races between process.exe and libicalss_cxx.a

[starius]

Following on from dcc40c7 and #219, libical build still fails sometimes when run with make -j 4.

I'm building libical as a part of MXE.
Build script: https://github.com/mxe/mxe/blob/master/src/libical.mk
Patch applied to libical: https://github.com/mxe/mxe/blob/master/src/libical-1-fixes.patch
Version of libical: 2.0.0

When I was rebuilding all MXE packages, libical build failed. It is not reproducible, but I have a log:

[ 91%] Linking CXX executable process.exe
[ 91%] Linking CXX static library ../../bin/libicalss_cxx.a
make[4]: Leaving directory '/home/dell/mxe/tmp-libical-i686-w64-mingw32.static/libical-2.0.0/build'
[ 91%] Built target icalss_cxx
[ 92%] Building C object src/test/CMakeFiles/regression.dir/regression-utils.c.obj
/home/dell/mxe/usr/bin/i686-w64-mingw32.static-ld: cannot find ../../bin/libicalss_cxx.a: No such file or directory
collect2: error: ld returned 1 exit status
src/test/CMakeFiles/process.dir/build.make:102: recipe for target 'src/test/process.exe' failed
make[4]: *** [src/test/process.exe] Error 1
make[4]: Leaving directory '/home/dell/mxe/tmp-libical-i686-w64-mingw32.static/libical-2.0.0/build'
CMakeFiles/Makefile2:1013: recipe for target 'src/test/CMakeFiles/process.dir/all' failed
make[3]: *** [src/test/CMakeFiles/process.dir/all] Error 2
make[3]: *** Waiting for unfinished jobs....

Full version: https://gist.github.com/78ee244a0a628f8b23799d66ca5c8d7a

It looks as a race between process.exe and libicalss_cxx.a.

Output of make -nd applied after the failed build:
https://gist.github.com/e948dec593f1ca84533a757e9d110360

The output of make -nd suggests that process.exe does depend on libicalss_cxx.a. I do not know why libicalss_cxx.a did not exist when process.exe was linking.

[winterz]

I verify this when cross-compiling.
Can you try this patch https://paste.kde.org/pi5x1vwen

[tonytheodore]

With the patch above, I'm seeing slightly different errors:

Scanning dependencies of target icalss_cxx
[ 78%] Building CXX object src/libicalss/CMakeFiles/icalss_cxx.dir/icalspanlist_cxx.cpp.obj
[ 79%] Linking C static library ../../bin/libicalvcal.a
[ 79%] Linking CXX static library ../../bin/libicalss_cxx.a
[ 79%] Built target icalvcal
Scanning dependencies of target testvcal
Scanning dependencies of target testmime
make[4]: *** No rule to make target `bin/libicalss_cxx.a', needed by `src/test/testvcal.exe'.  Stop.
make[4]: *** Waiting for unfinished jobs....
Scanning dependencies of target parser
Scanning dependencies of target process
Scanning dependencies of target copycluster
make[4]: *** No rule to make target `bin/libicalss_cxx.a', needed by `src/test/testmime.exe'.  Stop.
make[4]: *** Waiting for unfinished jobs....
make[4]: *** No rule to make target `bin/libicalss_cxx.a', needed by `src/test/parser.exe'.  Stop.
make[4]: *** Waiting for unfinished jobs....
make[4]: *** No rule to make target `bin/libicalss_cxx.a', needed by `src/test/process.exe'.  Stop.

Full log https://gist.github.com/tonytheodore/16ddf6df366557ef9b40dba64c582d71

[tonytheodore]

I do not know why libicalss_cxx.a did not exist when process.exe was linking.

It appears to be built twice:

[ 17%] Linking CXX static library ../../bin/libicalss_cxx.a
...
[ 91%] Linking CXX executable process.exe
[ 91%] Linking CXX static library ../../bin/libicalss_cxx.a

[tonytheodore]

I believe the cause of the double build is that icalvcal-static is added as a library, then it's output is renamed to icalvcal when not MSVC.

A similar renaming is done for ical, icalss and the _cxx variants.

[winterz]

in the master branch I just commited 825e25a that might fix this issue. I tested every combination I could think of. Please test.

[winterz]

no response in almost 3 months. closing. please re-open if you still encounter this problem in master branch.