qt 4.8.6 failed to build on 10.6

[Sean1708]

Upgrading qt failed, most likely with the error '_mm_alignr_epi8 not declared in this scope'.

Gist (as per the troubleshooting checklist) is here: https://gist.github.com/Sean1708/ea4873c3899bcd4606bc

[MikeMcQuaid]

Try and make this change to the formula file and see if that fixes it:

-    args << "-no-3dnow" if superenv?
+    args << '-no-3dnow' << '-no-ssse3' if superenv?

[ryanshaw]

@MikeMcQuaid that change made it work for me here.

[MikeMcQuaid]

@dakcarto ☝️

Perhaps we should enable it just for 10.6 (although I've a feeling we perhaps did that before).

[a5sk4s]

worked for me as well - thanks for such a quick workaround

[dakcarto]

@MikeMcQuaid wrote:

Perhaps we should enable it just for 10.6

Sure. I don't know enough about SSE (or what's supposedly fixed in the latest 4.8.6, but that doesn't work for 10.6) to really have a valid opinion here. Haven't compiled under 10.6 yet.

@ryanshaw @a5sk4s
Is everyone here saying that it is now building specifically for 10.6 (but wasn't before), or did anyone also have trouble with other OS versions?

[a5sk4s]

it was on 10.6.8 for me

[ryanshaw]

I misspoke earlier, I'm on 10.9.2 but also could not compile qt 4.8.6. That tweak mentioned by @MikeMcQuaid fixed it for me. Apologies for the confusion.

[dakcarto]

Here's the Qt source commit: https://qt.gitorious.org/qt/qt/commit/1c0a1ebba12c9c1fa02117b9a017c4ed3f7cd7c3

Seems with the cascading, that adding -no-ssse3 should keep any of the 4.x code blocks from being executed in <qt src>/src/corelib/tools/qstring.cpp's mergeQuestionMarks(). So I wonder why I had the opposite error in 10.7.5?

[dakcarto]

Ok, this patch fixed the compiling issue I had on 10.7.5.

Odd, though, because the same qstring.cpp existed in Qt 4.8.5. Something more going on with the SSE defines.

[andreasgerstmayr]

the commit 37c8e4b breaks the build on OS X 10.9.2:

