00:38
gef has quit [Ping timeout: 240 seconds]
00:38
gef has joined #pypy
01:09
gef has quit [Ping timeout: 252 seconds]
01:10
gef has joined #pypy
01:15
gef has quit [Ping timeout: 260 seconds]
01:17
gef has joined #pypy
01:23
gef has quit [Ping timeout: 240 seconds]
01:40
gef has joined #pypy
01:49
gef has quit [Ping timeout: 252 seconds]
02:09
jcea has quit [Ping timeout: 250 seconds]
02:23
gef has joined #pypy
03:14
gef has quit [Ping timeout: 260 seconds]
03:15
gef has joined #pypy
03:15
proteusguy has quit [Remote host closed the connection]
03:22
gef has quit [Ping timeout: 240 seconds]
03:22
gef has joined #pypy
03:32
oberstet has quit [Remote host closed the connection]
03:37
gef has quit [Ping timeout: 252 seconds]
03:38
gef has joined #pypy
03:53
gef has quit [Ping timeout: 252 seconds]
04:01
gef has joined #pypy
04:06
gef has quit [Ping timeout: 252 seconds]
04:20
<
cfbolz >
Bit worried by the label of their numbers: cinder_jit_noframe
04:20
<
cfbolz >
Ie they cheat?
04:34
gef has joined #pypy
04:43
<
mattip >
the "static python" seems like where the typing effort will eventually end up
04:51
<
mattip >
locally, when testing win64 on numpy, something strange.
04:52
<
mattip >
int_val = 2 ** 1024; for i in range(1000): np.longdouble(int_val)
04:52
<
mattip >
correctly returns "inf" for the first 750 iterations, but then returns "0"
04:55
<
mattip >
the code path dives into C, converts int_val into bytes and calls out to cpyext for PyOS_string_to_double()
04:56
<
mattip >
PyOS_string_to_double is marked as jit.dont_look_inside
04:58
<
mattip >
it is definitely some interaction with the JIT, when running with pypy --jit off the problem does not appear
05:41
<
cfbolz >
mattip: ouch
05:41
<
cfbolz >
mattip: on Linux 64 it's fine?
05:42
<
mattip >
linux64 uses a different code path and does not hit PyOS_string_to_double, so maybe one thing to try is to test that function on linux64
05:43
<
mattip >
another strange windows (maybe JIT?) problem: when running the whole numpy test suite at some point it segfaults
05:43
<
mattip >
but rerunning just that test passes
05:44
* mattip
rerunning the whole suite with "--jit off"
05:49
gef has quit [Ping timeout: 240 seconds]
05:51
<
mattip >
the PyOS_string_to_double seems tricky, if I add a printf of the return value just after the call (in C), the problem disappears
05:54
gef has joined #pypy
05:55
todda7 has joined #pypy
05:57
<
cfbolz >
mattip: hm, do you want to send me the PYPYLOG=jit:out output?
06:02
<
mattip >
let me try
06:07
todda7 has quit [Ping timeout: 240 seconds]
06:17
<
mattip >
and poof: I can no longer recreate the problem
06:59
<
cfbolz >
mattip: ppfff
07:00
<
mattip >
it reminds me of my attempts to get windconsoleio working: after a few hundred syscalls, something messed up
07:04
gef has quit [Ping timeout: 240 seconds]
07:07
gef has joined #pypy
07:30
isidentical has joined #pypy
07:41
oberstet has joined #pypy
07:42
gef has quit [Ping timeout: 240 seconds]
07:43
<
mattip >
When I see mojobake I feel like I do when reading pascal code: I learned all that once and am happy to forget it.
07:43
jacob22 has quit [Quit: Konversation terminated!]
07:44
jacob22 has joined #pypy
07:47
<
antocuni >
at first I read "fixes TESTS for you". This would be even cooler :)
07:56
<
LarstiQ >
as a more sophisticated measure of coverage than lines executed
07:56
gef has joined #pypy
07:58
<
antocuni >
it's an interesting area
08:12
<
cfbolz >
there is a python versfion
08:12
<
cfbolz >
been meaning to try it
08:56
glyph has quit [Read error: Connection reset by peer]
08:56
glyph has joined #pypy
09:06
Gustavo6046 has quit [Ping timeout: 260 seconds]
09:34
jacob22 has quit [Quit: Konversation terminated!]
09:39
jacob22 has joined #pypy
10:16
gef has quit [Ping timeout: 252 seconds]
10:17
gef has joined #pypy
10:29
<
mattip >
in the hacker news discussion about cinder there is a comment about PyPy "we really tried, six month project just to get it to run out codebase ..."
10:30
<
mattip >
I wonder whether it didn't run, or ran and didn't show improvement
10:30
<
cfbolz >
mattip: yeah, saw that too
10:30
<
cfbolz >
want me to try and find out?
