Issues building MINGW packages with clang #7589
Replies: 22 comments 116 replies
-
|
zlib: won't build shared library. See #7587 |
Beta Was this translation helpful? Give feedback.
-
|
pkg-config: couple of blow-ups in embedded glib. Stupid glib. Would have done pkgconf but it makedepends on cmake and I'm nowhere near getting its depends built yet. |
Beta Was this translation helpful? Give feedback.
-
|
Stupid hack that got it building: diff -ur glib/glib/gbacktrace.h glib/glib/gbacktrace.h
--- glib/glib/gbacktrace.h 2016-04-11 14:39:26.000000000 -0700
+++ glib/glib/gbacktrace.h 2021-01-03 21:45:16.025393700 -0800
@@ -58,7 +58,7 @@
#elif defined (__alpha__) && !defined(__osf__) && defined (__GNUC__) && __GNUC__ >= 2
# define G_BREAKPOINT() G_STMT_START{ __asm__ __volatile__ ("bpt"); }G_STMT_END
#else /* !__i386__ && !__alpha__ */
-# define G_BREAKPOINT() G_STMT_START{ raise (SIGTRAP); }G_STMT_END
+# define G_BREAKPOINT() G_STMT_START{ __builtin_trap(); }G_STMT_END
#endif /* __i386__ */
G_END_DECLS
diff -ur glib/glib/valgrind.h glib/glib/valgrind.h
--- glib/glib/valgrind.h 2016-04-11 14:39:26.000000000 -0700
+++ glib/glib/valgrind.h 2021-01-03 21:45:16.009865100 -0800
@@ -124,7 +124,7 @@
# define PLAT_x86_darwin 1
#elif defined(__APPLE__) && defined(__x86_64__)
# define PLAT_amd64_darwin 1
-#elif defined(__MINGW32__) || defined(__CYGWIN32__) \
+#elif defined(__CYGWIN32__) \
|| (defined(_WIN32) && defined(_M_IX86))
# define PLAT_x86_win32 1
#elif defined(__linux__) && defined(__i386__) |
Beta Was this translation helpful? Give feedback.
-
|
Oops, this was arm (aarch64) specific. Was going to keep those separate. Oh well |
Beta Was this translation helpful? Give feedback.
-
|
I had no problem building pkg-config for x86_64. |
Beta Was this translation helpful? Give feedback.
-
|
nor on i686 - just aarch64 so far (haven't tried armv7) |
Beta Was this translation helpful? Give feedback.
-
|
gmp: I tried |
Beta Was this translation helpful? Give feedback.
-
|
Same thing on libunistring, but |
Beta Was this translation helpful? Give feedback.
-
That solved gmp |
Beta Was this translation helpful? Give feedback.
-
|
Yeah, common issue with autotools (libtool to be specific), as you noticed autoreconf before applying patches solves it. |
Beta Was this translation helpful? Give feedback.
-
|
ncurses: |
Beta Was this translation helpful? Give feedback.
-
|
There was space or something with dashes. I think I was able to solve it. |
Beta Was this translation helpful? Give feedback.
-
|
Circular dependencies galore... largely involving cmake, probably because that was a 'goal' package to hopefully unlock more packages. |
Beta Was this translation helpful? Give feedback.
-
|
The easiest way is to disable some of cmake features to get minimal working build. |
Beta Was this translation helpful? Give feedback.
-
|
After actually looking at cmake (rather than just its PKGBUILD) it seems it bundles dependencies by default, so I'd probably have more success just using those rather than opting for system libraries. |
Beta Was this translation helpful? Give feedback.
-
|
mpdecimal: |
Beta Was this translation helpful? Give feedback.
