#pypy IRC log for Friday, 2011-03-18

etrepum (~bob@12.130.117.37) joined #pypy.00:00
lacMax lager is very loud.  what aboiut the rest?00:03
voidspacelac: have you been to the Mary Mac's tea room?00:03
lacno.00:03
hpk_ (~hpk@HSI-KBW-109-193-198-175.hsi7.kabel-badenwuerttemberg.de) left irc: Ping timeout: 260 seconds00:03
voidspaceit's southern food - country fried steak, meat loaf, okra, chicken and dumplings, cornbread00:03
derdon (~derdon@p54A6ECA2.dip.t-dialin.net) left irc: Ping timeout: 252 seconds00:03
arigato (~arigo@63.133.138.10) joined #pypy.00:03
voidspacefairly "heavy" but very lovely 00:04
voidspacehttp://www.marymacs.com/00:04
voidspaceit is a 15 minute walk or two marta stops00:05
voidspacethere is a very nice sushi place nearly next door to it as well00:05
voidspaceand there is always the chinese we went to on the very first day00:05
voidspaceany of them would be great for me :-)00:05
lacok, let us do MaryMacs.00:07
lacmeet when, where?00:07
nyxtom (~nyxtom@38.108.47.220) left irc: Quit: Computer has gone to sleep.00:07
lachmm, actually, its a bad fit for those who do not like fried, correct?00:08
voidspacelac: right00:08
lacthen I think I would rather eat someplace else.00:08
voidspaceis jacob22_ not a fan of fried?00:08
voidspaceoh, ok00:08
lacI'm the one who isn't00:08
voidspacethey have some roast meat or seafood dishes - but mostly fried00:09
voidspaceor a lot of fried anyway00:09
voidspacethe menu is here: http://www.marymacs.com/Portals/0/menus/marymacs_menu_dinner.pdf00:09
voidspaceor we choose somewhere else :-)00:10
AdemanI don't think I have hg working properly :-/00:10
jacob22__How about Nan Thai Fine dining? 3 metro stops away. A little pricy but rates a 27 for food in the Zagat.00:13
kenaan12ademan micronumpy 11fbb12ddc3b65 15/: Branch suffered from bad svn merge, micronumpy as it is here is mostly deprecated anyways.00:15
voidspacejacob22_: fine with me...00:16
arigato (~arigo@63.133.138.10) left irc: Ping timeout: 252 seconds00:21
htoothrotdid lamback come through with the windows builds or do you still need someone to build nightlies?00:27
nettok (~quassel@200.119.156.136) joined #pypy.00:28
mattheww (~mjw@cpc3-cmbg10-0-0-cust455.5-4.cable.virginmedia.com) left irc: Quit: Leaving00:28
alfredodeza (~alfredode@c-24-131-47-167.hsd1.ga.comcast.net) joined #pypy.00:31
alfredodezahey guys, I am having issues with the ``codecs`` module00:32
alfredodezait simply doesn't work under 1.4.1 :(00:32
alfredodezais this a known issue? 00:33
alfredodezaI found that fijal reported this was broken: https://codespeak.net/issue/pypy-dev/issue40200:33
nettok_ (~quassel@200.119.188.207) joined #pypy.00:35
nettok (~quassel@200.119.156.136) left irc: Ping timeout: 264 seconds00:36
Alex_Gaynor (~alex@76-206-244-32.lightspeed.cicril.sbcglobal.net) joined #pypy.00:40
fijalalfredodeza: hi00:43
mcfletch (~mcfletch@CPE0014bf07ffd2-CM001ac30d4aca.cpe.net.cable.rogers.com) left irc: Read error: Connection reset by peer00:43
fijalAlex_Gaynor: hi00:43
alfredodezahi fijal00:43
fijalalfredodeza: I think this is fixed on trunk00:43
fijalwhat sort of issue you have with codecs?00:44
alfredodezafijal: oh, ok, so this would be in trunk but not in 1.4.1?00:44
fijalI don't know00:44
fijalit really depends what sort of issue you have00:44
alfredodezafijal: basically the codecs.register(search_function) does not work at all00:44
fijalgreat00:45
alfredodezathe search function *translates* the encoding and it works perfectly under python2.5 and 2.600:45
fijalwell, it's really annoying00:45
fijalbecause cpython tests pass00:45
alfredodezait does not raise an exception in pypy at all, it just doesn't do any translation00:46
fijaldo you feel like coming up with a minimal program that works differently under pypy and under cpython00:46
fijaland putting it in the issue00:46
Alex_Gaynorfijal: hi00:46
fijalcan be the same issue for all I care00:46
alfredodezafijal: so this all stems from a package that I have built, I am not sure I could reproduce it in just a few lines of code00:47
alfredodezaif it doesn't involve my module00:47
fijalalfredodeza: the smaller the better00:47
fijalpackage is also fine00:47
fijalbut less so00:48
fijal(it does take more time to fix on average)00:48
alfredodezafijal: right. Well I will try to reproduce with as little code as I can and add to the original issue that was reported00:48
alfredodezafijal: I guess I need to be registered to submit issues right?00:49
alfredodezafijal: do I need special permissions to re-open closed tickets?00:50
alfredodeza(your issue #402 is closed)00:50
fijalnope00:51
fijalas in no special permissions00:52
fijalyou can create a new one as well00:53
fijalwhatever is easier00:53
alfredodezafijal: ok thanks! Just registered but hopefully it will not take me too long to get some examples posted00:54
mcfletch (~mcfletch@CPE0014bf07ffd2-CM001ac30d4aca.cpe.net.cable.rogers.com) joined #pypy.00:58
__name__ (~name@sburn/devel/name) left irc: Remote host closed the connection01:09
mcfletch (~mcfletch@CPE0014bf07ffd2-CM001ac30d4aca.cpe.net.cable.rogers.com) left irc: Read error: Connection reset by peer01:13
santagada (~santagada@187.113.192.150) joined #pypy.01:16
santagad_ (~santagada@187.113.192.150) joined #pypy.01:18
Nick change: santagad_ -> santagada_01:19
santagada (~santagada@187.113.192.150) left irc: Read error: Connection reset by peer01:19
mcfletch (~mcfletch@CPE0014bf07ffd2-CM001ac30d4aca.cpe.net.cable.rogers.com) joined #pypy.01:29
CIA-403alfredodeza 07roundup * 10#402/codecs are broken in pypy-c: [in-progress] The ``codecs`` module in pypy is behaving differently than CPython 2.5 and 2.6. The issue comes when you want to register a custom ... * 14https://codespeak.net/issue/pypy-dev/issue40201:30
alfredodezafijal: I was able to replicate the problem with as little code as possible in 2 files01:31
alfredodezafijal: hope that helps01:31
alfredodeza (~alfredode@c-24-131-47-167.hsd1.ga.comcast.net) left irc: Remote host closed the connection01:34
mcfletch (~mcfletch@CPE0014bf07ffd2-CM001ac30d4aca.cpe.net.cable.rogers.com) left irc: Read error: Connection reset by peer01:44
santagada (~santagada@187-7-73-104.ggs5102.3g.brasiltelecom.net.br) joined #pypy.01:49
santagada_ (~santagada@187.113.192.150) left irc: Ping timeout: 255 seconds01:52
fenrrir (~fenrrir@187.40.156.195) left irc: Ping timeout: 255 seconds01:57
santagada (~santagada@187-7-73-104.ggs5102.3g.brasiltelecom.net.br) left irc: Read error: Connection reset by peer02:00
santagada (~santagada@187-7-73-104.ggs5102.3g.brasiltelecom.net.br) joined #pypy.02:00
mcfletch (~mcfletch@CPE0014bf07ffd2-CM001ac30d4aca.cpe.net.cable.rogers.com) joined #pypy.02:00
Shinato (~John@osbk-4db1536e.pool.mediaWays.net) joined #pypy.02:06
Kaskuka (~John@osbk-4d0876b9.pool.mediaWays.net) left irc: Ping timeout: 248 seconds02:06
squiddy (~squiddy@x027.wh17.tu-dresden.de) left irc: Quit: Leaving02:12
lambacckin an earlier question it sounded like I could translate just the cpyext (and not the whole interpreter), how would I do that (if it is possible)?02:14
mcfletch (~mcfletch@CPE0014bf07ffd2-CM001ac30d4aca.cpe.net.cable.rogers.com) left irc: Read error: Connection reset by peer02:14
DasIch_ (~DasIch@p4FFDE340.dip.t-dialin.net) joined #pypy.02:20
etrepum (~bob@12.130.117.37) left irc: Quit: etrepum02:21
DasIch (~DasIch@p4FFDEF87.dip.t-dialin.net) left irc: Ping timeout: 260 seconds02:24
