Nightly build fails after adding parallel builds about deviation HOT 10 CLOSED

I relaunched it and it passed.
I suppose this is because of make -j2

from deviation.

OK. Let's wait for the next nightly before closing the issue.

from deviation.

i will take a look

from deviation.

where can I find the error? I didn't find error build from: https://travis-ci.org/DeviationTX/deviation/builds

from deviation.

I've been using dmake -j4 zips for local test builds. Sometimes it works and sometimes not. At least one problem seems to happen at the beginning when parallel jobs are trying to build the cm3 library at the same time. Maybe they're all putting the library in the same destination?

  CC      sync.c
  CC      dwt.c
  CC      dwt.c
  AR      libopencm3_stm32f1.a
  CC      sync.c
arm-none-eabi-ar: /git/deviation/src/libopencm3/lib/libopencm3_stm32f1.a: Error reading sync.o: File truncated
make[3]: *** [/git/deviation/src/libopencm3/lib/libopencm3_stm32f1.a] Error 1
make[2]: *** [lib/stm32/f1] Error 2
make[1]: *** [libopencm3/lib/libopencm3_stm32f1.a] Error 2
make: *** [zip_devo7e] Error 2
make: *** Waiting for unfinished jobs....
  CC      dwt.c
  AR      libopencm3_stm32f1.a

from deviation.

the cm3 lib only build once and place the fixed place in current Makefile. I can fix it by move it to individual objs/ folder. however it need change Makefile of libopencm3, which is not good practice.

In Makefile of libopencme, the target folder is not configurable as far as my read:
SRCLIBDIR:= $(realpath lib)
it is fixed to lib.

from deviation.

make -j4 seems to work fine for me. Shall we close this issue for now until the issue reappears?

from deviation.

It's a race condition. On my VM build box it still fails occasionally. With google's help came up with the following solution. Change lines 69-70 of target/common/devo/Makefile.inc to:

$(LIBOPENCM3):
    +flock $(LIBOPENCM3).lock $(MAKE) -C $(SDIR)/libopencm3 TARGETS=stm32/f1 lib; rm -f $(LIBOPENCM3).lock

This forces only one of the make jobs to build the libopencm3 library, which is the job process that gets the lock. The others wait till the library is built. When the library is built the lock is released. The waiting jobs still run the libopencm3 make, but the library is already up-to-date so it immediately returns.

But after this change, in my build environment, the build still fails because the optimize.ld file is shared for all targets. Need to fix that also...

from deviation.

please check PR #493

from deviation.

Fixed by #493

from deviation.