cfbolz changed the topic of #pypy to: PyPy, the flexible snake (IRC logs: https://botbot.me/freenode/pypy/ ) | use cffi for calling C | the secret reason for us trying to get PyPy users: to test the JIT well enough that we're somewhat confident about it
<Katharsis>
I had a chance to use matplotlib with PyPy 5.10 on the same machine before
<Katharsis>
fijal: any ideas what about the root of the problem?
<ronan>
Katharsis: maybe something has changed in matplotlib? Did you try with an older version?
<Katharsis>
ronan: the same bro
<arigato>
Katharsis: PyOS_InputHook was never supported in pypy, but we talked about it already. Right now I'm searching for the "search" functionality of botbot.me...
<arigato>
Katharsis: if you can progress and are able to give us an answer to that issue, please do so
<arigato>
even if it's "no it doesn't work"
<Katharsis>
cool, I'll try
<arigato>
google seems to say it has been a known bug for ages for matplotlib on OSX
<arigato>
we don't have anyone who is interested in pypy+matplotlib+osx, and apparently there hasn't been anyone to contribute the one-line fix or the information that it's not enough
marky1991 has quit [Read error: Connection reset by peer]
marky1991 has joined #pypy
marky1991 has quit [Remote host closed the connection]
asmeurer__ has quit [Quit: asmeurer__]
marky1991 has joined #pypy
asmeurer__ has joined #pypy
asmeurer__ has quit [Quit: asmeurer__]
inad922 has quit [Ping timeout: 240 seconds]
stevenja_ has quit [Remote host closed the connection]
stevenja_ has joined #pypy
stevenja_ has quit [Remote host closed the connection]
stevenja_ has joined #pypy
redj has quit [Ping timeout: 260 seconds]
asmeurer__ has joined #pypy
redj has joined #pypy
Rhy0lite has quit [Quit: Leaving]
stevenja_ has quit [Remote host closed the connection]
stevenja_ has joined #pypy
stevenja_ has quit [Remote host closed the connection]
stevenja_ has joined #pypy
asmeurer__ has quit [Quit: asmeurer__]
marky1991 has quit [Read error: Connection reset by peer]
marky1991 has joined #pypy
Hasimir has joined #pypy
asmeurer_ has joined #pypy
TheAdversary has joined #pypy
stevenja_ has quit [Remote host closed the connection]
stevenja_ has joined #pypy
jsza has quit [Quit: Connection closed for inactivity]
stevenja_ has quit [Ping timeout: 248 seconds]
asmeurer_ has quit [Quit: asmeurer_]
jsza has joined #pypy
<jsza>
hey all. I'm using cppyy on pypy3 (latest py3.5 branch) and seeing ~10x slowdown calling C++ class methods directly vs storing the method in a variable and calling that. same code works fine on pypy2: vmprof jitlog shows time spent in JIT on pypy2, and none on pypy3. any ideas? https://hastebin.com/okewisezes.py
<jsza>
I should mention I've tried on pypy3 translated with --translation-jit_opencoder_model=big to no change. thought it might be producing a too-long trace for whatever reason but it seems not
lritter has quit [Ping timeout: 260 seconds]
kristina has joined #pypy
<kristina>
i'm really confused ... until today i thought pypy was a python JIT and then i wanted to embed it and after a million toolchain woes it not playing nicely with perforce style builds i got it to build. now i have no clue what pypy is, it seems that jit is disabled by default, so it's a python interpreter + c transcompiler?
<kristina>
now i need to work out how to build it statically but now i'm not sure what it is, i mean i'm linking everything into a massive ELF blob, i presume i need to keep symbols that pypy uses so it can use cffi by dlsyming the main image.
marky1991 has quit [Read error: Connection reset by peer]
<kristina>
the environment in which this runs is bizzare, basically it doesn't have ld.so, it uses a custom linker instead designed around the idea of an extremely small OS image basically booting to a monolithic ELF blob, it makes deployment easier, stuff like lgpl libraries are pulled from sections by the dynamic linker and loaded, but i want to roll pypy itself into the monolith since there's a huge ipo pass that
<kristina>
goes over it.
<kristina>
could someone explain this to me if they have a second? also it seems that it wants a c compiler at runtime.
<kristina>
and some builtin libs call cc ignoring --cc setting and CC env var so instead of /usr/local/sdk/llvm-7.0/bin/clang they call cc which doesn't exist.
<kristina>
that's just running the bootstrapper thing (pypy/goal).
<kristina>
it seems to dishonor the env var for artifacts too at weird times like pypy ends up in the checkout instead of in the artifact tree.
adamholmberg has quit [Remote host closed the connection]
<kristina>
i usually convert most projects to cmake or gn for integration (if they're autotools or makefile based) but there's a fallback option to build a project using a script that basically just needs to provide a way of building + installing it to/from env vars passed from build system, this is reserved for projects that are extremely difficult to port to a build system dsl and require multi stage build.
<kristina>
ie. when a project depends on multiple builds of itself using different configurations (pypy seems to be in that category).
<kristina>
ie. do i have to ship libclang.a alongside and hack that into rpython (?) because i can't actually ship clang, i can only ship the driverless libclang and call it manually to build c code, i mean it supports virtual file system so headers can be overlaid but then it just gets compilcated, libllvm can lower to mc objects in memory too but like what am i even doing.
<kristina>
(please don't take it as an offensive thing, pypy seems like a great project but i'm beyond confused as i always had the impression that it was just a cpython alternative that used jit)
<kristina>
i did get it to build (almost, it builds but it writes some stuff to checkout, have to find a way to stop it from doing that, it cannot write artifacts in non artifact directory).
nopf has joined #pypy
<kristina>
is pypy not the right tool for the job (which is to have a fast, preferrably JIT'd language that could be embedded to write some code that isn't a CPU bottleneck in a dynamic(-ish) language.
<kristina>
well my first question is how do i get the bootstrap part to honor $CC or --cc even all the way through.
<kristina>
as well as CFLAGS and LDFLAGS.
<kristina>
some modules fail to build because it forgets about CFLAGS or CC.
adamholmberg has joined #pypy
<nimaje>
you should write bug reports for that, but do you really need to build pypy yourself or could you use a package from your package manager?
<kristina>
"the environment in which this runs is bizzare, basically it doesn't have ld.so, it uses a custom linker instead designed around the idea of an extremely small
<kristina>
OS image basically booting to a monolithic ELF blob, it makes deployment easier, stuff like lgpl libraries are pulled from sections by the dynamic linker and
<kristina>
loaded, but i want to roll pypy itself into the monolith since there's a huge ipo pass"
<kristina>
er.
stevenja_ has quit [Ping timeout: 240 seconds]
<kristina>
so preferably i would build it as a bitcode archive.
stevenja_ has joined #pypy
<kristina>
that way it can be rolled into the monolithic binary as part of IPO.
stevenja_ has quit [Remote host closed the connection]
stevenja_ has joined #pypy
<kristina>
it can load shared objects from segments, that is mostly for LGPL compliance. the rule of thumb is, if it can be included in the IPO pass, build a bitcode archive, if it can't because of licensing (aka LGPL), build a stripped down SO, run IPO within it with an export list, roll it into a segment.
<kristina>
it does some bizzare stuff too to remove thunks when loading such libraries.