antocuni changed the topic of #pypy to: PyPy, the flexible snake (IRC logs: https://botbot.me/freenode/pypy/ ) | use cffi for calling C | "PyPy: the Gradual Reduction of Magic (tm)"
Stefan_ has joined #pypy
inhahe_ has quit [Ping timeout: 260 seconds]
<Stefan_>
Hi, does anyone know why http://speed.pypy.org/ compares the current PyPy with CPython 2.7.2 which is 6.5 years (!) old? That seems pretty unreasonable to me.
<nimaje>
iirc because that was the fastes cpython 2.7.x in these benchmarks
<pjenvey>
that should maybe be mentioned on the front page of speed. it's frequently asked about
<Stefan_>
Hmm, then I'd find it very helpful if that was stated on the page. The way it is, it makes me think they're trying to trick me and that I can't trust them. Also, I'd be interested in the benchmark result of the *current* CPython 2 and 3, since that's what I'm using and thus that's what I'm interested in. If they're slower than 2.7.2, then I still want to know how slow they are.
<Joce>
Given that I have no homebrew or macports installed (and never have on that fresh install of high sierra), /usr/local/opt/ doesn't even exist, so it's not an appropriate folder for the linker to serach in
Joce has quit [Quit: Page closed]
exarkun has quit [Ping timeout: 272 seconds]
exarkun has joined #pypy
<Rotonen>
also quite many people install homebrew in $HOME to isolate it from expanding SIP enclaves across os versions
<fijal>
We should maybe clean up the os x builds at some stage
<fijal>
Right now they require some libraries built
<arigato>
ca3a17a0ae22 fixes a ctypes regression since 5.9, maybe worth a 5.10.1 release
<arigato>
(at some point)
exarkun has quit [Ping timeout: 264 seconds]
exarkun has joined #pypy
dddddd has joined #pypy
<LarstiQ>
iirc part of the osx/homebrew problem is standard OSX only coming with an ancient ssl
marr has joined #pypy
<Rotonen>
not accurate - a few macos versions ago apple stopped shipping the openssl dev headers
<Rotonen>
and in high sierra they actually ship libressl, not openssl, but still no dev headers
<Rotonen>
in apple land you're supposed to use their frameworks and cocoa apis for crypto, but most people don't want to bother, so there is a plethora of solutions for 'just doing the unixy thing'
<fijal>
LarstiQ: we distribute our own version of ressl with pypy on windows
<fijal>
I don't see why we can't do it for OS X
<fijal>
LarstiQ: I said "distribution" not "compilation"
<fijal>
it's ok if you need homebrew to compile pypy
<fijal>
well, "ok", but it's ok to install extra stuff
<LarstiQ>
Rotonen: btw, made any progress on win64, or waiting for the välipäivät?
<Rotonen>
LarstiQ: hoping to start looking into that later today, for now some more ikea to tackle
<Rotonen>
re: macos - one could pull in the older xcode package for compilation with the openssl headers, but treating both windows and macos uniformly as ~static binary targets sounds like the sane amount of effort over time
<mjacob>
arigato: regarding machine.platform() in soabi: i remember that some test failed and i showed stefan where this soabi string is created. not sure whether i participated in or reviewed this commit, but i'm responsible that this got through
antocuni has quit [Ping timeout: 252 seconds]
marr has joined #pypy
Joce has joined #pypy
yuyichao_ has joined #pypy
yuyichao has quit [Ping timeout: 265 seconds]
Joce has quit [Quit: Page closed]
<fijal>
Rotonen: so funnily enough, High Sierra has new functions in libc
<fijal>
maybe we can try to load them dynamically in the same binary
raynold has quit [Quit: Connection closed for inactivity]
<Rotonen>
fijal: most new macos versions will introduce both new syscalls and deprecate old ones, there are clang flags to make builds fail when you try to compile across os version targets where something would fall through the cracks (getentropy for ssl is an easy catch as that is easy to test for)
<fijal>
It's utimesat
<fijal>
The problem is cpython exposes it, so ideally we would check for it at runtime
<Rotonen>
also FUTIMENS
<Rotonen>
10.12 only introduced GETENTROPY, and it gets hairier towards the past as they used to do multiple things at once
<Rotonen>
which macos platforms do you target? the rolling two latest?