santagada (~santagada@187-7-73-104.ggs5102.3g.brasiltelecom.net.br) left irc: Remote host closed the connection02:27
mcfletch (~mcfletch@CPE0014bf07ffd2-CM001ac30d4aca.cpe.net.cable.rogers.com) joined #pypy.02:33
mcfletch (~mcfletch@CPE0014bf07ffd2-CM001ac30d4aca.cpe.net.cable.rogers.com) left irc: Read error: Connection reset by peer02:44
fijallambacck: you can't02:59
Alex_Gaynorlambacck: if you're developing it you can run tests though03:01
mcdonc (~mcdonc@ip72-209-213-54.dc.dc.cox.net) joined #pypy.03:02
mcfletch (~mcfletch@CPE0014bf07ffd2-CM001ac30d4aca.cpe.net.cable.rogers.com) joined #pypy.03:02
Alex_Gaynorfijal: apparently some twisted test is hanging forever, any idea what's up with that?03:04
exarkuntwisted.test.test_threadable.SynchronizationTestCase.testThreadedSynchronization03:04
Action: exarkun just failed to reproduce it on his laptop, though03:04
Alex_Gaynorsweeet03:04
exarkunit might be "hard to debug" failure not "it got fixed somewhere" failure03:05
Alex_Gaynorof course :)03:05
Action: exarkun sees some other failures though03:07
exarkunhow old is 42589?03:07
exarkunoh right, the test run finished, it's _good_ to see other failures, I forgot :)03:08
lac htoothrot: another windows build would be great03:09
kenneth_reitz (~kenneth_r@c-24-127-96-129.hsd1.va.comcast.net) joined #pypy.03:11
htoothrotis there some existing buildbot config that I can copy/look at?03:14
Alex_Gaynorfijal: why are benchmarks running twice tonight?03:14
lacAlex_Gaynor: today you mean? because they didn't run last night.  So Armin restarted them when he woke up.  Or are they running three times?03:16
Alex_Gaynorlac: they're runing 2x consecutively this evening, the second of which is overlappying with other builds so it iwll get nonsense results03:17
lacarmin jacob and I have nothing to do with this, we were just at dinner.  I think its an error.03:17
laccan you stop things?03:17
Alex_GaynorI can abort it from the web UI03:17
mcfletch (~mcfletch@CPE0014bf07ffd2-CM001ac30d4aca.cpe.net.cable.rogers.com) left irc: Quit: Leaving.03:18
Alex_Gaynornvm it just finished running03:18
Alex_GaynorI suppose its fine03:18
lachtoothrot: amaury_ would know where one was.  I don't.  Maybe Alex or fijal does?03:19
Alex_Gaynorhtoothrot: there's a repo on bitbucket, not sure if it contains the client configs03:19
arigato (~arigo@63.133.138.10) joined #pypy.03:25
htoothrotyeah, poked around there, but i don't see any of the slave configs, i'll check with amaury tomorrow, I won't be able to set it up until this weekend anyways03:29
nettok_ (~quassel@200.119.188.207) left irc: Ping timeout: 250 seconds03:32
Moku (~John@osbk-4db15276.pool.mediaWays.net) joined #pypy.03:47
Trundle (~andy@python/site-packages/trundle) left irc: Remote host closed the connection03:47
lachtoothrot: thank you very much.03:48
Shinato (~John@osbk-4db1536e.pool.mediaWays.net) left irc: Ping timeout: 248 seconds03:48
davisagli (~davisagli@davisagli.com) left irc: Excess Flood03:58
davisagli (~davisagli@davisagli.com) joined #pypy.03:58
lac (~quassel@63.133.138.10) left irc: Read error: Connection reset by peer04:04
lac (~quassel@63.133.138.10) joined #pypy.04:05
ElectronicRU (~Pythonist@5f4489b5.dynamic.mv.ru) left irc: Read error: Operation timed out04:05
s00fla (~root@li264-16.members.linode.com) joined #pypy.04:14
Alex_Gaynorlac: you guys broke the tests, badly: http://buildbot.pypy.org/summary?branch=%3Ctrunk%3E04:15
Alex_GaynorIt may make sense to replace all those tests with a sphinx build though04:15
Alex_Gaynor    sphinx-build -W -b html -d {envtmpdir}/doctrees . {envtmpdir}/html04:16
Alex_Gaynor    sphinx-build -W -b linkcheck -d {envtmpdir}/doctrees . {envtmpdir}/linkcheck04:16
Alex_Gaynoris a pretty good setup for getting any issues04:16
lacwell, since we renamed all the files, it would make sense they all broke.04:19
Alex_Gaynorthat fine, but the then the tests should be opdated :)04:22
daniloaf (~daniloaf@189.71.125.186) left irc: Remote host closed the connection04:22
lacyes, I forgot about them.  I'm sorry.04:22
daniloaf (~daniloaf@189.71.125.186) joined #pypy.04:22
Moku (~John@osbk-4db15276.pool.mediaWays.net) left irc: Ping timeout: 248 seconds04:24
Moku (~John@osbk-4db15795.pool.mediaWays.net) joined #pypy.04:24
fijal (~fijal@71-208-188-71.hlrn.qwest.net) left irc: Ping timeout: 240 seconds04:25
lacyes, config/test_makerestdoc.py is all about how we aren't generating the docs now04:27
Alex_Gaynor:)04:28
brad_smith (~brad_smit@ip68-98-36-157.ph.ph.cox.net) joined #pypy.04:28
lacand config/test_pypyoption.py  has a separate bit for testing the existence of doc files, and I wasn't even aware of this.04:30
lacOk, something to do tomorrow.  right now I am too fried.  but it should be quite easy to make the errors go away.  deciding how we really want to test the new sphinx stuff will take more thought and consensus building.04:31
exarkundunno if this will make any difference, but Twisted is about to switch to sphinx and someone is going to be writing a bunch of unit tests for automating sphinx doc building for our release process04:34
fijal (~fijal@71-208-182-156.hlrn.qwest.net) joined #pypy.04:38
Alex_Gaynorexarkun: my 2 lines are pretty effective :)04:38
ElectronicRU (~Pythonist@5f4489b5.dynamic.mv.ru) joined #pypy.04:38
exarkunAlex_Gaynor: How can I be sure!  Where is the unit test!04:39
exarkun:)04:39
Alex_Gaynorexarkun: subprocess.Popen or somehthing :)04:40
exarkunI don't really know what kind of thing lac was thinking of, just thought I'd throw the coincidence out there04:40
Alex_Gaynorfor comparison Django doesn' unit test our docs04:40
Hodgestar (~Hodgestar@gluon.za.net) left irc: Ping timeout: 250 seconds04:43
arigato (~arigo@63.133.138.10) left irc: Quit: See you04:44
Hodgestar (~Hodgestar@gluon.za.net) joined #pypy.04:45
voidspace (~voidspace@python/psf/voidspace) left irc: Quit: voidspace04:46
pingvenoI've tracking the status of the PEP to add in a /usr/bin/python2 symlink for CPython. Will Pypy add in something like that?04:53
Alex_Gaynorpingveno: we don't presently have an install script, so I don't know04:55
pingvenoAh, so there really isn't a standard executable name to begin with?04:56
exarkunthe standard name is pypy/translator/goal/pypy-c ;)04:57
Alex_Gaynorexarkun: in the release builds its just pypy04:58
pingvenoYeah, that's what I was thinking of when I said no standard name.05:00
pingvenoIn the future, perhaps add in a pypy2?05:00
daniloaf (~daniloaf@189.71.125.186) left irc: Quit: Saindo05:02
daniloaf (~daniloaf@189.71.125.186) joined #pypy.05:03
Alex_GaynorATM there isn't any ambiguity05:03
Alex_Gaynorpingveno: of more value, IMO, would be asking the virtualenv maintainers to have it add python205:04
exarkunI think a motivation of the PEP is that the python2 name should be available so that people can start using it05:04
exarkunSo that by the time there is ambiguity you have already avoided it.05:04
pingvenoThat's what I was pondering with adding a pypy2, even before PyPy implements Python 3.05:06
Shanita (~John@osbk-4db157b4.pool.mediaWays.net) joined #pypy.05:07
Moku (~John@osbk-4db15795.pool.mediaWays.net) left irc: Ping timeout: 248 seconds05:07
pingvenoAllow people to have a shebang that dependably points to PyPy implementing Python 2.05:07
exarkunPyPy could just as easily say that it will always call the Python 2 executable "pypy" and the Python 3 executable "pypy3"05:08
pingvenoIt does have the luxury of not already having Arch screwing around with it :P05:09
