77792 - Wrong compiler option for Solarisx86 building freebl

Status: RESOLVED FIXED

(NSS :: Build, defect, P3)

Description

[Margaret Chan]

Trying to build PSM2 on Solaris x86, the build failed with segmentation fault.
That appears to be a compiler specific problem which we have addressed to
the compiler group.  However, we've also noticed that the compilation 
is picking up Solaris Sparc specific options (-xarch=v8) which should not be
there:

/dist/include  -O -KPIC -DSVR4 -DSYSV -D__svr4 -D__svr4__ -DSOLARIS -Di386 -DSO
LARIS2_7 -D_SVID_GETTOD -xarch=v8 -DXP_UNIX -UDEBUG -DNDEBUG
-DMP_API_COMPATIBLE
 -I/usr/dt/include -I/usr/openwin/include
-I/h/rachael/export3/builds/forte/5.7/
i386/mtchan042401/mozilla/dist/include/nspr  -I../../../../dist/public/security
-I../../../../dist/private/security -I../../../../dist/include  mpi.c cc:
Warning: illegal option -xarch=v8 cc: Warning: multiple use of -K option,
previous one discarded. cc: Warning: illegal option -xarch=v8
"mpi.c": [mp_to_unsigned_octets]: internal: SIGNAL: Segmentation Fault
cc: mcs failed for mpi.c
gmake[4]: *** [SunOS5.7_i86pc_OPT.OBJ/mpi.o] Error 2
gmake[4]: Leaving directory
`/h/rachael/export3/builds/forte/5.7/i386/mtchan0424
01/mozilla/security/nss/lib/freebl'

Comment 1

[Wan-Teh Chang]

Margaret, thanks for the bug report.  We don't build and test
our code on Solaris x86 ourselves but we will be happy to review
and merge patches.

Can you look at the Makefile and see where we need to more specifically
test for CPU_ARCH being sparc?

Comment 2

[Margaret Chan]

Sure.  I will look into that and try to come up with a patch.

Comment 3

[Nelson Bolyard (seldom reads bugmail)]

Margaret:

Thanks for the pointer to the solaris/x86 system where I can build
with gcc and Forte'.

I made a one line change to coreconf/SunOS5.7_i86pc.mk

 CPU_ARCH               = x86
+ARCHFLAG               =
 OS_DEFINES             += -Di386

and that seemed to solve the immediate problem by eliminating the 
-xarch=v8

I've now built NSS (tip of trunk) with both gcc and Forte' on 
that system, and have checked in the coreconf makefile changes
(files SunOS5*pc.mk) to the tip of the trunk.  

I consider this fixed, but won't mark it so until we figure out
how you can use these fixes (since you're not pulling from the 
tip of the trunk.)

Comment 4

[Margaret Chan]

Thanks for the fix, Nelson.  Well, we were pulling out using the NSS_CLIENT_TAG,
which was used by PSM2.0.  Unless they pull out from a new tag, it does not
seem like that we will get this anyhow.  For now, we'll patch it.  Not really a
big problem.  I guess I will check with Javier to see if there is any plan to
pull out from a new tag for PSM2.0.  If so, then perhaps we can apply to
whatever branch that he is going to tag to.      

Comment 5

[Nelson Bolyard (seldom reads bugmail)]

I'll play Karnak and predict that the client tag will move soon
for both security/coreconf and probably also security/lib/freebl.
Lots of build problems for various platforms have been solved in 
those directories since the current client tag was created.  
The tag's gotta move, or a new tag must be created, IMO.

Comment 6

[Wan-Teh Chang]

Nelson: the NSS_CLIENT_TAG has been moved on the modified files,
so I am marking this bug fixed.

Comment 7

[Wan-Teh Chang]

The fix is in 3.3 as well.