Re: Bug#730216: please consider allowing pypy to build on machines with less ram.
Hi peter (2013.11.22_21:57:12_+0200)
> I did consider filing a bug asking them to only build pypy stuff on
> architectures that had pypy but I decided to file a bug on pypy first
> and see what the response was.
Yeah, I suspect that libraries building for PyPy are going to have to
have arch-specific PyPy Build-Depends. And pybuild will have to learn to
ignore pypy on the archs without it.
Right now, PyPy should run on all Debian architectures (mostly without
JIT). The exceptions are Hurd, and archs where we don't have big enough
buildds. There are also known issues on big-endian that nobody has
bothered to fix yet.
> Sorry if my post wasn't clear, what I meant to say is
> "I still think it is preferable to grind swap on a buildd for a few
> days than to not have the package built at all"
I don't want to do that without porter approval. We do this on s390x,
where swap is ridiculously fast. I should probably contact the other
There are also, many buildds that don't have enough swap :P
> We accept build times in the days for several other packages, I
> don't see why we shouldn't accept them for pypy. Or do you think a
> pypy build with only 1GB of ram would take more than a "few days"?
More on the scale of a few weeks. I successfully built on an armhf
i.MX51 board, in about a week, at some point.
An unfortunate concern is that occasionally (quite rarely) pypy locks up
during builds, so having a timeout is useful.
> I also have armhf machines with more than 1GB of ram but AIUI policy
> requires that packages be buildable on the autobuilders :/.
I don't think that would violate the policy. It is buildable, just not
in a reasonable timeframe.
H: +27 21 461 1230 C: +27 72 419 8559