jacob22__ (~jacob@63.133.138.10) left irc: Ping timeout: 250 seconds05:09
Alex_Gaynorexarkun: I do believe that's what cpython said :)05:11
exarkunAlex_Gaynor: Clearly not05:11
Alex_Gaynorfijal: what's "abort bad loop"?05:12
pingvenoWasn't the problem more that there wasn't an official policy to start with, so people did whatever they felt like?05:13
pingvenoTheir gut feeling!05:14
kenneth_reitz (~kenneth_r@c-24-127-96-129.hsd1.va.comcast.net) left irc: Quit: Computer has gone to sleep.05:18
Kaskuka (~John@osbk-4d08b10d.pool.mediaWays.net) joined #pypy.05:20
Shanita (~John@osbk-4db157b4.pool.mediaWays.net) left irc: Ping timeout: 248 seconds05:21
daniloaf (~daniloaf@189.71.125.186) left irc: Quit: Saindo05:25
etrepum (~bob@c-67-180-34-190.hsd1.ca.comcast.net) joined #pypy.05:34
cco3 (~conley@c-69-181-140-72.hsd1.ca.comcast.net) left irc: Ping timeout: 250 seconds05:45
nyxtom (~nyxtom@67.205.234.176) joined #pypy.05:46
carljm (~carljm@spinach.threepines.org) left irc: Quit: leaving05:49
carljm (~carljm@spinach.threepines.org) joined #pypy.05:50
Alex_Gaynorfijal: what are the semantics of int_divide(i0, i1) where i1 is 0?06:12
Shinato (~John@osbk-4d08b10d.pool.mediaWays.net) joined #pypy.06:33
Kaskuka (~John@osbk-4d08b10d.pool.mediaWays.net) left irc: Ping timeout: 248 seconds06:33
hakanardo_antocuni: yes the test_array_sum* tests check the interaction between the jit and the array06:40
hakanardo_different array types require different support in the jit backend06:40
hakanardo_I remember having problem with 'f'-arrays (32 bit floats) for example, that armin fixed in the jit backend if I remember correctly...06:41
hakanardo_so I think we need some tests that verifies that the jit support all the operations needed by al types of arrays 06:42
Alex_Gaynorhakanardo: eh, FWIW every time I compile (on 32-bit) I get a bunch of messages about the JIT not supporting single floats06:44
hakanardo_ok, I don't know but I would guess there is just enough support to make the array work06:46
Action: Alex_Gaynor shrugs06:47
hakanardo_i.e maybe something getitem and setitem support but no arithmetics?06:47
Alex_GaynorI thought I saw a messages about steitem not getting jitcode, but maybe not06:47
Ademancolor me annoyed, I have three failures in pypy/jit/metainterp/test with my code disabled, were those tests broken in default recently?06:52
Ademaner, two of those tests are mine, one is in test_virtual06:53
Alex_GaynorAdeman: metainterp/test_virtual is clena for me06:59
AdemanAlex_Gaynor: yeah for me too currently, I'm wondering about around the time I branched though. I've probably broken something obscurely, but it'd have to be a *very* obscure break to affect tests when I've cut out my optimizeopt07:00
Alex_Gaynorshrug, merge in default07:01
AdemanAlex_Gaynor: I've been scared of merging since before the hg transition. hg merge -r default  ?07:02
nekto0n_ (c397cf21@gateway/web/freenode/ip.195.151.207.33) joined #pypy.07:03
Alex_Gaynorhg merge default I think07:03
antocuni (~antocuni@host17-122-dynamic.0-79-r.retail.telecomitalia.it) joined #pypy.07:08
nekto0n_hi there! I'm trying to build cx_Oracle on Win Xp machine, and setup.py says "error: Unable to find vcvarsall.bat"07:21
panni_ (hannes@ip-178-203-73-7.unitymediagroup.de) left irc: Read error: Connection reset by peer07:21
Alex_Gaynornekto0n_: there's actually an RPython cx_Oracle equivilant, it may be easier to build pypy using that then to try to use the C version07:23
nekto0n_Alex_Gaynor: if cx_Oracle is written in RPython, so it must be already in the binary?07:25
Alex_Gaynornekto0n_: yes, RPython modules must be included at translation time, however oracle is not included by default07:25
nekto0n_Alex_Gaynor: Oh... I'm afraid pypy is too memory hungry, I'm not sure if I can translate it. What about C version? :)07:27
Alex_Gaynornekto0n_: not sure, when amaury gets on you might ask him if he has a binary included cx_oracle (he wrote it)07:27
nekto0n_Alex_Gaynor: what is wrong with "vcvarsall.bat", I think there's a problem with %PATH% or something, can someone look at it? I will provide all info you need07:30
Alex_Gaynornekto0n_: I can't help with this, sorry, I know nothing about oracle or windows07:32
nekto0n_Alex_Gaynor: no no no, I think it's pypy who complains about vcvarsall.bat, cause I could find that string in cx_Oracle sources07:34
Alex_GaynorI'm falling back to "I know nothing about windows" then :/07:34
nekto0n_Alex_Gaynor: oh, I accept this :)07:35
Alex_Gaynornekto0n_: :)07:36
antocuninekto0n_: AFAIK, vcvars.bat is distributed with visual studio07:39
antocuninot sure how it is supposed to be found by the installation script07:40
Alex_Gaynorantocuni: E       OSError: [Errno 24] Too many open files; when running the new pypy_c tests07:40
antocuniuh?07:40
Action: antocuni looks07:40
antocunino clue about it07:43
nekto0n_antocuni: I've got this script, even appended it to the path, but no luck07:44
antocuninekto0n_: no clue then, sorry (I'm in the same corner as alex -- "know nothing about windows")07:44
Alex_Gaynorantocuni: can you reproduce the error?07:44
antocuniAlex_Gaynor: the only file that I open is enclosed by a with statement07:44
Action: antocuni downloads the last nightly07:45
Alex_Gaynorantocuni: could it be the subprocess?07:46
antocunimaybe07:46
antocunibut, in that case, it means that subprocess needs to be fixed probably07:46
antocunibecause it leaks07:46
nyxtom (~nyxtom@67.205.234.176) left irc: Ping timeout: 240 seconds07:47
antocuninekto0n_: I can build a pypy with cx_oracle if you want07:48
antocunibut only for linux07:48
kkris (~kris@93-82-42-106.adsl.highway.telekom.at) joined #pypy.07:48
nekto0n_antocuni: that would be nice!07:48
Alex_Gaynorantocuni: could this gc_was_enabled stuff prevetn desctructors closing files in pypy?07:49
antocuniAlex_Gaynor: yesish. IIRC, gc.disable in pypy disables destructors07:50
Alex_GaynorI wonder what the reason for that is, seems a bit pointless07:50
antocuniAlex_Gaynor: but anyway it seems that subprocess always reset gc.enabled, so I'm not sure it's an issue07:51
antocuniand the test works fine on my machine07:51
Alex_Gaynor:/07:51
antocuninekto0n_: do you want linux 32 bit or linux 64 bit?07:51
nekto0n_nekto0n_: 3207:52
Alex_Gaynorantocuni: how many tests fail for you?07:52
antocuniAlex_Gaynor: on test_pypy_c_new? They all pass07:52
antocuni(on 64 bit07:52
antocuni)07:52
Alex_Gaynorantocuni: ah wonderful :)  there are failures for me on 32 bit07:53
antocunihow many? On buildbot, there is one07:53
antocuni(other than this too many files)07:53
Alex_Gaynorah I should mention that my trunk is currently modified07:53
Alex_GaynorI was trying to optimize func_defaults07:53
nekto0n_btw, not so long ago (about a month) I tried to translate pypy into binary on Ubuntu x64 and 4Gb of RAM, almost succeded, but when gcc started - my memory was exhausted, can pypy release memory after C code has been generated?07:53
Alex_Gaynorantocuni: http://paste.pocoo.org/show/355429/07:54
MostAwesomeDudenekto0n_: It used to be that you could go into the temp folder and run make yourself; dunno if that's still possible.07:54
antocuninekto0n_: translation on 64 bit with 4GB of RAM should "just work"07:55
nekto0n_MostAwesomeDude: ah, sure, almost forgot about that. But still, is it rational to keep all that memory?07:56
nekto0n_antocuni: you mean "with 4Gb and swap enabled"? :)07:56