-
|
This hack fixed mpdecimal: diff --git a/mingw-w64-mpdecimal/PKGBUILD b/mingw-w64-mpdecimal/PKGBUILD
index e0df453..c259e78 100644
--- a/mingw-w64-mpdecimal/PKGBUILD
+++ b/mingw-w64-mpdecimal/PKGBUILD
@@ -19,6 +19,7 @@ sha256sums=('15417edc8e12a57d1d9d75fa7e3f22b158a3b98f44db9d694
cfd2acde8dfa0ca'
prepare() {
cd $srcdir/${_realname}-${pkgver}
patch -Nbp1 -i ${srcdir}/mpdecimal-2.5.0-msys-mingw.patch
+ sed -i -e 's/ -fPIC//g' libmpdec/Makefile.in libmpdec++/Makefile.in
autoreconf -fiv
}
|
Beta Was this translation helpful? Give feedback.
-
|
I've got an idea how to make Clang ignore fpic without listing it in help but I don't know yet how much effort it will take. |
Beta Was this translation helpful? Give feedback.
-
|
libuv: This looks familiar but I don't recall what if anything the outcome was... https://gitter.im/msys2/msys2?at=5faf347fb4283c208a63557a |
Beta Was this translation helpful? Give feedback.
-
|
Every binary that links libclang or libllvm may fail, maybe libc++ or ucrt helps here somehow but |
Beta Was this translation helpful? Give feedback.
-
|
maybe when they had to be statically linked? I don't think it would be as big of a mmap problem for tools to link to DLLs (or more precisely to import libraries) |
Beta Was this translation helpful? Give feedback.
-
|
Few times in the past it crashed for me when creating libclang Dll so it's not always due to static libs. |
Beta Was this translation helpful? Give feedback.
-
|
these are leftover definitions from intrin.h and can be safely removed. |
Beta Was this translation helpful? Give feedback.
-
|
I tried again without a workaround the other day, and the issue was still there. Tracked at msys2/CLANG-packages#6 |
Beta Was this translation helpful? Give feedback.
-
|
Opening issues in Clang repo would be much easier to track, we could use labels for arch specific issues, etc. |
Beta Was this translation helpful? Give feedback.
-
|
If you prefer... I didn't go that route because the packages in question aren't actually in that repository. |
Beta Was this translation helpful? Give feedback.
-
|
done, and will do so for anything I find moving forward. I decided to 'start over' fresh after changing the default win32_winnt value just to be safe, and my aarch64 machine is currently busy seeing just how long it will take to bootstrap the aarch64 clang on it. |
Beta Was this translation helpful? Give feedback.
-
|
well i can actually rebuild clang with clang itself now :) so it is definatly an improvement. |
Beta Was this translation helpful? Give feedback.
-
|
tesseract-ocr fails to link to libomp.
|
Beta Was this translation helpful? Give feedback.
-
|
heh will be fun seing how codelite likes to be compiled on clang now :) |
Beta Was this translation helpful? Give feedback.
-
|
xavs is already crossed out on my list, I think somebody already fixed it |
Beta Was this translation helpful? Give feedback.
-
|
oki then :) |
Beta Was this translation helpful? Give feedback.
-
|
mingw-w64-db nameclash between atomic_init and clangs libc++ function of the same name. the partial fix is renaming all occurences of atomic_init to atomic_init_db atleast that was what they had to do on APPLE where i got the fix from. The store define could also be renamed to dbstore but i had a hard time finding where it is actually called because it only seems to be one place in tcl_compat.c while it is defined in a billion places for every architecture so urgh... |
Beta Was this translation helpful? Give feedback.
-
|
https://community.oracle.com/tech/developers/discussion/3952592/bug-build-failure-on-linux-with-clang-libc-atomic-issues |
Beta Was this translation helpful? Give feedback.
-
|
time for an update :) |
Beta Was this translation helpful? Give feedback.
-
|
maybe this will also take care of tesseract-ocr ? -fopenmp=libomp since it looks like clang default links to libgomp if using just -fopenmp. |
Beta Was this translation helpful? Give feedback.
-
|
yes it works :) no need for it in LIBS anymore. |
Beta Was this translation helpful? Give feedback.
-
|
This can also be used in the other packages that fail i would reckon ?. |
Beta Was this translation helpful? Give feedback.
-
|
drmingw this one is partially fixed, so feel free to build on my work :) it does build with clang* now but i had to make some rather nasty changes to get it going. |
Beta Was this translation helpful? Give feedback.
