
 // Test that tellp() does not break the stringstream after INT_MAX, due to use
 // of pbump() that accept int.
 
 #include <cassert>
 #include <climits>
 #include <sstream>
 #include <string>

libcxx/utils/libcxx/test/features.py

DEFAULT_FEATURES += [
   # Tests that require support for std::shared_mutex and std::shared_timed_mutex in the built library
   Feature(name='availability-shared_mutex-missing',
     when=lambda cfg: BooleanExpression.evaluate('stdlib=apple-libc++ && target={{.+}}-apple-macosx10.{{(9|10|11)(.0)?}}', cfg.available_features)),
 
   # Tests that require support for aligned allocation in the built library. This is about `operator new(..., std::align_val_t, ...)` specifically,
   # not other forms of aligned allocation.
   Feature(name='availability-aligned_allocation-missing',
     when=lambda cfg: BooleanExpression.evaluate('stdlib=apple-libc++ && target={{.+}}-apple-macosx10.{{(9|10|11|12)(.0)?}}', cfg.available_features)),
+
+  # Tests that require 64-bit architecture
+  Feature(name='32-bit-pointer',
+          when=lambda cfg: sourceBuilds(cfg, """
+            int main(int, char**) {
+              static_assert(sizeof(void *) == 4);
+            }
+          """)),
 ]