MostAwesomeDudenekto0n_: Dunno how it's gonna release all that RAM, exactly.07:56
antocuninekto0n_: yes, I fear you need swap, even if then you don't actually use it07:56
antocunibecause fork() does copy-on-write07:57
antocuniAlex_Gaynor: what's that?07:57
Alex_Gaynorantocuni: patch i wrote/am testing07:57
Alex_Gaynorantocuni: the idea is to turn Function.w_defaults into a W_TupleObject so the JIT always sees the array as being constant07:57
antocuniI see07:58
antocuniAlex_Gaynor: you cannot use w_defs.wrappeditems07:58
bivab (~david@fwstups.cs.uni-duesseldorf.de) joined #pypy.07:58
Alex_Gaynorwhy not? and is there another way to get the values so the JIT can see it as constant07:58
antocuniAlex_Gaynor: because in theory you cannot know the precise type returned by space.newtuple07:59
Alex_GaynorI suppose my usage of W_TupleObject is therefore illegal as well07:59
antocuniyes08:00
antocuniAlex_Gaynor: I think it's fine to do that for experimenting08:00
antocuniand see if you actually get good speedups08:00
Alex_Gaynorreading a simple trace it looks like it more or less works08:00
antocunibut then, before merging you should encapsulate all of that in some space method08:01
antocuniah, cool08:01
Alex_Gaynornot sure how to really benchmark it though, ideally I could push to a branch and let speed.pypy figure it out08:01
antocuniwell, if traces look better, then it's already good :-)08:02
Alex_Gaynoryes, I wonder if it's messing up the tests because its moving stuff around08:03
antocuniAlex_Gaynor: also, I'm a bit confused08:04
antocuniyou never promote self.w_defs08:04
antocunihow can the jit know that it's constant?08:04
nekto0n_antocuni: do you use fork directly or popen to invoke a compiler? popen uses exec*() and thus new process shouldn't as much virtual memory as parent did.08:04
Alex_Gaynorantocuni: yes I do w_defs = jit.hint(self.w_defs, promote=True)08:04
antocuniah, overlooked that08:05
antocuniAlex_Gaynor: also, do this code work with your patch? http://paste.pocoo.org/show/355432/08:08
Alex_Gaynorantocuni: it gets assigned08:09
antocuniok08:09
Alex_Gaynorwhat the behavior is supposd to be for that I don't even know08:09
antocuniAlex_Gaynor: no clue either :-)08:11
Alex_GaynorI guess its a question of whether CPython uses PyTuple_GetItme or PyObject_GetItem08:11
Alex_Gaynor(or heraps PyTuple_Check vs PyTuple_CheckExact)08:12
antocunianyway, your patch seems fine, if you encapsulate the usage of wrappeditems and W_TupleObject inside the space08:12
antocuniAlex_Gaynor: this snippet prints 4: http://paste.pocoo.org/show/355435/08:12
antocuninot sure if it's sane or not :-)08:12
Alex_Gaynorantocuni: on CPython?08:12
antocuniyes08:12
Alex_Gaynorantocuni: that's fine I guess, depends what happens when you override __getitem__ :)08:13
antocuniand also on pypy08:13
Moku (~John@osbk-4d08b10d.pool.mediaWays.net) joined #pypy.08:26
Shinato (~John@osbk-4d08b10d.pool.mediaWays.net) left irc: Ping timeout: 248 seconds08:27
ronny (~ronny@pida/ronny) joined #pypy.08:27
kkris (~kris@93-82-42-106.adsl.highway.telekom.at) left irc: Quit: Leaving.08:32
squiddy (~squiddy@x027.wh17.tu-dresden.de) joined #pypy.08:35
nekto0n_is ssl module disabled in pypy-1.4.1-windows?08:43
antocuninekto0n_: unfortunately, to build pypy with the oracle module you need an oracle client installed08:43
antocunithus, I cannot, sorry08:43
nekto0n_antocuni: thanks for the effort, I'll try to enable swap and translate it myself, how do I enable oracle support?08:45
Alex_Gaynorpypy/translator/goal/translate.py -Ojit pypy/translator/goal/targetpypystandalone.py --withmod-oracle08:46
antocuninekto0n_: ./translate.py -Ojit targetpypystandalone.py --withmod-oracle08:46
verte (~verte@python/site-packages/verte) joined #pypy.08:46
antocuninekto0n_: also, "Please set ORACLE_HOME to the root of an Oracle client installation"08:46
nekto0n_antocuni: roger that08:47
verteis there a standard way to override the annotation on the return value of the method?08:47
vertemmh I'll try to remove this special case instead08:50
Alex_Gaynorverte: ExtRegistry or something like that08:50
verteah, could go that route08:51
vertethink I will just create llmemory.InvalidArrayLengthOffset08:52
verte (~verte@python/site-packages/verte) left irc: Ping timeout: 252 seconds09:02
michaelh (c1beac85@gateway/web/freenode/ip.193.190.172.133) joined #pypy.09:03
Alex_Gaynor (~alex@76-206-244-32.lightspeed.cicril.sbcglobal.net) left irc: Ping timeout: 255 seconds09:07
hpk_ (~hpk@HSI-KBW-109-193-198-175.hsi7.kabel-badenwuerttemberg.de) joined #pypy.09:16
verte (~verte@python/site-packages/verte) joined #pypy.09:19
Alex_Gaynor (~alex@76-206-244-32.lightspeed.cicril.sbcglobal.net) joined #pypy.09:21
verte (~verte@python/site-packages/verte) left irc: Client Quit09:21
bivab (~david@fwstups.cs.uni-duesseldorf.de) left irc: Quit: bivab09:36
antocuniAlex_Gaynor: indeed, this snippet fails with "too many open files" on pypy09:46
antocuniah, but it works if I explicitly do pipe.std{out,err}.close()09:48
kkris (~kris@80-123-34-145.adsl.highway.telekom.at) joined #pypy.09:49
antocunibtw, I think that subprocess.Popen objects should be context managers, so one could use "with" to automatically handle that09:52
kenaan12antocuni default 11491a8d97fba5 15/pypy/module/pypyjit/test_pypy_c/test_model.py: use communicate() to make sure that stdout/err are actually closed, else we leak file descriptors09:55
kenaan12antocuni default 115d58c1ce7fc4 15/pypy/doc/: merge heads09:55
antocuniok, the "too many files" error should be fixed now09:56
nekto0n_ (c397cf21@gateway/web/freenode/ip.195.151.207.33) left irc: Ping timeout: 252 seconds10:00
CIA-403afa 07roundup * 10#402/codecs are broken in pypy-c: To decode a source file, pypy calls: unicode(text, encoding) When CPython uses the equivalent of: codecs.getreader(encoding)(input_stream).r ... * 14https://codespeak.net/issue/pypy-dev/issue40210:01
Moku (~John@osbk-4d08b10d.pool.mediaWays.net) left irc: Ping timeout: 248 seconds10:04
Shanita (~John@osbk-4db15322.pool.mediaWays.net) joined #pypy.10:04
lac (~quassel@63.133.138.10) left irc: Ping timeout: 276 seconds10:10
Guest19749 (~kost-bebi@195.95.210.229) joined #pypy.10:12
Nick change: Guest19749 -> kost-bebix10:12
amaury_ (5bc3480e@gateway/web/freenode/ip.91.195.72.14) joined #pypy.10:17
ElectronicRU (~Pythonist@5f4489b5.dynamic.mv.ru) left irc: Read error: Connection reset by peer10:17
ElectronicRU (~Pythonist@5f4489b5.dynamic.mv.ru) joined #pypy.10:17
verte (~verte@python/site-packages/verte) joined #pypy.10:22
verte2 (~verte@120.154.221.62) joined #pypy.10:23
verte2 (~verte@120.154.221.62) left irc: Client Quit10:23
Kaskuka (~John@osbk-4db16b0e.pool.mediaWays.net) joined #pypy.10:29
Shanita (~John@osbk-4db15322.pool.mediaWays.net) left irc: Ping timeout: 255 seconds10:29
Action: antocuni found a suboptimal trace produced by the jit10:41
antocunihttp://paste.pocoo.org/show/355469/10:41
antocuniin theory, the second guard could be avoided because it's the same as the first one10:41
antocunialthough the JIT is not aware of it10:41
antocuniah no, of course in theory the length of array could be mutated inside the loop, and thus we still need the guard10:44
kenaan12antocuni default 116dfd34bb257f 15/pypy/module/pypyjit/test_pypy_c/test_pypy_c_new.py: port and adapt test_array_sum from test_pypy_c10:48
kenaan12antocuni default 11cd14b43b01e4 15/pypy/module/pypyjit/test_pypy_c/test_pypy_c_new.py: remove the XXX, it does not make any sense :-)10:49