<Rotonen>
as in shuffle about what is a weakref and what is a hardref in the frameworks (mostly relevant for complex xcode projects)
<fijal>
yeah 2 or 3 latest
<fijal>
ones before had quite stupid bugs
<fijal>
there was one which made our complex code not work
<fijal>
(like complex numbers)
<fijal>
version of clang I believe
mat^2 has joined #pypy
<Rotonen>
compiling on 10.13 and supporting back to 10.10 is fairly straight forwards
<Rotonen>
if you want to support back to 10.9, hoop jumping starts
<Rotonen>
i think apple itself only supports the latest one these days, even for security, unless horrendously severe
<Rotonen>
so 2-3 latest is fair, not everyone can chase their hardware requirement targets
<Rotonen>
although now that everything first requires opengl3 hardware to render everything uikit and soon enough metal2 hardware, things will settle down
<fijal>
it's not that straightforward
<fijal>
because things
marr has quit [Ping timeout: 265 seconds]
<fijal>
not unattanable
<fijal>
maybe I should work on that a bit, being the only OS X based developer of pypy
<Rotonen>
if it sounds helpful, i'll try to clear out the red tape at $WORK and ask if i can share you what we do for our cross-target static-for-sandboxing cpython builds
<Rotonen>
our platform native bits, both mac and windows, are the rare odd closed source things we do
<Rotonen>
most of our products are gpl v2
<fijal>
Rotonen: great. I will be on holiday next two weeks, but I'll read IRC. Any help would be helpful!
dowwie has quit [Read error: Connection reset by peer]
yuyichao_ has quit [Quit: Konversation terminated!]
yuyichao_ has joined #pypy
<kenaan>
ltratt default 8390220c0526 /rpython/rlib/rvmprof/cintf.py: OpenBSD doesn't currently support vmprof.
marvin_ has quit [Remote host closed the connection]
marvin has joined #pypy
marvin is now known as Guest1045
Guest1045 has quit [Remote host closed the connection]
marvin_ has joined #pypy
mattip has joined #pypy
jcea has joined #pypy
yuyichao_ has quit [Ping timeout: 268 seconds]
tayfun26 has quit [Quit: tayfun26]
realitix has quit [Ping timeout: 265 seconds]
yuyichao_ has joined #pypy
realitix has joined #pypy
realitix has quit [Client Quit]
mat^2 has quit [Quit: Leaving]
marky1991_2 has joined #pypy
antocuni has joined #pypy
<Rotonen>
oh yay, cpython 2.7 has not been brought to the future, will actually need to grab vs 2010, gnarl ahoy
<Rotonen>
https://bugs.python.org/issue30742 <- or at least i'll take the word of steve dower from june on that and not start to chase the apparent "trivial few things" :-)
<Rotonen>
LarstiQ: so that's a no for progress today, vs 2010 wants to do a recursive update dance with service packs and all - maybe can achieve cpython compilation still today
<Rotonen>
mattip: i want the ide, that is the thing ms has done well
<mattip>
you should be able to use any ide with older compiler. Just specify it in the project
<Rotonen>
or rather a visual debugger will speed up debugging for me to warrant grabbing vs 2010
<Rotonen>
yes and no, rather not have it nag about the project files etc.
