Hi Felix,
Sorry for the long delay, but I'm back to the issue, because I need to
fix it.
Please ignore what I wrote if you already fixed the issue, and if so,
please tell me how (I di't fix it correctly yet).
Zhang Xiaofei a écrit :
For people who don't know, I'm colleague of the author of the original
message, we build on Fuloong Minis, which has same processors as Gdium
machines has.
I observed the build output and found the switch still in use. The -fpic
switch may have been moved to soleenv/inc/unxlng.mk in the fix of i89237
"Unify the linux .mk files". If that's the case I guess there's other issues
causing the breakage.
Maybe, maybe not.
About optimisation, what I have remarked, that the default value is
CDEFAULTOPT*=-O2 in solenv/inc/unxlngx.mk
And in solenv/inc/unxlngmips.mk, there is :
CDEFAULTOPT=-Os
Like Caolan explained.
Now, reading gcc documentation, I found that the last value passed wins.
So we can consider the CDEFAULTOPT*=-O2 wins, because -Os is written
before the include unxlng.mk (if I'm not wrong) in unxlngmips.mk.
This could explain the new behavior too.
Till now we have tested DEV300_m58 and DEV300_m60, the breakage took place
in the latter but not the former one.
Confirmed.
By the way we tried the following workaround: adding -mips2 (or -mips3) for
gcc globally, the build could continue but office will crash with SIGBUS,
Yes, I confirm the build work for me in comphelper too using -mips3, but
use -mips3 is imho not the right option, because (yet reading the
documentation) mips3 is for 64 bits, and we do 32 bits chroot with linux32.
In fact, to save time, I started the build using mips3 and I started
reading the documentation in meantime ( bad move ;-) , but once I'll
have verified the build is ok, I'll restart one using
-march=mips32
Did you try that one ? (I didn't saw it in the build, but maybe I was
blind). Or maybe thre is a better one ?
this can be solved by using configure --with-alloc=system as another
workaround.
Yes, but the root is imho elsewhere, and it really worth to read the
full gcc documentation, because mips(el there) seems to have a *lot* of
possiblities (really nice proc imho).
As conclusion, for me, some gcc option has been modified, and since
there is an optimisation issue, or maybe we simply hit a missing
parameter ( like -march=aValue )
To be continued
Thanks,
Eric
--
Education Project:
http://wiki.services.openoffice.org/wiki/Education_Project
Projet OOo4Kids : http://wiki.ooo4kids.org/index.php/Main_Page
L'association EducOOo : http://www.educoo.org
Blog : http://eric.bachard.free.fr/news
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscr...@openoffice.org
For additional commands, e-mail: dev-h...@openoffice.org