antocunithe nice thing of being alone on working on pypy today is that I can push freely without having to merge :-)10:49
ElectronicRU (~Pythonist@5f4489b5.dynamic.mv.ru) left irc: Read error: Connection reset by peer11:05
hakanardo_antocuni: identifying that i9 is loop invariant and thus know that the length is constant should not be hard11:05
ElectronicRU (~Pythonist@5f4489b5.dynamic.mv.ru) joined #pypy.11:05
antocunihakanardo_: uhm, I think that you need at least a promotion on the length of the array11:08
antocunior am I wrong?11:09
hakanardo_right11:09
hakanardo_a more interesting case would be to use  while i<len(img): though11:09
hakanardo_which we might already handle correctly?11:10
antocuniah, maybe11:14
Action: antocuni tries11:14
hakanardo_I just did, there is only one int_lt in that case11:15
antocunicool11:15
hakanardo_no, the cool thing would be 0 int_lt11:15
antocuniuhm, you need at least one guard to exit the loop11:16
vertein the prologue.11:16
hakanardo_antocuni: right, of course11:16
hruske (~Gasper@188-230-156-183.dynamic.t-2.net) joined #pypy.11:19
verteoutre11:19
vertewhy would an pypy.jit.backend.x86.regloc.AddressLoc get annotated as an int?11:22
lac (~quassel@63.133.138.10) joined #pypy.11:23
verteack, ok.11:33
vertejit (and gc) is kind of a special case, because it can play fast and loose with types, moreso than regular rpython code. a function with a name like 'addr_add_const' should make this clear: it's going to take addrs with arbitrary types and just add things to them.11:37
vertethis means, if you're trying to capture type information against the function, you're going to get whacky stuff like [LENGTH_DESCR, 'chars'] as the adress kind for one call, and something completely different for another.11:38
ElectronicRU (~Pythonist@5f4489b5.dynamic.mv.ru) left irc: Read error: Connection reset by peer11:38
ElectronicRU (~Pythonist@5f4489b5.dynamic.mv.ru) joined #pypy.11:39
kenneth_reitz (~kenneth_r@c-24-127-96-129.hsd1.va.comcast.net) joined #pypy.11:52
alfredodeza (~alfredode@c-24-131-47-167.hsd1.ga.comcast.net) joined #pypy.12:02
alfredodeza (alfredode@c-24-131-47-167.hsd1.ga.comcast.net) left #pypy.12:03
Shinato (~John@osbk-4db15c0b.pool.mediaWays.net) joined #pypy.12:10
Kaskuka (~John@osbk-4db16b0e.pool.mediaWays.net) left irc: Ping timeout: 248 seconds12:10
kenaan12lac default 11f9ce284a688a 15/pypy/doc/temp_index.rst: pypy/doc/temp_index.rst was leftover junk from the sphinxification which should not have been moved.12:17
kenaan12lac default 115dfd2fa3a69c 15/pypy/config/test/test_pypyoption.py: change .txt to .rst12:17
kenaan12lac default 1153c2426b09b8 15/: merge heads12:17
nekto0n_ (c397cf21@gateway/web/freenode/ip.195.151.207.33) joined #pypy.12:20
ronnyhmm, i never actually investigated how to nicely test local stuff12:25
vertelocal stuff?12:26
ronnyverte: before pushing12:27
ronnyhmm, the simplify tests pass, so i'll push12:28
kenaan12RonnyPfannschmidt default 11868496e73f01 15/pypy/translator/simplify.py: translator: remove unused list instanciation in the isinstance simplifier12:28
verteI test stuff I know to be relevant before committing, and test as much as reasonable before pushing. sometimes you're going to break the build and have to live with that if you don't want to wait an hour to get a full set of results.12:31
ronnyoh12:32
ronnyhmm12:32
ronnywoo, thi is actually my first commit as me12:32
verteI would like to buy something with more cores and ram so I could have an automatic build on every commit, but can't really justify that sort of $$12:32
verteneat12:32
verteif you expect to be breaking things frequently, branch12:33
ronnyi expect not to break things12:34
ronnyi'll see to get used to the codebase a bit12:34
antocunironny: that's optimistic ;-)12:34
ronnydoing cleanups&stuff12:34
verteso I'm kind of at an impasse and I don't know what is going to be less work. teach the annotator to ignore strongly typed adr annotations that come out of jit, or implement parametric polymorphism in the annotator. I guess I have more experience fooling the annotator.12:39
ronnyantocuni: i have a feeling the codebase is littered with small bits of dead code12:40
antocunironny: yes, it's likely12:41
ronnypyflakes pypy/translator|grep 'assigned to but never used' looks scary12:42
antocunicould you paste the output, please?12:42
Taggnostr (~quassel@dyn57-215.yok.fi) left irc: Read error: Connection reset by peer12:43
ronnyantocuni: http://paste.pocoo.org/show/35551212:46
ronnyantocuni: i suppose code cleanup iterations would be rather helpfull12:47
lacso I am looking at pypy/doc/test_redirections.py12:47
antocunironny: yes, indeed12:47
Action: hpk_ looking at the mess of test failures after the sphinx documentation merge 12:47
ronnybbl, food12:47
antocunironny: but keep in mind that sometimes having a variable with a meaningful name is more readable than using '_', even if the variable is unused12:48
lachpk: all the test_check_documentation ones should be fixed since they are looking for .txt files which are .rst now12:49
hpk_the question is which tests are now still make sense at all 12:49
hpk_not too many because sphinx does many of them 12:49
Action: lac nods12:49
Action: hpk_ wouldn't mind to help fixing on a branch instead of hurrying because it's on default now 12:50
hpk_what about the ":config:" ReST roles? 12:51
hpk_is there any sphinx-cusotmizing code that deals with that? 12:51
lacrevert the checkin, go back to working on dmalcolm's fork?  I am afraid he might delete it12:52
vertefork it?12:52
lacbut we could check that into a branch if I could only figure out how to bash mercurial into doing that.  That was what I was trying to do yesterday and gave up in frustration.12:52
antocunihakanardo_: what is test_array_intimg supposed to test? The comment says something about the number of guards: is there any particular guard that we expect to be removed?12:53
hpk_let me first just see how serious things are  - i have about an hour time right now12:53
hpk_lac:  what about the :config: ReST roles? 12:53
hpk_i see they still show up in the html file so they don't seem to work 12:53
lacright.12:54
hpk_lac:  did you or David or Armin look at the pypy specific documentation generation? 12:54
lacarmin had nothing to do with this.  And david and I looked at it briefly, but thought that getting rid of most of it was the whole idea.12:54
lacso that we could move from needing a codespeak account.12:55
hpk_lac:  moving away and going to sphinx is fine 12:55
lacwhat is left in that which is needed to migrate?12:55
hpk_but ignoring the pypy specific docutils-customization is not so good 12:55
hpk_lac:  i am trying to figure it out 12:55
hpk_:config: roles are the one thing 12:56
hpk_in general docutils was the base of both the old way and the new way of doing things 12:57
hpk_so it doesn't make sense to completely ignore the customization i suppose 12:57
verte (~verte@python/site-packages/verte) left irc: Quit: ~~~ Crash in JIT!12:58
Taggnostr (~quassel@dyn57-215.yok.fi) joined #pypy.12:59
Rhy0lite (~dje@nat/ibm/x-zrossdkqjnzpexxt) joined #pypy.13:02
derdon (~derdon@p54A6D9EF.dip.t-dialin.net) joined #pypy.13:03
kenaan12hpk42 default 1113ef5b1f6617 15/pypy/doc/: fix failing tests after sphinx-merge by removing py.test support for running documentation13:06