-
|
fixed appstream-glib linker script fixed via lld's --Map flag. |
Beta Was this translation helpful? Give feedback.
-
|
still a work in progress but i might have flang ready for the the clang packages. |
Beta Was this translation helpful? Give feedback.
-
|
its a 3 stage bootstrap, i allready have the bootstrap compiler ready using our patches, next i need to build the flang compiler itself which is an out of source compile, lastly i need to build the flang runtimes. |
Beta Was this translation helpful? Give feedback.
-
|
It uses LLVM 10, so might require some of the patches we were using back then. I have no idea if it will produce binaries compatible with our LLVM 12 (probably 13 within 2 months). |
Beta Was this translation helpful? Give feedback.
-
|
look at the branches theres also a 12 branch :) im using that. |
Beta Was this translation helpful? Give feedback.
-
|
our 12 patches apply cleanly to it though i had to use a later patch from martin for the flang code, moving sys/stat.h up in one of the headers. |
Beta Was this translation helpful? Give feedback.
-
|
binaries are compatible with our own, though im still working on producing the flang compiler and runtime itself the classic flang build is pretty much the same code we use besides the added interface for the flang compiler. I do have a PKGBUILD for that part ready if someone wants to experiment with building the flang compiler itself. |
Beta Was this translation helpful? Give feedback.
-
|
P.s kirigami-qt5 is unbuildable with clang :S |
Beta Was this translation helpful? Give feedback.
-
|
probably for the best, i cant even wrap my head around this xD. |
Beta Was this translation helpful? Give feedback.
-
|
D'oh. Go to enter a bug with LLVM, and it does a little search for you on your summary. Found https://bugs.llvm.org/show_bug.cgi?id=38109 |
Beta Was this translation helpful? Give feedback.
-
|
The best workaround for the short term if you just want to get it to build is probably your first suggestion: take the export off the class and put it on the static in the class definition. But you will get warnings advising you to
So here's a patch that does that --- kirigami2-5.83.0/src/libkirigami/platformtheme.h.orig 2021-06-05 02:20:11.000000000 -0700
+++ kirigami2-5.83.0/src/libkirigami/platformtheme.h 2021-08-17 23:21:19.466150200 -0700
@@ -352,7 +352,7 @@
// of these events.
template<typename T>
-class KIRIGAMI2_EXPORT PropertyChangedEvent : public QEvent
+class PropertyChangedEvent : public QEvent
{
public:
PropertyChangedEvent(PlatformTheme *theme, const T &previous, const T ¤t)
@@ -367,7 +367,7 @@
T oldValue;
T newValue;
- static QEvent::Type type;
+ static KIRIGAMI2_EXPORT QEvent::Type type;
};
using DataChangedEvent = PropertyChangedEvent<std::shared_ptr<PlatformThemeData>>;
@@ -376,6 +376,12 @@
using ColorChangedEvent = PropertyChangedEvent<QColor>;
using FontChangedEvent = PropertyChangedEvent<QFont>;
+template<> KIRIGAMI2_EXPORT QEvent::Type DataChangedEvent::type;
+template<> KIRIGAMI2_EXPORT QEvent::Type ColorSetChangedEvent::type;
+template<> KIRIGAMI2_EXPORT QEvent::Type ColorGroupChangedEvent::type;
+template<> KIRIGAMI2_EXPORT QEvent::Type ColorChangedEvent::type;
+template<> KIRIGAMI2_EXPORT QEvent::Type FontChangedEvent::type;
+
}
} // namespace KirigamiHowever, this results in different exports than GCC made with the original code, lacking the destructors, typeinfos, and vtables for the class template instantiations. |
Beta Was this translation helpful? Give feedback.
-
|
well it seems to work with plasma... guess only the KDE folks would know if something gets broken by it. |
Beta Was this translation helpful? Give feedback.
-
|
patched in #9530 - further investigation revealed that GCC exports for those classes did not change with that patch, and that no dependent packages import those symbols (or even that DLL). |
Beta Was this translation helpful? Give feedback.