tools/qstring.cpp:3603:22: error: use of undeclared identifier '_SIDD_UWORD_OPS'
    const int mode = _SIDD_UWORD_OPS | _SIDD_CMP_RANGES | _SIDD_UNIT_MASK;
                     ^
tools/qstring.cpp:3603:40: error: use of undeclared identifier '_SIDD_CMP_RANGES'
    const int mode = _SIDD_UWORD_OPS | _SIDD_CMP_RANGES | _SIDD_UNIT_MASK;
                                       ^
tools/qstring.cpp:3603:59: error: use of undeclared identifier '_SIDD_UNIT_MASK'
    const int mode = _SIDD_UWORD_OPS | _SIDD_CMP_RANGES | _SIDD_UNIT_MASK;
                                                          ^
tools/qstring.cpp:3605:34: error: use of undeclared identifier '_mm_cmpestrm'
    const __m128i offLimitMask = _mm_cmpestrm(rangeMatch, 2, chunk, 8, mode);
                                 ^
/private/tmp/qt-6VoF/qt-everywhere-opensource-src-4.8.6/bin/moc -DQT_SHARED -DQT_BUILD_CORE_LIB -DQT_NO_USING_NAMESPACE -DQT_NO_CAST_TO_ASCII -DQT_ASCII_CAST_WARNINGS -DQT3_SUPPORT -DQT_MOC_COMPAT -DQT_USE_QSTRINGBUILDER -DQLIBRARYINFO_EPOCROOT -DHB_EXPORT=Q_CORE_EXPORT -DQT_NO_DEBUG_PLUGIN_CHECK -DQT_NO_DEBUG -DQT_HAVE_MMX -DQT_HAVE_SSE -DQT_HAVE_MMXEXT -DQT_HAVE_SSE2 -DQT_HAVE_SSE3 -D_LARGEFILE64_SOURCE -D_LARGEFILE_SOURCE -I../../mkspecs/unsupported/macx-clang-libc++ -I. -I../../include -I../../include/QtCore -I.rcc/release-shared -Iglobal -I../../tools/shared -I../3rdparty/harfbuzz/src -I../3rdparty/md5 -I../3rdparty/md4 -I.moc/release-shared -D__APPLE__ -D__GNUC__ io/qbuffer.h -o .moc/release-shared/moc_qbuffer.cpp
tools/qstring.cpp:3608:13: error: use of undeclared identifier '_mm_blendv_epi8'
    chunk = _mm_blendv_epi8(chunk, questionMark, offLimitMask);
            ^
clang++ -c -pipe -stdlib=libc++ -mmacosx-version-min=10.7 -I/opt/X11/include -I/usr/local/opt/d-bus/lib/dbus-1.0/include -I/usr/local/opt/d-bus/include/dbus-1.0 -O2 -arch x86_64 -fvisibility=hidden -fvisibility-inlines-hidden -Wall -W -fPIC -DQT_SHARED -DQT_BUILD_CORE_LIB -DQT_NO_USING_NAMESPACE -DQT_NO_CAST_TO_ASCII -DQT_ASCII_CAST_WARNINGS -DQT3_SUPPORT -DQT_MOC_COMPAT -DQT_USE_QSTRINGBUILDER -DQLIBRARYINFO_EPOCROOT -DHB_EXPORT=Q_CORE_EXPORT -DQT_NO_DEBUG_PLUGIN_CHECK -DQT_NO_DEBUG -DQT_HAVE_MMX -DQT_HAVE_SSE -DQT_HAVE_MMXEXT -DQT_HAVE_SSE2 -DQT_HAVE_SSE3 -D_LARGEFILE64_SOURCE -D_LARGEFILE_SOURCE -I../../mkspecs/unsupported/macx-clang-libc++ -I. -I../../include -I../../include/QtCore -I.rcc/release-shared -Iglobal -I../../tools/shared -I../3rdparty/harfbuzz/src -I../3rdparty/md5 -I../3rdparty/md4 -I.moc/release-shared -o .obj/release-shared/qdatastream.o io/qdatastream.cpp
5 errors generated.
make[2]: *** [.obj/release-shared/qstring.o] Error 1
make[2]: *** Waiting for unfinished jobs....
make[1]: *** [release] Error 2
make: *** [sub-corelib-make_default-ordered] Error 2

when I revert the changes of this commit the build works.

[dakcarto]

@andihit Try this formula with included patch for that issue (noted above).

However, did not have the issue here on 10.9.2 with current formula.

HOMEBREW_VERSION: 0.9.5
ORIGIN: https://github.com/Homebrew/homebrew.git
HEAD: 9362781dd8822650b7e3714cff5d1aa004f116d3
HOMEBREW_PREFIX: /usr/local
HOMEBREW_CELLAR: /usr/local/Cellar
CPU: 8-core 64-bit haswell
OS X: 10.9.2-x86_64
Xcode: 5.1
CLT: 5.1.0.0.1.1396320587
GCC-4.2: build 5666
Clang: 5.1 build 503
X11: 2.7.5 => /opt/X11
System Ruby: 1.8.7-358
Perl: /usr/bin/perl
Python: /usr/local/bin/python => /usr/local/Cellar/python/2.7.6_1/Frameworks/Python.framework/Versions/2.7/bin/python2.7
Ruby: /usr/bin/ruby

[MikeMcQuaid]

@dakcarto I'm guessing it'll depend on your hardware whether you have this problem or not. Could you submit this patch to Qt's Gerrit: https://codereview.qt-project.org/

[dakcarto]

Qt Gerrit patch set uploaded for review and bug report added.

[MikeMcQuaid]

Perfect, thanks @dakcarto. Happy to apply that patch in a PR too now it's reported. You're a model open-source citizen 👼

[Sean1708]

Thanks guys, works perfectly now.

[dakcarto]

@Sean1708 Thanks for the feedback. What worked for you? The current formula in main tap, or did you specifically try the one I noted here?

[dakcarto]

@MikeMcQuaid Do you mind taking some time to review the Qt bug report?

If there is a flaw in my logic, or superenv is doing something I am unaware of, I'd appreciate knowing. Having some difficulty getting the reviewer (who also created the issue with their commit) to understand the situation.

Could be that I just don't know enough about how SS(S)E optimization is triggered at the compiler level. Otherwise, I think my argument/patch in the report is pretty sound.