kenaan12antocuni default 112b484cf91faf 15/pypy/module/pypyjit/test_pypy_c/test_pypy_c_new.py: port test_array_intimg from test_pypy_c13:09
kenaan12antocuni default 1138a48b8331fe 15/pypy/module/pypyjit/test_pypy_c/model.py: improve the reporting (thanks to pytest like _assert magic)13:09
kenaan12antocuni default 1177adefbb81e7 15/: merge heads13:09
hpk_lac:  i removed the tests 13:10
lacok13:11
lacnow we need to decide what we really want to do.13:11
laca lot of the things seem to be already done by sphinx13:11
hpk_is there a way to sphinx-generate just one out of the many doc files? 13:13
Action: hpk_ so far has only used sphinx on smaller projects where it didn't matter so much to run all docs13:14
kenaan12antocuni default 117b22560b4e3f 15/pypy/module/pypyjit/test_pypy_c/test_pypy_c_new.py: don't rely on the exact offset13:15
Nick change: skulltone -> kneetone13:16
lacwe never found a way, but regenerating the whole lot was surprisingly fast13:19
kenaan12antocuni default 11927529207222 15/pypy/module/pypyjit/test_pypy_c/test_pypy_c_new.py: fix the test on 32bit, where we get slightly different code13:19
lambacckI'm getting a build failure on windows.13:21
amaury_lambacck: which one?13:21
lambacckjit x8613:21
lambacckmsvc 200813:21
amaury_error message, please :-)13:22
lambacckerror A2075:jump destination too far : by 1 byte(s)13:22
lambacckin implement_13.lbl.s(129113) :13:22
hakanardo__ (~ardo@putten.maths.lth.se) joined #pypy.13:23
hakanardo__antocuni: test_array_intimg is actually a real world image transform I was optimizing13:23
antocunihakanardo__: ah, ok. At the end, I opted to check the whole loop, since it's not too long13:24
antocunisee 2b484cf91faf13:24
hakanardo__the test should make sure to remove the array-bound-check of the calculated index i-64013:24
hakanardo__ok13:24
antocunihakanardo__: then it fails :-(13:25
antocunithe bound check is still there13:25
hakanardo__yes, it would need something like assert len(intimg)==640*480 to kill the bound...13:26
hakanardo__I suppose I never got as far as I intended with this...13:27
antocunihakanardo__: I don't see how it's possible to remove it. Even if we know that i < 640*480, we can't we be sure that i-640 < 640*48013:29
antocunibecause e.g. i could be -sys.maxint and i-640 be very large13:29
gtaylor (~gtaylor@99-5-124-9.lightspeed.gnvlsc.sbcglobal.net) joined #pypy.13:30
hakanardo__right, so how about if we add assert i>0 as well?13:30
hakanardo__i>=0 that is13:30
antocuniyes, then the guard is killed13:31
antocunieven without the assert on len(intimg)13:31
antocuni(this would kill yet another guard :-))13:31
hakanardo__cool, what one realy would want though is to do is "for i in range(len(img)):" and not insert any asserts13:32
hakanardo__ie had the jit figure ot that i is positive by itself13:33
antocuniyes, but that's a bit harder I fear13:33
hakanardo__indeed13:33
antocunibecause we would need a way to pass the information nonneg=True from the annotator to the jit13:34
kenneth_reitz (~kenneth_r@c-24-127-96-129.hsd1.va.comcast.net) left irc: Remote host closed the connection13:34
antocunihakanardo__: ok, with this loop I kill both array bound checks13:35
hakanardo__antocuni: without the assert on the len of intimg how can the upper bound be killed?13:36
hakanardo__there are 3 accesses to intimg, is realy all 3 killed?13:36
antocunihakanardo__: well, there is still one guard, for the condition of the loop13:37
antocunibut it's the only one left, apart for the overflows13:37
hakanardo__ok, so you have changed the loop check to "while i<len(intimg)"?13:38
antocunihakanardo__: so far, I've tried with this: http://paste.pocoo.org/show/355538/13:39
antocunibut I suppose it'd still work if I use while <len(intimg) and kill the assert13:39
hakanardo__there you have the assert on the len13:40
antocuniok, I suppose there is a bit of misunderstanding :-)13:40
antocunion the original loop, we had three guards13:40
CIA-403alfredodeza 07roundup * 10#402/codecs are broken in pypy-c: I want to avoid a discussion about encoding correctness as it works for the intended purpose in CPython (e.g. is *not* calling compile at any tim ... * 14https://codespeak.net/issue/pypy-dev/issue40213:40
antocunione for the loop condition, one for accessing img, and one for accessing intimg13:41
antocuniwith the assert i>=0, we kill the guard on intimg13:41
antocuniwith the other two asserts, we kill the guard on img13:41
ojii (~ojii@adsl-84-226-255-220.adslplus.ch) joined #pypy.13:44
ojiipypy is supposed to be able to install PIL, right?13:44
antocuniojii: I think so, although I never tried by myself13:44
ojiimy fail or pypy fail? http://dpaste.com/514046/13:46
fenrrir (~fenrrir@187.40.154.178) joined #pypy.13:46
antocuniojii: no clue, sorry :-(13:47
antocunimaybe amaury_ has a better clue13:47
antocuniwhich version of pypy are you using?13:47
ojiiPython 2.5.2 (e503e483e9ac, Dec 21 2010, 12:02:48), [PyPy 1.4.1]13:48
DanielHolth (~dholth@2002:62b4:2270:1:224:8cff:fe26:8c67) left irc: Remote host closed the connection13:48
antocuniojii: could you try with a recent nightly build please? http://buildbot.pypy.org/nightly/trunk/13:48
antocunithe one from last night is fine13:49
whitelynx (~whitelynx@li117-47.members.linode.com) joined #pypy.13:49
DanielHolth (~dholth@2002:62b4:2270:1:224:8cff:fe26:8c67) joined #pypy.13:50
ojiiworks antocuni 13:51
voidspace (~voidspace@python/psf/voidspace) joined #pypy.13:51
antocuniojii: good :-)13:51
ojiibtw: is there a ubuntu PPA for pypy?13:53
asabil (~asabil@195.159.219.65) joined #pypy.13:53
ojii(that works with ubuntu 10.10)13:53
antocuninot that I know13:53
hakanardo__antocuni: Without assert I get 4 guards, one for the loop and one for each array access13:54
hakanardo__asserting i>=0 has no effect13:55
ojiithanks a lot antocuni, now let's play around with this ;-)13:55
hakanardo__adding aa 3 guards kill all the array bound checks though13:55
hakanardo__hmm, maybe my pypy is too old...13:56
antocunihakanardo__: mine is from last night13:56
hakanardo__I'll upgrade13:56
antocuniI double checked and I confirm that assert i>=0 kills one guard13:56
amaury_ojii: indeed, the fPIC issue was fixed after 1.4.113:57
harrison (~sr@adsl-69-209-217-94.dsl.chcgil.sbcglobal.net) joined #pypy.13:57
hakanardo__I suppose it could kill the guard on i-640 due to unrolling, but without any assert linking the len(intimg) to the constnat in the loop check I don't see how the other could be removed13:58
antocuniyes, I don't have any idea either14:00
hakanardo__I'll dig into it14:01
lizardo (~lizardo@189.2.128.130) joined #pypy.14:02
ojiiisn't pypy supposed to be faster than cpython? If I run the django-cms test suite using pypy, it takes 3 times longer (and one test fails)14:03
kenaan12antocuni default 11f187f9701d1d 15/pypy/module/pypyjit/test_pypy_c/test_pypy_c_new.py: (antocuni, hakanardo around): improve the test, and check that we remove a couple of guards thanks to the asserts14:03
kenaan12antocuni default 11263e6f36aa8d 15/pypy/module/pypyjit/test_pypy_c/test_pypy_c_new.py: rewrite the test, to kill one guard14:03
CIA-403afa 07roundup * 10#402/codecs are broken in pypy-c: Since PyPy does not use the streamreader to decode source files, yes, your codec should be changed and implement its hacks in the "decode" method ... * 14https://codespeak.net/issue/pypy-dev/issue40214:04
voidspace (~voidspace@python/psf/voidspace) left irc: Quit: voidspace14:05
ronnyojii: a testsuite is most likely completely useless as benchmark14:06