<mattip>
ok, whatever is easiest for your workflow
<Rotonen>
i will take the cpython recommended beaten path as that gets the least in my way when actually tackling the issue
<Rotonen>
yeah, i do not mind prep time for achieving a workflow
<Rotonen>
on the plus side win 10 seems to grab the recursive updates without user intervention, which is definite improvement
<Rotonen>
vs. rebooting and checking for updates and finding the next bunch again
jamesaxl has quit [Read error: Connection reset by peer]
jamesaxl has joined #pypy
<kenaan>
mattip py3.5 a4194a67868f /pypy/module/cpyext/test/test_eval.py: make test fail not hang on win32
<mattip>
the hacks to run -A tests on py3.5 make it very difficult to find hanging tests
<mattip>
there must be a better way to directly run tests using the pypy3 interpreter
raynold has joined #pypy
<arigato>
mjacob: machine.platform(): thanks, that's fine
antocuni has quit [Ping timeout: 265 seconds]
EWDurbin has joined #pypy
EWDurbin has left #pypy [#pypy]
jamesaxl has quit [Read error: Connection reset by peer]
jamesaxl has joined #pypy
dowwie has joined #pypy
dowwie has quit [Changing host]
dowwie has joined #pypy
<fijal>
arigato: hi
<arigato>
hi
* fijal
is flying above africa
<fijal>
arigato: thanks for all the work
<arigato>
heh, I was about to say thanks to you for all the work about doing the release
pilne has joined #pypy
<fijal>
well, matti did most as far as I'm concerned :)
<fijal>
I need to clean up OS X install, it's a complete mess right now
<fijal>
the binary that we ship is not really easily runnable
<Alex_Gaynor>
fijal: tbh, people on macOS should probably just use the pypy from homebrew
<fijal>
Alex_Gaynor: which grabs pypy from us ;-)
<fijal>
but yes
<Alex_Gaynor>
fijal: it compiles its own pypy
<fijal>
ok
<Alex_Gaynor>
pyenv grabs the official one
<fijal>
maybe ports do that?
<Alex_Gaynor>
I don't know what ports do; I haven't paid attentino to macports in years
kipras`away is now known as kipras
adamholmberg has joined #pypy
adamholmberg has quit [Ping timeout: 264 seconds]
<arigato>
if it was only me, I'd interpret these comments as "ok cool let's not make an official OS/X binary at all"
<Alex_Gaynor>
pyenv uses the prebuilt binaries. the only other option is pyenv buliding it on every users machine, which would suck. homebrew has CI which builds the binaries for all of it's users.
altendky has joined #pypy
<mattip>
pypy2-v5.9.0-osx64.tar.bz2 on bitbucket downloads has 1900 downloads in 10 weeks
<mattip>
maybe they are taking from portable builds?
<Alex_Gaynor>
no, they use that one; look at some of the other versions though, I think they were relatively slow to add 5.9.0
<mattip>
either there is caching somewhere or pypy is not very popular
<mattip>
on another note, I cannot seem to get the own cpyext tests to pass after merging avoid-roundtrip,
<mattip>
test_tupleobject used to call api.PyTuple_New, but that has moved to C?
<fijal>
mattip: I would expect a lot of people to use homebrew or docker
<njs>
arigato: tbh I'm not sure why you have official linux binaries either, when squeaky's are better for ~everyone
<njs>
(or maybe squeaky's should be official or something)
<mattip>
njs: do they play well with cextensions that may link to system libraries like lxml, requests?
<njs>
mattip: AFAIK? why wouldn't they?
marky1991_2 has quit [Ping timeout: 264 seconds]
<mattip>
isn't there a problem with mixing incompatible versions somewhere along the line?
<njs>
mattip: I believe the main differences are that his builds are compatible with every common distro's glibc, and they vendor the libraries like ssl that they actually need themselves and can't necessarily be found on the distro
<njs>
mattip: I can imagine some cases where you might have dll hell issues, but I can't think of any reason why squeaky's would have them worse than the current official binaries :-)
<kenaan>
mattip default 678518de33f8 /pypy/module/cpyext/test/test_tupleobject.py: api.Py_DecRef and api.PyTuple_New no longer exist, work around for tests
<mattip>
issue 2617 looks like serious breakage with the past, worthy to bump the major pypy version
<mattip>
like in moving from 5 to 6
marky1991_2 has quit [Ping timeout: 256 seconds]
marky1991_2 has joined #pypy
jamesaxl has quit [Quit: WeeChat 1.9.1]
mattip has left #pypy ["bye"]
marky1991_2 has quit [Remote host closed the connection]