cfbolz changed the topic of #pypy to: #pypy PyPy, the flexible snake | IRC logs: and | Matti: I made a bit of progress, the tests now only segfault towards the end
reneeontheweb has quit [Quit: Client closed]
<mattip> builds did not trigger last night, even though there were changes that should have triggered builds
<mattip> I cannot log in to to see what is wrong
Guest96 has joined #pypy
jacob22 has quit [Ping timeout: 240 seconds]
jacob22 has joined #pypy
mgorny has quit [Quit: No Ping reply in 60 seconds.]
mgorny has joined #pypy
mattip has quit [Ping timeout: 260 seconds]
mattip has joined #pypy
<antocuni> cfbolz: if I remember correctly, at some point years ago you showed me a way to visualize vmprof data inside chrome web tools
<antocuni> does this thing still exist and works today?
<cfbolz> argh
<cfbolz> antocuni: that was a terrible hack
<cfbolz> I am not sure I can still find it
<cfbolz> antocuni: we need to do something about profiling though
<antocuni> yes
<cfbolz> and indeed I think a good approach is to leverage one of the existing UIs
<antocuni> a friend tells me that he has a python program that at some point slows down for few seconds and he wanted to use vmprof to investigate
<antocuni> but I don't think our existing tools are good for that use case
<antocuni> reusing existing UI: yes definitely, I think we clearly proved that we are not able to write UIs 😅
<cfbolz> I don't know that's the conclusion, more that maintaining them is work
<cfbolz> and we all like our text stuff too much ;-)
<antocuni> well, writting good UIs require a completely different skillset and attitude than writing JITs and VMs, so I'm not surprised that our UIs are not as good as our VMs
<cfbolz> anyway, the profiler space (even within python but even more so generally) is really exploding
<cfbolz> and many of the UIs take in general data that you can convert to
<cfbolz> but we would have to take the time and write the glue code necessary
<fijal> heh, it seems we have been ahead of the curve on both testing and profiling
<fijal> I remember concluding "no way you can make a commercial CI tool successful"
<fijal> very happy to be proven wrong, I guess
<antocuni> fijal: yes, I remember you saying that :)
<cfbolz> fijal: yes, but I think the commercial tools are very different than what we were thinking about. I certainly wasn't thinking of sentry
<fijal> oh what do you mean, you mean that I missed taking over the world by more than one simple step? ;-)
<cfbolz> yep
<fijal> blatant lies!
<fijal> cfbolz: yes, I think the main takeaway from vmprof is that people don't care about very precise measurments, they want *something* done in an easy way presented in a nice UI
<cfbolz> fijal: no, it's more extreme ;-)
<fijal> yes?>
<cfbolz> "we don't care about performance unless it's on thousands of servers and costs real money"
<fijal> yes, we kinda knew that
<fijal> ... by which time it's often too late to really do anything
<cfbolz> but we didn't build a profiler for that mindset ;-)
<fijal> we tried
<cfbolz> we certainly cannot profile 1000 servers
<fijal> we haven't built it partly because we don't have 1000 servers
<fijal> so it's a bit hard to think how you would do that
<fijal> I definitely thought *a lot* how to make vmprof usable in production
<fijal> (but never got around to it and never had a client willing to fund it either)
<fijal> as opposed to benchmarks
<fijal> but *also* it kinda generally does not work to profile functions/loops in this setup *really*
<fijal> cfbolz: interestingly enough, what it does convince me of is that "modeling in VR" is the future
<fijal> that maybe I should rely a bit more on intuition what feels like the right tool
<cfbolz> heh
<fijal> why "heh"?
<cfbolz> well, it's good
<fijal> remote work is another one of those....
<antocuni> I've used vmprof in production, for gambit. It justed worked well
<fijal> antocuni: we never had a good way to "Stream" results and combine them from multiple servers
<fijal> I nearly landed a contract to do that, but acquisitions etc.
<antocuni> true, that'a a different issue
lritter has joined #pypy
<cfbolz> (it's an open issue in general what a JIT looks like for such a setting)
<cfbolz> or an IDE
<fijal> yep
<fijal> I think it's an interesting topic :)
<fijal> but I'm bad at raising VC money ;-)
<fijal> ok, can I get an idiot check please?
<fijal> this should show 8, right?
<cfbolz> I don't know varargs, but I suppose so
<cfbolz> fijal: does it work on py3?
<mattip> works on linux w/python3
<fijal> no it does not
<fijal> I can try on older OS X too
<fijal> (well, not right now)
<fijal> I'll try cffi though
<fijal> mattip: I think also wiht py2
<mattip> doesn't work with arm64 on mac
<fijal> no
<fijal> but I'm failing to get cffi going at all, can't find the symbols
<mattip> n is passed in correctly, but the va_arg macro doesn't work properly
<fijal> are the symbols somewhat special with va_xxx?
<fijal> mattip: yes, it's a known problem, I think
slav0nic has joined #pypy
<fijal> arigato: ping?
<fijal> I'm really struggling with cffi I'm not sure what's the story (probably mac speciifc?)
<fijal> I don't understand how picking up the libraries really works (or how I can make it work)
<fijal> ok, so cffi works, ctypes does not, including both ctypes for python 2.7 and 43.9
<fijal> 3.9
<fijal> I think I'll try to compile 2.7 from scratch, using system libffi
Guest96 has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
indyZ has left #pypy [#pypy]
<fijal> this is madness
<fijal> mattip: I *think* this is a form of memory protection, but I'm not sure
Guest96 has joined #pypy
Guest96 has quit [Client Quit]
jacob22 has quit [Ping timeout: 248 seconds]
<mattip> fijal: maybe the buildbot machine does not have mercurial or the mercurial cannot work with heptapod,
<mattip> so it cannot track changes and trigger builds?
Guest96 has joined #pypy
jacob22 has joined #pypy
Guest96 has quit [Quit: Textual IRC Client:]
jacob22 has quit [Ping timeout: 256 seconds]
Corbin has quit [Quit: Corbin]
Corbin has joined #pypy
jacob22 has joined #pypy
reneeontheweb has joined #pypy
Techcable has quit [Ping timeout: 276 seconds]
Techcable has joined #pypy
lritter has quit [Ping timeout: 248 seconds]
<exarkun> anyone know if slipcover works on PyPy?
leshaste has joined #pypy
<leshaste> is it possible to install igraph for pypy?
<leshaste> I just tried and it says You will need the Python headers to compile this extension.
<nimaje> that sounds like the pypy package you installed doesn't contain header files and you need to install some pypy-headers or pypy-dev package with your os package manager or something
reneeontheweb has quit [Ping timeout: 252 seconds]
slav0nic has quit [Ping timeout: 248 seconds]