10:30
<
cfbolz >
mattip: all in all I'm not too impressed though. this "throw over the wall" approach with lots of language changes, not sure it will get picked up
10:33
<
cfbolz >
I really don't get these big companies that really depend on python. somehow it should be strategically important for them to improve cpython/the ecosystem. they have the cash! but even the very biggest ones do very little for the core VM (few exceptions exist of course, MS, Red Hat?)
10:37
gef has quit [Ping timeout: 252 seconds]
10:38
Dejan has joined #pypy
10:38
Dejan has joined #pypy
10:38
Dejan has quit [Changing host]
10:40
lritter_ has joined #pypy
10:47
<
mattip >
Microsoft and Google are now "Visionary Sponsors", and the PSF is investing in hiring here and there
10:48
<
mattip >
and there are people who get eqivalent time to contribute, like vstinner and others
10:50
lritter_ has quit [Remote host closed the connection]
10:50
<
mattip >
cfbolz: any thoughts about the jit-instance-ptr-eq-improvements branch? It would be a shame to let it get stale
10:51
<
mattip >
and py3.7-newtext-const-arg-caching looks promising too
10:53
<
cfbolz >
mattip: sorry, yeah, I'm a bit distracted with teaching
10:53
infernix has quit [Ping timeout: 248 seconds]
10:53
<
mattip >
ahh, cool. I didn't realize you are back to teaching
10:53
<
cfbolz >
mattip: both are "ready", instance-ptr-eq had a relatively good test run
10:54
gef has joined #pypy
10:54
<
cfbolz >
I fixed the test_pypy_c one
10:55
<
mattip >
let's see :)
11:01
<
cfbolz >
mattip: thanks for looking.
11:01
<
cfbolz >
mattip: newtext gives ~5% on some interpreter only benchmarks
11:02
<
cfbolz >
and sorry, as usual I am better at starting things than at finishing them ;-)
11:05
infernix has joined #pypy
11:21
gef_ has joined #pypy
11:24
gef has quit [Ping timeout: 265 seconds]
11:54
gef has joined #pypy
11:56
gef_ has quit [Ping timeout: 246 seconds]
12:14
proteusguy has joined #pypy
12:24
jcea has joined #pypy
12:42
<
cfbolz >
mattip: hm, the test_verify1 is weird
12:52
jcea has quit [Ping timeout: 250 seconds]
13:21
<
mattip >
something about guard values and registers
13:21
<
mattip >
is there such a thing as too many guards?
13:23
<
cfbolz >
mattip: where do you see something about guard values?
13:24
<
mattip >
in the 32-bit stdout
13:25
<
cfbolz >
ugh, even weirder 😅
13:38
<
mattip >
the cinder guy answered on the link from above: they tried it a few years ago
13:39
<
mattip >
hmm, jitting + threading means new jit code for each thread, which eats memory
13:39
<
cfbolz >
mattip: ouch, really? the machine code should be sahred
13:51
<
cfbolz >
of course test_verify passes locally
14:12
<
mattip >
the comments about their experience seem to be second hand
15:44
isidentical has quit [Quit: isidentical]
16:06
Gustavo6046 has joined #pypy
16:18
Gustavo6046_ has joined #pypy
16:18
Gustavo6046 has quit [Ping timeout: 260 seconds]
16:18
muke has joined #pypy
16:20
<
muke >
Hey cfbolz, I'm having some trouble with the test for the linear float loop - the IR compiles and executes but only seems to return it's argument, but I can't see any reason that would be from looking at the IR
16:20
Gustavo6046_ is now known as Gustavo6046
16:20
<
muke >
everything else is the same as with the regular linear loop, is there anything that comes to mind as to why floats would be different here?
16:51
oberstet has quit [Quit: Leaving]
17:10
isidentical has joined #pypy
17:22
jcea has joined #pypy
18:08
todda7 has joined #pypy
18:34
<
krono >
say I want hg-up my repo to the commit of the latest py2.7 release, which would that be/
18:55
Gustavo6046 has quit [Ping timeout: 240 seconds]
18:59
Gustavo6046 has joined #pypy
19:05
gef has quit [Ping timeout: 240 seconds]
19:13
gef has joined #pypy
19:36
muke has quit [Remote host closed the connection]
19:43
isidentical has quit [Remote host closed the connection]
19:58
muke has joined #pypy
20:17
<
cfbolz >
muke: sorry, not around this evening, will try to look tomorrow
20:18
<
muke >
no worries, there's no rush for this
20:25
isidentical has joined #pypy
21:05
dnshane has joined #pypy
21:11
jcea1 has joined #pypy
21:11
jcea has quit [Quit: jcea]
21:11
jcea1 is now known as jcea
21:17
jcea has quit [Quit: jcea]
21:17
muke has quit [Quit: Leaving]
21:18
jcea has joined #pypy
22:28
isidentical has quit [Quit: isidentical]
22:35
Gustavo6046 has quit [Ping timeout: 250 seconds]
22:48
Gustavo6046 has joined #pypy