pypy 5.0.0 on i386

[David Naylor]

On Tuesday, 15 March 2016 09:43:15 Bartłomiej Rutkowski wrote:
> > On 14 Mar 2016, at 19:47, Konstantin Belousov <kostikbel at gmail.com> wrote:
> > 
> > Hello,
> > is pypy supposed to build on i386 ?  I get the following error on the
> > attempt to build pypy in i386 current jail on current amd64 box.
> > 
> > uname -a
> > FreeBSD sandy32 11.0-CURRENT FreeBSD 11.0-CURRENT #919
> > r296579+bdd6db3(sysv_sem)-dirty: Thu Mar 10 12:43:41 EET 2016    
> > root at pooma.home:/usr/home/kostik/work/build/bsd/DEV/obj-amd64/usr/home/ko
> > stik/work/build/bsd/DEV/src/sys/X  i386
> > 
> > The immediate cause of the error is that vmprof_config.h is populated with
> > the snippet that uses mc_rip mcontext member to get instruction counter
> > value (seemingly).  This is a valid attempt on amd64, which has %rip,
> > but i386 has %eip for instruction counter and mcontext member is called
> > mc_eip.
> > 
> > My vmprof_config.h content, I am not sure if the file is generated:
> > #define HAVE_SYS_UCONTEXT_H
> > #if defined(__FreeBSD__) || defined(__APPLE__)
> > #define PC_FROM_UCONTEXT uc_mcontext.mc_rip
> > #else
> > #define PC_FROM_UCONTEXT uc_mcontext.gregs[REG_RIP]
> > #endif
> > 
> > Script of the build is attached.
> > 
> > Thanks for the advise.
> > <1.log>
> 
> To be perfectly honest - I dont know, I havent done the update to 5.0.
> However, since previous versions did, I see no reason why this wouldnt, and
> I'd expect that update provided by a committer to be well tested on all
> supported platforms. I will make my test run, but it will take some time.

Hi,

Apologies, it was my fault for not properly testing the PyPy update before 
pushing it.  I incorrectly assumed there wouldn't be any issues between i386 
and amd64.  I'll ensure to run both poudriere builds before committing.  

This has been fixed in r411414 and reported upstream with bug #2262.  

@kib: thank you for the report, your explanation around eip vs rip was most 
instrumental in finding a solution.  

P.S. It appears that the PyPy team will be issuing a bug fix quite soon, if 
anyone would like to submit fixes for some unit tests ;-).  

Regards,
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 949 bytes
Desc: This is a digitally signed message part.
URL: <http://lists.freebsd.org/pipermail/freebsd-python/attachments/20160319/931bd177/attachment.sig>