[GuillaumeDIDIER]

The commit has broken qt for me on 10.9.2
Reverting the changes worked.
It was broken in string.cpp ~3600
Reverting the change made it work.
Having investigated the source code I think that the patch in the formula seems to be the right thing to do.
Also has someone looked into 10.6 breakage to see why qt+superenv gets messed up.

[MikeMcQuaid]

@GuillaumeDIDIER Can you try @dakcarto's patch.

[MikeMcQuaid]

@dakcarto Think you're probably fighting a losing battle there. Can we just specify more configure arguments when using superenv to e.g. disable SSE4.2?

[MikeMcQuaid]

Can anyone still having issues here (@justinclift and @robsonpeixoto too) confirm if this fixes them for you:

-    args << "-no-3dnow" << "-no-ssse3" if superenv?
+    args << "-no-3dnow" << "-no-ssse3" << "-no-sse4.1" << "-no-sse4.2" if superenv?

[MikeMcQuaid]

I suspect given the resistance of getting this patch merged upstream we may need to remove the no-X superenv args, fix the underlying issue and submit a patch upstream. This should be possible given it's working in Qt5.

[robsonpeixoto]

@MikeMcQuaid same problem

tools/qstring.cpp:3603:22: error: use of undeclared identifier '_SIDD_UWORD_OPS'
    const int mode = _SIDD_UWORD_OPS | _SIDD_CMP_RANGES | _SIDD_UNIT_MASK;
                     ^
tools/qstring.cpp:3603:40: error: use of undeclared identifier '_SIDD_CMP_RANGES'
    const int mode = _SIDD_UWORD_OPS | _SIDD_CMP_RANGES | _SIDD_UNIT_MASK;
                                       ^
tools/qstring.cpp:3603:59: error: use of undeclared identifier '_SIDD_UNIT_MASK'
    const int mode = _SIDD_UWORD_OPS | _SIDD_CMP_RANGES | _SIDD_UNIT_MASK;
                                                          ^
tools/qstring.cpp:3605:34: error: use of undeclared identifier '_mm_cmpestrm'
    const __m128i offLimitMask = _mm_cmpestrm(rangeMatch, 2, chunk, 8, mode);
                                 ^
tools/qstring.cpp:3608:13: error: use of undeclared identifier '_mm_blendv_epi8'
    chunk = _mm_blendv_epi8(chunk, questionMark, offLimitMask);
            ^
5 errors generated.

[MikeMcQuaid]

I guess we should perhaps just apply @dakcarto's patch in the mean time until we work out a better solution.

[justinclift]

Things still barf on OSX 10.7.5 when using this line:

args << "-no-3dnow" << "-no-ssse3" << "-no-sse4.1" << "-no-sse4.2" if superenv?

@dakcarto's patch worked though.

[dakcarto]

@MikeMcQuaid Regarding the addition of -no-sse4.x flags, due to the cascading-up changes between 4.8.5 and .6 setting any of the lower flags to no automatically sets the higher ones to no as well. Although the compiler may support SSE4.x (and be compiling the rest of Qt that way), having any -no-X effectively negates including the appropriate headers for the SSE-specific code in Qt.

I really have no clue as to the logic of why the patch is not being applied upstream. I can't think of anything to add to the upstream bug report, or any other way to fix it. Removing all -no-ss(s)e flags and trying to patch the specific SSSE3 issue for some compilers may be the only alternative (as you noted).

[MikeMcQuaid]

People may want to test out #28985 as it seems to fix all these issues.

[dakcarto]

@MikeMcQuaid That PR fixes the build issues here on 10.7.5. Thanks!

[robsonpeixoto]

@MikeMcQuaid That PR fixes the build issues here on 10.9.2. Thanks!

[Sean1708]

@dakcarto Sorry I didn't get back to you earlier, the main tap worked for me on the day I posted that comment.

[jacknagel]

Fixes for this just went into master, let us know if there are any further issues.

[mattmartini]

@jacknagel On 10.8.5 I was seeing the "error: use of undeclared identifier '_mm_blendv_epi8'" issue. After pulling the latest master the build worked.

[justinclift]

Yep, worked here too. qt 4.8.6 compiled fine.

[MikeMcQuaid]

@jacknagel Thanks for this.