FS#25855 - [pypy] Crashes immediately with Illegal hardware instruction, needs rebuild.

Attached to Project: Community Packages
Opened by Sverd Johnsen (sjohnsen) - Thursday, 01 September 2011, 19:34 GMT
Last edited by Sven-Hendrik Haase (Svenstaro) - Sunday, 04 September 2011, 17:31 GMT
Task Type Bug Report
Category Packages
Status Closed
Assigned To Sven-Hendrik Haase (Svenstaro)
Architecture All
Severity Medium
Priority Normal
Reported Version
Due in Version Undecided
Due Date Undecided
Percent Complete 100%
Votes 0
Private No

Details

Hi,

pypy dumps core immediately with "illegal hardware instruction", this is actually expected and documented on their site. I'm not sure how useful pypy is without SSE2 or how many people want to run it on "ancient" hardware but given that it's in the i686 repos I suppose this should be fixed.

Adding --jit-backend=x86-without-sse2 should probably take care of that.
This task depends upon

Closed by  Sven-Hendrik Haase (Svenstaro)
Sunday, 04 September 2011, 17:31 GMT
Reason for closing:  Fixed
Comment by Sven-Hendrik Haase (Svenstaro) - Thursday, 01 September 2011, 20:04 GMT
Actually I think SSE2 hardware is be reasonable to assume in people's machines. I also think it is currently probably is more useful than taking it out and considering how few people this would likely concern, I'm inclined to just close this and tell you to use ABS to compile your own custom version of pypy without SSE2 support.
Comment by Sverd Johnsen (sjohnsen) - Friday, 02 September 2011, 10:33 GMT
  • Field changed: Percent Complete (100% → 0%)
If SSE2 hardware is reasonable to assume, why do we have a i686 repository?
Comment by Sven-Hendrik Haase (Svenstaro) - Friday, 02 September 2011, 11:21 GMT
Not all SSE2 capable hardware is x86_64 compatible (early P4, Pentium M). Pypy mailing list seems to suggest that pypy jit is not really well tested without SSE2. We still have 40% of people that run i686 and that's why we have that repo. About all desktop processors in the last 10 years support SSE2 and I think it is reasonable to assume that people who'd want to run pypy have that hardware. In conclusion, I will not disable that option for our users.

I have, however, made you a package because building pypy is unreasonable on old hardware. http://178.63.102.135/svens_stuff/pypy-1.6-1-i686.pkg.tar.xz

I will also leave this bug open for a little to gather some opinions in case more users feel like you.
Comment by Sven-Hendrik Haase (Svenstaro) - Saturday, 03 September 2011, 22:43 GMT
Actually I'll close the bug to get it off my list. Please reopen it if you want to add opinions.
Comment by Eric Belanger (Snowman) - Sunday, 04 September 2011, 02:59 GMT
For the record, the packages in the i686 repo are supposed to run on all i686 systems. SSE2 is not supported by all i686 CPUs so it should be disabled for the i686 package. You can keep it for the x86_64 package though. Many package do that (see blender, for e.g.).

However, I don't use pypy and I don't know how it work well without SSE2. If it doesn't work at all without SSE2, then I guess we will need to keep the SSE2.
Comment by Sven-Hendrik Haase (Svenstaro) - Sunday, 04 September 2011, 04:59 GMT
Very well, will put the non-sse2 package into [community] and see what people say. It seems functional but I don't know about speed.
Comment by Sven-Hendrik Haase (Svenstaro) - Sunday, 04 September 2011, 15:20 GMT
Done in -2.

Loading...