<nimaje>
hm, seems like you have to limit the precision of the inputs for that continued fractions stuff, it would be way cooler (and likely not possible) if you really could do that exact and just have to decide on the precision you want for your results
<cfbolz>
nimaje: yeah, there are a few shortcuts
<cfbolz>
eg also the ints really need to be machine ints because of the use of numpy
infernix has quit [Ping timeout: 240 seconds]
lritter has quit [Ping timeout: 260 seconds]
infernix has joined #pypy
<mattip>
according to NEP 29, NumPy will be dropping support for python3.9 in April 2024
<mattip>
so during 2023 we should release at least pypy3.10
<mgorny>
mattip: another important goal is moving away from python2 for rpython
<mgorny>
people are on my arse pointing out that we're giving pypy a lot more time than any other project in gentoo
<mattip>
mgorny: refactoring rpython to python3 is a big project, I am not sure it will happen without sponsorship
<mattip>
but I can understand that it causes problems for a compile-from-source distro
<mgorny>
mattip: i'm afraid i can't offer anything myself, i'm unemployed and leading towards going broke ;-)
<mattip>
oy - that sounds bad
<mgorny>
but i can pass the message on if you announce something
<mgorny>
well, it's better than how i felt in my last job
<mattip>
we have not had much luck with go-fund-me type projects
<mgorny>
plus the frustration that i'm wasting my life earning money that i have no time to use, and inflation is eating my savings away
<mattip>
better to have a single funder provide most of the funds
<mgorny>
i'm afraid i have no clue how to organize something like that
<mgorny>
i mean, i could ask Gentoo Foundation if they're willing to sponsor something like that but i'm not sure they could justify spending large money on it
<mgorny>
plus it probably involves writing proposals, plans, scary stuff
<mattip>
it would be weird for the Gentoo Foundation to sponsor that, I agree. How does Gentoo bootstrap compiling a compiler?
<fijal>
mattip: do you know if we can organize people to even do it?
<mgorny>
mattip: generally either using precompiled binaries from somewhere, or using a cross-compiler
<mattip>
fijal: I am optimistic we would find people, yes
<fijal>
ok
<fijal>
you would need to find a corporation IMO
<fijal>
maybe PSF, but I kinda doubt it
<fijal>
I don't think anyone actually cares about "pypy uses internally some python version"
<mgorny>
mattip: "weird", maybe. but i've always been pointing out first-class pypy3 support on gentoo as one of its key advantages, so if we want to keep that advantage...
<fijal>
well, except gentoo, or course
<mattip>
mgorny: so couldn't you justify downloading pypy2.7 to build pypy3 in the same way?
<mgorny>
mattip: that's what i do now but this be a growing problem
<mgorny>
new arch will still require bootstrapping using python2
<mgorny>
then python2 will stop building again and we'll have to maintain even more patches to keep it alive
<mgorny>
fijal: i think fedora wants to kill python2 too, and they prefer building pypy3 purely from source as well
<mgorny>
don't know about debian
<mgorny>
there's also the option of trying summer of code but it'd be hard to find a competent student from the general pool
<mattip>
in order for pypy to support a new architecture, it needs a new backed tailored to that architecture, so it doesn't make sense to build PyPy3 for a new architecture until there is upstream support
<mgorny>
might work if you had someone ready and willing
<mattip>
"upstream support" == a working pypy2 for that architecture
derpydoo has quit [Ping timeout: 256 seconds]
jcea has joined #pypy
<arigato>
my point of view is that if someone really wants to port rpython to python3 he can do it but it would be a big waste of time imho. but who am I to do what people should do or not do
<arigato>
...who am I to say...
jcea has quit [Quit: jcea]
jcea has joined #pypy
<arigato>
sorry, didn't mean to sound so negative. the point is that it's a lot of work, and when it's done we would be dependent on the bytecode format, which changes every time there is a new 3.x+1, so someone would need to keep updating it, even if we don't add any new features to RPython
<antocuni>
or we can declare that the RPython compiler is not written in Python 2; it's written in the language Foo, and pypy provides an implementation of it
<arigato>
great plan, +1
<antocuni>
I suspect that if you factor out the "Python 2 is bad" logic, then pypy's current architecture becomes much less controversial
<antocuni>
and to solve the bootstrapping problem, we can also provide a minimal version of the Foo language implemented in C. Let's call it CFoo 2.7 😅
<arigato>
yes, I suspect that all this "let's reimplement RPython so that it is Python 3" would be a lot more efforts than just maintain a minimal portable version of CPython-er-sorry-CFoo 2.7
<antocuni>
it might even be easier to _rewrite_ RPython more or less from scratch to be a saner language than trying to port it to py3k (I'm not completely sure about this statement, but the fact that it _might_ be true should already tell something)