ojiironny, not using it as a benchmark really, just wanted to see if it passes ;-)14:06
ojiiwas surprised by the big difference tho (150s vs 49s)14:06
ronnyi wonder where that is from, it shouldnt be that much slower without jit14:07
ojiiI think I use jit14:08
mdeval (~miguel@134.76.235.21) joined #pypy.14:08
ojiironny, if you want to test it for yourself: http://dpaste.com/514105/14:09
ronnyi need to translate pypy first14:09
CIA-403alfredodeza 07roundup * 10#402/codecs are broken in pypy-c: Fair enough. I am not sure if you want to close this issue since PyPy is not "wrong" but just doing things a bit different. * 14https://codespeak.net/issue/pypy-dev/issue40214:10
kenaan12RonnyPfannschmidt default 11467cb1d0f61c 15/pypy/translator/c/funcgen.py: translator: remove unused list/dict instances in the c funcgen14:14
ronnywhere should i put  a script that downloads the latest pypy build to translate pypy14:15
Alex_Gaynorojii: the problem with test suites is they tend to hit a TON of code one or 2 times, no inner loops which are run frequently14:16
ojiiAlex_Gaynor, good point14:16
ojiiany idea why one test would fail though?14:17
ronnys/without jit/without a chance for the jit to do good/14:17
Alex_Gaynorojii: link to the failure?14:18
ojiiAlex_Gaynor, http://dpaste.com/514137/14:19
Alex_Gaynorstrange14:19
ojiiit passes on cpython 2.5-2.7 though14:19
hakanardo__antocuni: using pypy-c-jit-42773-61fefec7abc6-linux64 from nightly I still get 3 array bound check guards without asserts and asserting i>=0 has no effect14:32
hakanardo__actully the test passes without the assert i >= 014:33
antocunihakanardo__: uhm indeed, the test still passes14:34
antocunibut then it's wrong14:34
asenchiAnyone else having issue with coverage building?14:34
fijalhi14:34
hakanardo__which makes sense since there will be a i>=0 test in the preamble14:34
hakanardo__the array bound check should be 0<i<len(array), right?14:35
antocunihakanardo__: ah, indeed14:35
hakanardo__then there is i+1, so after the preamble we would know that i<=114:36
hakanardo__sorry, i>=114:36
fijalAlex_Gaynor: no idea what's wrong with twisted's thread14:37
fijalalso, grep pyjitpl.py for bad loop if you want to know what are conditions14:37
ronnyrandom tought, should bin/py.py include the 'yo dawg meme' about interpreters?14:37
fijalno14:38
Alex_Gaynor (~alex@76-206-244-32.lightspeed.cicril.sbcglobal.net) left irc: Ping timeout: 255 seconds14:41
DanielHolthanybody play with the old gccxml ctypes generator?14:49
fijalyeah14:51
dmalcolm (~david@c-24-61-12-82.hsd1.ma.comcast.net) joined #pypy.14:51
ojii (~ojii@adsl-84-226-255-220.adslplus.ch) left irc: Read error: Connection reset by peer14:57
kenneth_reitz (~kenneth_r@c-24-127-96-129.hsd1.va.comcast.net) joined #pypy.14:59
harrison (~sr@adsl-69-209-217-94.dsl.chcgil.sbcglobal.net) left irc: Remote host closed the connection15:00
harrison (~sr@adsl-69-209-217-94.dsl.chcgil.ameritech.net) joined #pypy.15:01
__name__ (~name@chello080108038152.1.11.vie.surfer.at) joined #pypy.15:01
__name__ (~name@chello080108038152.1.11.vie.surfer.at) left irc: Changing host15:01
__name__ (~name@sburn/devel/name) joined #pypy.15:01
ronnyare there more fractrals now?15:12
arigato (~arigo@63.133.138.10) joined #pypy.15:17
fijalarigato: hi15:18
fijalarigato: where in the world are you?15:18
arigatohi15:19
arigatostill at the Hyatt hotel15:19
fijaloh15:20
fijalI thought you left yesterday15:20
ronnyhmm, what is $memofield_get_slot_tp_function_600215:21
fijalnoone knows for sure :)15:21
ronnylooks like a cashcow - everyone seems to want it and nobody seems to have it15:22
ronnyseems to be something from cpyext15:23
Action: ronny greps for $memofield15:23
ronnyso some "specialize" annotation does weird stuff15:24
ronnyhmm, i dont want to investigate now ^^15:24
amaury_fijal, ronny: hey, I know what $memofield_get_slot_tp_function_6002 is about15:37
amaury_but I've never managed to suppress the warning15:37
amaury_fyi, it's generated by cpyext15:38
ronnyamaury_: thats as far as i got15:40
ronnyi dont understand how it got to be15:41
kenaan12arigo enable-opts 1185eb5570280a 15/: Close branch.15:44
fijal (~fijal@71-208-182-156.hlrn.qwest.net) left irc: Ping timeout: 240 seconds15:45
ronnyhow many fractals are actually in the translation process? i lost count15:46
xorAxAxronny: about 415:46
xorAxAxactually, its only one fractal15:47
xorAxAxzoomed to different regions15:47
xorAxAxyou can add new regions15:47
arigato (~arigo@63.133.138.10) left irc: Quit: See you15:50
fijal (~fijal@65.101.242.214) joined #pypy.15:51
harrison (~sr@adsl-69-209-217-94.dsl.chcgil.ameritech.net) left irc: Remote host closed the connection15:51
harrison (~sr@adsl-69-209-217-94.dsl.chcgil.ameritech.net) joined #pypy.15:52
apoirier_away (~apoirier@sakura.nagare.org) joined #pypy.15:55
Nick change: apoirier_away -> apoirier15:56
fijalamaury_: I want this special destructor16:02
fijal"but this also has a raw malloc associated"16:02
arigato (~arigo@63.133.138.10) joined #pypy.16:06
jacob22__ (~jacob@63.133.138.10) joined #pypy.16:07
amaury_fijal: I don't understand anything16:08
fijalheh :)16:08
fijalok, that was your idea16:08
amaury_or do you want a leak detector?16:08
fijalto have a special kind of finalizer16:08
fijalthat says "by the way this object has a raw malloced memory, please free it"16:08
amaury_ah16:09
amaury_my old idea was to do this at lexical level16:09
amaury_you cannot rely on the gc here16:09
fijalyes16:10
fijalbut sometimes you really want a GC16:10
fijallike in the example of numpy arrays16:10
fijalbtw16:10
arigatoit can also be done with a hint in the __del__ method (if I understood you correctly)16:10
amaury_what is wrong with lltype.scoped_malloc?16:10
fijalarigato: yes16:10
fijalamaury_: same thing that's wrong with C++ method of dealing with that16:10
kkris (~kris@80-123-34-145.adsl.highway.telekom.at) left irc: Ping timeout: 246 seconds16:10
fijalsometimes your objects escape lexical scope16:10
amaury_so: what is wrong with the gc?16:11
fijalyou have a finalizer16:11
fijalwith all complicated semantics16:11
arigatoyes, including forcing the object to be "old" immediately16:12
antocunifijal: do you want an hint like "this __del__ also calls lltype.free" or "instances of class A owns a raw-malloced area of memory, please free it automatically"?16:12
fijalthe latter please16:12
fijalor16:12
arigatoI would prefer a hint "this is a lightweight finalizer"16:12
fijal"that __del__ doesn't do anything besides freeing memory"16:13
fijalalso prevents JIT from making stuff virtual16:13
fijaletc.16:13
kkris (~kris@93-82-42-99.adsl.highway.telekom.at) joined #pypy.16:13
amaury_are semantics similar to CPython tp_dealloc?16:20
ElectronicRU (~Pythonist@5f4489b5.dynamic.mv.ru) left irc: Read error: Connection reset by peer16:26
ElectronicRU (~Pythonist@5f4489b5.dynamic.mv.ru) joined #pypy.16:26
fprimex (~fprimex@brent-macbook.sc.fsu.edu) joined #pypy.16:35
arigatobringing jitypes2 up-to-date is a major mess16:36
fijal:/16:37
fijalbecause?16:37
arigatoright now I'm stuck at _ctypes/function.py16:37
arigatowhich was heavily refactored on both sides16:37
fijalwow pypy is fast16:40
Action: fijal translating on tannit16:40
antocuniarigato: :-(16:42
mvt (~mvantelli@87.213.45.85) left irc: Quit: Leaving16:42
antocunimy fault probably, I left it laying there for too long16:42
fijaletrepum: ping16:42
