05:51 <glyph> so I wanted to go give PyPy a try on Apple Silicon since it's been a while since https://foss.heptapod.net/pypy/pypy/-/issues/3566 landed

05:53 <glyph> but I cannot find a nightly build to download

05:54 <glyph> the only two macosx_arm64's on https://buildbot.pypy.org/nightly/py3.9/ are 0 bytes

06:10 <glyph> I am falling pretty deep into this rabbit hole, I downloaded a pypy2 that seems to be functional and now I'm translating pypy3 myself, I don't really know why I am doing this given that I have no idea if the relevant changes are even on the py3.9 branch

06:10 <glyph> I see some mandelbrots though so cool I guess

06:20 <LarstiQ> :)

06:21 <LarstiQ> glyph: from reading what fijal and mattip write here, my impression is that it's still a little early for the Apple silicion pypy

06:25 <glyph> LarstiQ: Yeah, I mean, I'm expecting to report bugs

06:25 <glyph> LarstiQ: I am not gonna look at mandelbrot ascii art getting generated on my own terminal and then think to myself "I bet this is pretty robust production software, let me put it into production"

06:26 <glyph> LarstiQ: any pointers for me to not just waste a bunch of time translating something that's gonna fail with a traceback after 2 hours?

06:26 <glyph> like what branch to check out, etc

06:30 <LarstiQ> I don't remember seeing that mentioned on irc, I'd have to dig for a relevant looking commit

06:31 <glyph> LarstiQ: also, just curious, is any of this stuff cached? or is every translation the same multi-hour exercise

06:35 <glyph> okay now we're to the compilation stage and dang this is the first thing that has caused this machine's fans to turn on that didn't involve 3D rendering

06:35 <glyph> and they are _very_ on

06:40 <glyph> ↪ Thu Jun 16 23:39:10 PDT 2022 (★ PATH="${PATH}:/Users/glyph/Downloads/pypy-c-jit-105731-00a230410d68-macosx_arm64/bin" pypy ../../rpython/bin/rpython --make-jobs=12 --opt=jit) (1286 seconds elapsed)

06:41 <glyph> it runs pip

06:41 <glyph> well, this was definitely a surprise]

06:41 <glyph> I feel like 21 minutes is a pretty good translation time?

06:47 <glyph> runs twisted too, dang, this like, 100% works great, good job everybody

06:50 <LarstiQ> glyph: no caching on translations, that idea keeps resurfacing throughout the years but is an actual research project

06:51 <LarstiQ> and yes, translation time is pretty good on that fancy hardware :)

06:51 <glyph> LarstiQ: I'd much rather get parallelism than caching, not that my opinion means anything here :)

06:51 <glyph> 2 minute translation times could really improve development though I'd imagine

06:52 <LarstiQ> perhaps, but pypy devs are used to not translating all the time and developing with tests

06:52 <LarstiQ> I myself am mostly a lurker though

06:52 <glyph> woohoo and it's 2x as fast as cpython on this hardware

06:53 <glyph> at least in this one extremely stupid microbenchmark

06:53 <LarstiQ> :)

06:53 <LarstiQ> (my understanding of the translation is that it's a whole-program analysis, making the caching and parallelism a bit trickier)

06:54 <glyph> yeah I certainly understand what it's a difficult problem

06:54 <glyph> frankly if arigo can't figure it out I have no hope of even approaching the problem

06:55 <LarstiQ> glyph: if you have a bachelor/master/phd student for cfbolz I'm sure he'd be happy to supervise such work ;)

06:59 <cfbolz> glyph: yeah, it's our oldest piece of technical debt

13:29 <mattip> glyph: thanks for the heads up, I didn't notice the macos package upload was empty.

13:29 <mattip> should be fixed now

13:30 <mattip> fwiw, there is still something fishy with vararg and libffi, see for instance this issue

13:30 <mattip> https://foss.heptapod.net/pypy/pypy/-/issues/3762