-
|
hmm another one was qtwebkit, had to remove -fno-keep-inline-dll-exports since the clang linker does not support it. |
Beta Was this translation helpful? Give feedback.
-
|
crossing fingers. |
Beta Was this translation helpful? Give feedback.
-
|
Beta Was this translation helpful? Give feedback.
-
|
Shebang is ignored when launching script directly (like |
Beta Was this translation helpful? Give feedback.
-
|
#9419 - clangs succeeded. gcc is still going. UPDATE all succeeded building now. |
Beta Was this translation helpful? Give feedback.
-
|
well found out why it was working here, even though it specifies python in the python scripts it was calling msys2's python2.7.exe, lucky me but in the long run it is probably better now that it works with python3. |
Beta Was this translation helpful? Give feedback.
-
|
tulip: This does look odd... the inner struct is not templatized so I don't know why they are trying to explicitly instantiate it (rather than the outer class). |
Beta Was this translation helpful? Give feedback.
-
|
Tulip is being one of those packages where you fix one thing and something else turns up. After the template instantiation, there was explicit linking to |
Beta Was this translation helpful? Give feedback.
-
|
hmm ill see if i can cook up something :/ |
Beta Was this translation helpful? Give feedback.
-
|
I put what I got so far into #9437. Good luck 😁 |
Beta Was this translation helpful? Give feedback.
-
|
oh boy... guess ill have my work cut out xD |
Beta Was this translation helpful? Give feedback.
-
|
ruby also seems to have some problems still, funny though the 32 bit version works but the 64 bit version cannot install gems (or rather it can but only a few like the deprecated ruby-hpricot), guess it is up to the maintainer to fix it properly for clang ucrt. |
Beta Was this translation helpful? Give feedback.
-
|
clang32 also worked in autobuild. Your magical build machine is now jinxed 😉 |
Beta Was this translation helpful? Give feedback.
-
|
heh damn magics newer work as intended xD |
Beta Was this translation helpful? Give feedback.
-
|
well here is something odd, seems it depends on some other gem because after installing all other gems it worked 🎃 |
Beta Was this translation helpful? Give feedback.
-
|
texlive also works fine with clang32 so it can be enabled for this arch :) |
Beta Was this translation helpful? Give feedback.
-
|
tesseract-ocr need to remove -Wl,--as-needed from configure.ac and makefile.am for clang build, works fine without it for clang but im not sure what the repercussions would be for mingw* gcc. |
Beta Was this translation helpful? Give feedback.
-
|
040-non-static-declaration.zip |
Beta Was this translation helpful? Give feedback.
-
|
intel-tbb fixored for clang :) |
Beta Was this translation helpful? Give feedback.
-
|
just need to get freeimage going and then we might be able to build ogre3d with clang as well. |
Beta Was this translation helpful? Give feedback.
-
|
allmost got inkscape going, small mistake in there code omitting unistd.h from the autosave code. and then there is this ?!? |
Beta Was this translation helpful? Give feedback.
-
|
btw one question, atomic builtins do not seem to get built for standalone clang which leads to some odd build errors with x86_64, example vkd3d-proton cannot be built by the standalone x86_64 clang because of this. amx extensions will fail. |
Beta Was this translation helpful? Give feedback.
-
Or not, seems to already be fixed in |
Beta Was this translation helpful? Give feedback.
-
|
had a hunch it might be because those needed to be inlined great :) |
Beta Was this translation helpful? Give feedback.
-
|
still broken in clang-13.0 |
Beta Was this translation helpful? Give feedback.
-
|
It was commited after LLVM 13 branching. |
Beta Was this translation helpful? Give feedback.
-
|
ah that sucks :/ maybe provide a patch untill then ? |
Beta Was this translation helpful? Give feedback.
-
I don't know whether this repo will end up being used or if variants will end up being copied to CLANG-packages, so I don't want to spam issues for anything and everything as I encounter it. It seems like the threaded structure of 'discussions' could work well for this.
Beta Was this translation helpful? Give feedback.
All reactions