These are chat archives for dropbox/pyston

20th
Apr 2015
Michael Arntzenius
@rntz
Apr 20 2015 02:55
hm, perhaps they're pushing new objects on the front of the list, which puts pyston itself & libgcc and whatever other libraries we probably spend most of our time in at the back of the list?
that would produce pessimal behavior
A. F. Dudley
@AFDudley
Apr 20 2015 15:25
This project is very interesting to me, I am trying to understand how this is different from writing a pypy-llvm backend.
hmm, i found a blog post...
A. F. Dudley
@AFDudley
Apr 20 2015 15:30
I guess since 2.7 is being targeted pep 484 is out of the question? Does the type speculation system support type hinting anyway?
Chris Toshok
@toshok
Apr 20 2015 15:40
@rntz they do an insertion sort based on descending pc_begin. depending on how allocations happen that can definitely result in pushing everything to the front
@AFDudley pypy jits traces, an llvm backend would just change the way the code is generated
pyston jits an entire method at a time
Chris Toshok
@toshok
Apr 20 2015 18:53
weird, getting cmake build failures locally for pyston_release (granted, this is on my django-fixes branch):
/usr/bin/../lib/gcc/x86_64-linux-gnu/4.9/../../../../include/c++/4.9/cstddef:51:11: error: no member named 'max_align_t' in the global namespace
Daniel Agar
@dagar
Apr 20 2015 21:09
gcc 4.9 was working before?
Michael Arntzenius
@rntz
Apr 20 2015 22:03
hm, is there a good way to selectively override certain symbols, so that Pyston gets a different __cxa_begin_catch or _Unwind_Resume function?
Kevin Modzelewski
@kmod
Apr 20 2015 22:11
you might be able to just define those functions
and depending on the linkage they were built with the link will either work or not
ex that's how we override abort()
I'm not very familiar with how it all works but it's fairly easy to test :P
Chris Toshok
@toshok
Apr 20 2015 22:13
@undingen good call on std::lower_bound, looking at it now
Michael Arntzenius
@rntz
Apr 20 2015 22:16
@kmod Yeah, I figured that wouldn't work because it didn't seem likely that standard library functions that important would be weak symbols. But I was wrong!
Chris Toshok
@toshok
Apr 20 2015 22:39
really hope github isn’t being ddos’ed again
Chris Toshok
@toshok
Apr 20 2015 23:14
is closing/reopening a pull request an acceptable way to kick travis-ci?
Daniel Agar
@dagar
Apr 20 2015 23:20
yes
Chris Toshok
@toshok
Apr 20 2015 23:20
thanks
Daniel Agar
@dagar
Apr 20 2015 23:20
but if you have access to the repo I think you can restart it
Chris Toshok
@toshok
Apr 20 2015 23:21
yeah I’ve done that before, but have had issues with the github page not reflecting the state of the new build
Kevin Modzelewski
@kmod
Apr 20 2015 23:25
I think they do slightly different things --
rerunning the build in travis will rerun on the exact same code
but closing + reopening will remerge and run on that
I think if you're the PR author you can push -f new code and travis-ci will rebuild that
Michael Arntzenius
@rntz
Apr 20 2015 23:41
@kmod did we add some special new thing in the lz4 directory? I'm getting cmake errors related to it when I switch to master branch
Chris Toshok
@toshok
Apr 20 2015 23:41
you probably need git submodules init + update
i ran into that locally when I rebased
Michael Arntzenius
@rntz
Apr 20 2015 23:42
ah, that seems to have worked
I tried update, but not init. I should probably learn what git submodule's actual behavior is one of these days.
Chris Toshok
@toshok
Apr 20 2015 23:43
it’s easy - git submodule’s behavior is broken :)
Kevin Modzelewski
@kmod
Apr 20 2015 23:46
yeah it'd be nice if it just initialized it for you
Michael Arntzenius
@rntz
Apr 20 2015 23:51
@kmod hey, can we talk about how we store exc_info? it seems relevant to what I'm now doing (trying to get unwinding to work even through calls to _Unwind_Resume)