Fix Mingw cross-compilation. #2329
Conversation
|
+cc @brondani |
src/disk_interface_test.cc
Outdated
| @@ -171,7 +171,7 @@ TEST_F(DiskInterfaceTest, StatCache) { | |||
| EXPECT_GT(disk_.Stat("subdir/subsubdir", &err), 1); | |||
| EXPECT_EQ("", err); | |||
|
|
|||
| #ifndef _MSC_VER // TODO: Investigate why. Also see https://github.com/ninja-build/ninja/pull/1423 | |||
| #if 0 // TODO: Investigate why. Also see https://github.com/ninja-build/ninja/pull/1423 | |||
There was a problem hiding this comment.
should be #ifndef _WIN32 then, otherwise we would disable it for Linux, too.
There was a problem hiding this comment.
Ok, I'll do it, but note that there is already an #ifndef _WIN32 just before this test declaration, because this is (currently) the only platform with a StatCache.
There was a problem hiding this comment.
Ah I see! Then we probably shouldn't touch this line at all, should we? The test succeeded on AppVeyor for MinGW: https://ci.appveyor.com/project/nico/ninja/builds/47865014/job/eg5b0uic6jgrrida
There was a problem hiding this comment.
The test definitely fails when building with my Debian mingw installation, and running on a Windows 10 machine though, which is why I had to disable it there too. I have no idea why.
[3/398] MSVCHelperTest.NoReadOfStderrto stderr
[207/398] DiskInterfaceTest.StatCache
*** Failure in /work2/github-digit/ninja/src/disk_interface_test.cc:175
disk_.Stat("subdir", &err) == disk_.Stat("subdir/.", &err)
*** Failure in /work2/github-digit/ninja/src/disk_interface_test.cc:178
disk_.Stat("subdir", &err) == disk_.Stat("subdir/subsubdir/..", &err)
[398/398] BuildLogRecompactTest.Recompact
I suspect the issue is with the MSVCRT.DLL runtime being used, not the compiler that builds the code.
There was a problem hiding this comment.
For the record, I am building with CMake, not the ./configure.py script which does not support cross-compilation.
There was a problem hiding this comment.
I see. Still it runs fine in some cases and one of them is the CI. That's why I would like to keep it enabled.
There was a problem hiding this comment.
All right, we'll keep a local patch for our own CI then. I have removed the change to disk_interface_test.cc here.
digit-google
force-pushed
the
fix-mingw-compilation
branch
from
0769f5a
to
00aa84f
Compare
The previous pull request broke compilation with the Mingw toolchain, which does not provide <atlcore.h>. Moreover, this massive header was only used to expand a macro that generates a static volatile variable under the hood (which is not necessary here). Replace it with with a simpler and more portable code path to fix the build.
digit-google
force-pushed
the
fix-mingw-compilation
branch
from
00aa84f
to
eddafdd
Compare
|
Have you seen this warning? |
|
I was not aware of this warning. I have pushed digit-google@f6e83fa as a work-around. Searching online for an explanation only finds a few pages that state that this is a GCC 8+ bug, without proof however. |
The previous pull request broke compilation with the Mingw toolchain, which does not provide <atlcore.h>. Moreover, this massive header was only used to expand a macro that generates a static volatile variable under the hood (which is not necessary here).
Replace it with with a simpler and more portable code path to fix the build.
Also, ensure that a broken DiskInterfaceTest, which was only disabled when using the Microsoft compiler, is also disabled for Mingw (note that the whole test code is inside a #ifdef _WIN32 .. #endif block).
After this patch, a
ninja_test.exegenerated on Linux with the Mingw toolchain runs and succeeds properly on a Windows machine.