<arigato>
_aegis_: fwiw, I suspect that MAP_JIT does something unexpected on its pages after a fork(), but I didn't manage to find any documentation for MAP_JIT so far
mattip has quit [Remote host closed the connection]
<arigato>
maybe write a small test in C? do a mmap(MAP_JIT), write a \x42 in the page, call os.fork(), and try to read from both parent and child. and then try to write, too, and check that the page is private (changes don't show up in the other process)
<arigato>
(this is required for pypy's jit)
<arigato>
4b6061b47541 might in theory be a reason for a few translations randomly having worse performance
oberstet has quit [Remote host closed the connection]
oberstet has joined #pypy
mattip has joined #pypy
xcm has quit [Remote host closed the connection]
xcm has joined #pypy
<njs>
fijal: imagining a skeptical viewer, the two questions that jump to mind: (1) "what, python 2.7.2? what kind of cooked benchmarks are these", (2) "yeah sure twisted, whatever, they still can't handle the C API or python 3"
<fijal>
njs: good feedback!
<fijal>
njs: can you write it to python-dev?
<njs>
I'm certainly not going to write that to python-dev :-)
<njs>
I would prefer to be lazy and not write it to pypy-dev either
<fijal>
pypy-dev
<fijal>
hahaha ok fair :)
<fijal>
I'll write it
antocuni has joined #pypy
<kenaan>
antocuni release-pypy3.5-7.x aaff09311f03 /pypy/: fix the version number AGAIN
<kenaan>
antocuni release-pypy3.6-7.x e83543812df5 /pypy/doc/conf.py: fix the version :(
<kenaan>
arigo release-pypy2.7-7.x c8805ee6d784 /rpython/conftest.py: Crashes with too-old versions of hypothesis. Turn crashes into warnings
<kenaan>
arigo release-pypy3.5-7.x 928a4f70d3de /rpython/conftest.py: Crashes with too-old versions of hypothesis. Turn crashes into warnings
<kenaan>
arigo release-pypy3.6-7.x fb40f7a5524c /rpython/conftest.py: Crashes with too-old versions of hypothesis. Turn crashes into warnings
<_aegis_>
arigato: I already posted a follow up to the pull request and filed an issue with apple, since this will affect any jitted language with fork()
<_aegis_>
MAP_JIT pages are simply not shared with the fork()ed process
<_aegis_>
I assume because they only used MAP_JIT in Safari's javascript implementation and that never forks they never cared
ronan has joined #pypy
mattip has quit [Read error: Connection reset by peer]
ronan has quit [Read error: No route to host]
ronan__ has joined #pypy
mattip has joined #pypy
<tos9>
cfbolz: hello hello, I did
<tos9>
(unfortunately)
<tos9>
when I was on my way to the airport the uber driver told me about the airport strike
<tos9>
So I was kind of hoping they'd cancel my flight and I'd be forced to stay :)
marky1991 has quit [Remote host closed the connection]
marky1991 has joined #pypy
marky1991 has quit [Remote host closed the connection]
<tos9>
antocuni: so I used victor's perf package and confirmed that at least on my machine, that microbenchmark is 25% faster using struct.unpack than ord, both when I read from /dev/zero and when I read from /dev/random (which was the same result as my 1GB file)
<tos9>
(which I still don't understand either obviously :), but figured I'd spend more time on it later and make believe I was still sprinting)