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
inhahe__ has quit [Read error: Connection reset by peer]
Hasimir has joined #pypy
lritter has joined #pypy
inhahe__ has joined #pypy
<kenaan>
arigo py3.5 7f62e74d2342 /: Merged in toumorokoshi/pypy/py3.5 (pull request #608) Adding detection of gcc to sysconfig_pypy (fixes #2809)
<kenaan>
Yusuke Tsutsumi py3.5 4c61881c4c06 /lib_pypy/_sysconfigdata.py: Adding detection of cc and gcc to build variables
<kenaan>
Yusuke Tsutsumi py3.5 b3a56b6a0940 /: Addressing Code Review Feedback Refactoring to move the init_posix code to a central location, inside of s...
<kenaan>
toumorokoshi py3.5 29a5d3ceb791 /: Close branch py3.5.
<kenaan>
Yusuke Tsutsumi py3.5 c1a718b1710d /lib_pypy/_sysconfigdata.py: minor change to recover branch.
<kenaan>
Yusuke Tsutsumi py3.5 019e1a27a95c /lib-python/3/test/test_sysconfig.py: Adding unit tests
<kenaan>
arigo py3.5 0811e990ecee /lib_pypy/_sysconfigdata.py: Remove these two entries because they appear also later. In the case of SHLIB_SUFFIX, it appears with a different val...
<arigato>
ronan: just a quick note, unicode-utf8 is meant to be merged into default and from there into py3.5. but that's the theory
<cfbolz>
arigato: do you have an opinion on the bytecode question m
<cfbolz>
?
<arigato>
not really
<arigato>
I fear there are various places where we assume something about the bytecode a bit implicitly
<arigato>
but even so, it's probably a good idea to mimic cpython 3.6's
<cfbolz>
Ok
<cfbolz>
arigato: yes, that's my opinion as well
<cfbolz>
arigato: I even volunteer to do it, I think
<arigato>
thanks :-)
<cfbolz>
arigato: I haven't worked on nice straightforward features in a while, I kind of forgot how relaxing that is
<arigato>
doesn't change much for the jit, but before that, the bytecode should probably be some kind of list of r_ushort
<cfbolz>
arigato: ah, why?
<cfbolz>
Right, I see
<cfbolz>
We don't want to load two chars
<arigato>
to avoid several operations just to read one of them
<arigato>
yes
<cfbolz>
arigato: what do we do for co_code then?
<arigato>
it's fine if we just build a string on demand
<cfbolz>
OK
<arigato>
(I think)
<cfbolz>
Think so too
<cfbolz>
arigato: I'm kind of curious whether it helps the interpreter at all
<arigato>
CPython got a very small number of percents
<arigato>
I'm personally skeptical :-)
<cfbolz>
arigato: so probably a little bit irrelevant
<arigato>
yes, I'd expect they would have gotten the same result by doing a hack that reads the 16-bit argument in a single go
<arigato>
because nowadays x86 cpus are good at non-aligned reads
<arigato>
but of course it's not clean because on other cpus you're not allowed to do non-aligned reads
<cfbolz>
arigato: makes sense
antocuni has joined #pypy
<arigato>
ah wait
<arigato>
no no, wait
jacob22__ has joined #pypy
<arigato>
it's done with 16-bit units, but one unit contains the opcode and the oparg
<arigato>
so really, it's two independent bytes again
<arigato>
hah, they even need to check if we're big-endian or little-endian to know how to extract the bytes
<arigato>
so in all cases, it's one byte containing the opcode, followed by one byte containing the oparg or 0
<arigato>
and the EXTENDED_ARG opcode can be used to add more than 8 bits to the oparg, but with 8 extra bits at a time
<arigato>
so I guess the main gain is more likely to be reduced memory usage
<arigato>
and it makes not much sense to use a list of r_ushort in pypy, a plain string would do
antocuni_ has joined #pypy
jacob22__ has quit [Ping timeout: 264 seconds]
antocuni has quit [Ping timeout: 264 seconds]
antocuni_ is now known as antocuni
<kenaan>
arigo default b929ba72ac2b /rpython/jit/metainterp/pyjitpl.py: Remove an unused line
<arigato>
...or no, memory usage is likely not really changed, and the main gain is only the removal of this unpredictable branch
<cfbolz>
arigato: I think I read somewhere that memory is a tiny bit better
<cfbolz>
But who knows
<mattip>
arigato: unicode-utf8 is mergable to default, but we held off to see if unicode-utf8 needed changes for python3
<arigato>
mattip: ok
<mattip>
so the plan at the sprint was to merge default into unicode-utf8, which is done, then merge unicode-utf8 into py3.5 via a new unicode-py3.5 branch
<mattip>
which it turns out is hard
<mattip>
especially since I don't really understand the underlying high-level design wrt rpython/rlib, interpreter/unicodehelper and objspace/std/unicodeobject on the one hand,
<mattip>
and how surrogates are meant to work on the other hand
<kenaan>
arigo reverse-debugger-updated c8c3a21ba942 /rpython/rtyper/lltypesystem/: in-progress (not working so far): trying to get get_raw_address_of_string() to never raise ValueError
lesshaste has quit [Quit: Leaving]
<kenaan>
arigo reverse-debugger-updated 3b526d4fd961 /rpython/rtyper/lltypesystem/: Test was already passing, just confused about debug_print().
<kenaan>
arigo reverse-debugger-updated 265844e2ae4e /rpython/rtyper/lltypesystem/test/test_rffi.py: improve the test
forgottenone has quit [Quit: Konversation terminated!]