+Masahiro
Hi Andreas,
On 29 January 2015 at 11:05, Andreas Bießmann andreas.devel@googlemail.com wrote:
Hi Simon,
I add some more descriptive text.
On 29.01.15 18:41, Simon Glass wrote:
Hi Andreas,
On 31 December 2014 at 05:44, Andreas Bießmann andreas.devel@googlemail.com wrote:
Hi Simon,
while test-building 2015.01-rc4 I encountered following strange behaviour of buildman:
---8<---
This time create non existing /tmp/bar and run the build for all avr32 boards on the last commit.
andreas@andreas-pc % ./tools/buildman/buildman -b buildtest -o /tmp/bar -v avr32 boards.cfg is up to date. Nothing to do. Building 1 commit for 10 boards (6 threads, 1 job per thread)
Ok, it starts 6 threads ...
01: Prepare v2015.01-rc4 avr32: + atngw100mkii 0 4 6 /10 0:00:02 : atngw100mkii
Ouch ... six errors, what's going on here? Let's print the errors and summary ...
./tools/buildman/buildman -b buildtest -o /tmp/bar -v avr32 82.57s user 16.90s system 249% cpu 39.899 total andreas@andreas-pc % ./tools/buildman/buildman -b buildtest -o /tmp/bar -v -lsed
+(grasshopper,atngw100,favr-32-ezkit,atstk1006,atstk1004,hammerhead) Could not find linker script.
w+(atngw100mkii,atstk1002,atstk1003,mimc200) ../tools/kwbimage.c:803:8: warning: ‘headersz’ may be used uninitialized in this function [-Wmaybe-uninitialized]
Oups some complains about missing linker script and something is wrong with kwbimage .... Let's see what's going on with grasshopper, I know this builds fine, so let's see:
andreas@andreas-pc % ./tools/buildman/buildman -b buildtest -o /tmp/bar -v -lsed grasshopper boards.cfg is up to date. Nothing to do. Summary of 1 commit for 1 boards (1 thread, 6 jobs per thread) 01: Prepare v2015.01-rc4 avr32: + grasshopper +(grasshopper) Could not find linker script. +(grasshopper) make[1]: *** [prepare1] Error 1 +(grasshopper) make: *** [sub-make] Error 2
Ok, missing linker script. That is wrong, I know grasshopper builds fine! So run another build for just the grasshopper board with another output directory (/tmp/grasshopper) ... thus a fresh build.
andreas@andreas-pc % ./tools/buildman/buildman -b buildtest -o /tmp/grasshopper -v grasshopper boards.cfg is up to date. Nothing to do. Building 1 commit for 1 boards (1 thread, 6 jobs per thread)
Ok, this time it starts just one thread.
Cloning repo for thread 0 01: Prepare v2015.01-rc4 avr32: + grasshopper 0 1 0 /1 grasshopper
Yep, it builds fine (besides the warning) ...
./tools/buildman/buildman -b buildtest -o /tmp/grasshopper -v grasshopper 14.11s user 2.69s system 183% cpu 9.171 total andreas@andreas-pc % ./tools/buildman/buildman -b buildtest -o /tmp/grasshopper -v -lsed grasshopper boards.cfg is up to date. Nothing to do. Summary of 1 commit for 1 boards (1 thread, 6 jobs per thread) 01: Prepare v2015.01-rc4 avr32: + grasshopper w+(grasshopper) ../tools/kwbimage.c: In function ‘kwbimage_set_header’: w+(grasshopper) ../tools/kwbimage.c:803:8: warning: ‘headersz’ may be used uninitialized in this function [-Wmaybe-uninitialized]
Ok, the warning is about kwbimage.c. I know this warning and it will be fixed soon.
So there must be something wrong when it builds more than one board at a time. I guess it has something to do with the threads.
Unless I am much-mistaken this is a Kbuild/Makefile problem rather than anything to do with buildman. You can use -T to control the number of threads.
buildman complains about missing linker script for most boards which is an error when building all avr32 boards. While it detects the correct warning for still not fixed kwbimage.c maybe-uninitialized when building just the single board which had an error before. Both builds are based on v2015.01-rc4 and built in different locations.
Sorry for not getting back sooner - twice I read your email and tried to understand what is going on.
Sorry for being not clear with my error description. I hope the additional thoughts may help.
Yes I understand now.
This happens on current Debian stable (which has python 2.7.3-4+deb7u1) but not with Debian testing (which has python 2.7.8-2) nor with my MAC box (which has Python 2.7.9, thanks to macports).
I'm copying Masahiro, as I recall he fixed a problem something like this recently.
Regards, Simon