arigatono, it may actually not be too bad, but I'm fighting with understanding hg merges more deeply16:43
etrepumfijal: pong16:43
fijaletrepum: should we pull simplejson from here https://github.com/simplejson/simplejson/tree/pypy-support to be the default json for pypy?16:43
etrepumfijal: you should probably leave json to be the default json, they aren't 100% API compatible16:43
etrepumjson is simplejson 1.x16:43
fijalpffff16:44
fijalwe want a fast one though16:44
etrepumit's not fast in cpython either16:44
fijalhow different is the API?16:44
fijalyeah16:44
fijalbut C thingie is faster16:44
etrepumin that version, probably not much16:44
fijalok16:45
etrepumthe pure python simplejson beats the cpython json with C extensions in most cases16:45
fijalI see16:45
etrepum(on cpython)16:45
fijalthere is a real problem there though16:45
etrepumand that is?16:45
fijaland I don't care if we cheat and are simply better at updating stdlib16:45
fijalbut we want a fast json16:45
lucian_ (~lucian@78-86-217-168.zone2.bethere.co.uk) joined #pypy.16:45
etrepumalso, simplejson doesn't plug directly into the stdlib, things need to get transformed for that to happen16:46
etrepums/simplejson/json mostly but I think the structure and test locations might have changed too (although the structure thing is only really relevant to the c extension so not very relevant)16:47
kost-bebix (~kost-bebi@195.95.210.229) left irc: Quit: kost-bebix16:47
arigatoantocuni: I'm not sure, but I think you messed up something in df1066ff4a7516:47
cco3 (~conley@c-69-181-140-72.hsd1.ca.comcast.net) joined #pypy.16:47
arigatoit looks like the "hg merge default" is incomplete16:47
fijalI see16:48
etrepumI agree that cheating is probably the best thing to do, but it does deviate from cpython and it's more work16:48
fijalit does deviate in the sense of API?16:48
fijalor it does deviate in the sense of more changes to stdlib?16:48
etrepumif we cheat, I would prefer to do it after we're done fixing simplejson for pypy16:48
amaury_arigato: I've looked at several failures in ctypes16:48
fijalright16:49
amaury_and it seems that we use the CArgObject wrong16:49
etrepumit deviates mostly in that there is more functionality in simplejson than json16:49
fijalthere are several questions, all of them valid16:49
etrepumBUT there were dumb things that people subclassed or whatever in json that don't work anymore16:49
fijalfrom "how fast is pypy on this particular piece of code" up to "how fast we can load json using different interpreters"16:49
lacif anybody here has any interest in being the next ctypes guru and maintainer for python-dev now would be the perfect time to volunteer.16:49
amaury_arigato: at least, not at all like CPython's PyCArgObject16:49
fijallac: have you seen that code?16:49
etrepumand the API changed slightly such that simplejson will return ASCII str instead of unicode to save RAM where json always returns unicode16:50
etrepumpeople don't like that one16:50
etrepum(cpython purity over practicality people)16:50
fijalthat's easy to "fix" right?16:50
etrepumyeah you just change ''.join to u''.join16:50
fijalmore importantly we have an optimization in mind16:50
etrepumthere might be other things16:50
fijalthat would make this irrelevant16:50
lacfijal: yes, and so if somebody was in the middle of rewriting it, this would be the best way to get their rewrite to be part of CPython, should that be desired.16:51
fijalascii-only unicode should take the same amount of RAM eventually16:51
fijallac: it's not16:51
fijalit's a C beast16:51
ronnyfijal: factor style or internal type differences?16:51
etrepumyeah, if ascii-only unicode had better storage I could simplify things16:51
etrepumconsiderably16:51
etrepumwhich the jit would probably like too16:51
fijalyes16:51
lacok, so my intial idea that nobody here wants to touch it _AT ALL_ was more correct. :-)16:52
fijalthat's thought out and clearly doable16:52
fijalbut not yet implemented16:52
fijalit'll happen though16:52
fijalbut yeah, we can't have everything *today*16:52
etrepumif we deviate the unicode behavior from simplejson then it'll use more RAM16:52
fijalalthough that would be cool :)16:52
fijalright16:52
arigatolac: I'd be more happy if nobody showed up and people stopped adding more and more obscure features to it16:52
fijalarigato: you can be a maintainer16:53
fijalthat fixes segfaults and says "no"16:53
exarkun+116:53
lambacckis there a way to only translate cpyext and not everything else if all I am changing is cpyext?16:53
fijalbut I don't think python-dev works this way16:53
fijallambacck: no16:53
fijallambacck: but you can test things without translating16:53
amaury_and you can post your changes if you want a review16:54
fijaleh16:55
fprimex (~fprimex@brent-macbook.sc.fsu.edu) left irc: Quit: http://www.fprimex.com16:55
fijalarigato: do you remember the issue that allocating lots of raw memory didn't trigger gc collection?16:55
fijalor collected too much16:55
fijal(I don't remember)16:55
lambacckwindows address space limits mean that I have to do the translate with cpython and it takes sooooo loooong :(16:55
xorAxAxlambacck: pypy doesnt have separate compilation on the trunk16:56
fijallambacck: implement get_total_memory for windows16:56
arigatofijal: it doesn't trigger gc collection16:57
fijalarigato: and it's not fixed?16:57
arigatodon't think so16:57
fijalok16:57
arigatoit's still in the XXX at the start of minimark.py16:58
fijalit's essential for merging numpy-exp16:58
lambacckxorAxAx: I don't understand what that means16:58
antocuniarigato: in which sense incomplete? Looking at the bitbucket changeset it says "delta 0 bytes", which means (I think) that it's a merge without any diff attached16:58
xorAxAxlambacck: it means "no"16:58
amaury_lambacck: I can compile pypy with pypy, it works16:58
amaury_on win3216:58
amaury_it uses up to 1.9Gb RAM16:58
lambacckamaury_: which version of pypy are you using to build?16:59
panni_ (hannes@ip-178-203-73-7.unitymediagroup.de) joined #pypy.16:59
arigatoantocuni: following the changes, it looks like _ctypes/function.py was modified in trunk but the changes are not added by df1066ff4a75 into jitypes216:59
arigatoor I'm messing up something16:59
amaury_lambacck: 1f6ef71f3d19, Mar 01 201117:00
lambacckOkay will try with just built version17:01
stakkars_ (~tismer@72-254-60-112.client.stsn.net) left irc: Ping timeout: 240 seconds17:01
arigatoantocuni: e.g. 964669bb3f60 adds a number of FUNCFLAG_USE_LASTERROR17:01
Action: antocuni looks17:02
arigatothese FUNCFLAG_USE_LASTERROR are still in trunk, but not in jitypes2 now17:02
arigatoalthough there was several merges from trunk to jitypes2 inbetween17:02
arigatoah, the merge ff00970bf708 adds them to function.py in jitypes217:02
fijaleh17:03
antocuniarigato: yes, I remember that at some point I had this code in jitypes2 as well17:03
fijalkenaan: wake up17:03
arigatoantocuni: er17:04
antocuniarigato: yes, it's in _call_funcptr17:04
arigatoforget it17:04
arigatoit's still in jitypes2 too17:04
arigatoI guess I'm getting majorly confused by diffs17:05
antocuniwhich app do you use to browse hg history, btw?17:05
arigatohglog17:05
fijal (~fijal@65.101.242.214) left irc: Remote host closed the connection17:05
arigato.../hg/arigo/hack/hg/hglog17:06
Action: antocuni clones the repo17:06
ronnyarigato: can i see that one?17:06
antocuniarigato: can I kill your svn user/arigo?17:07
antocuni(I mean: is everything migrated in hg?)17:07
arigatoronny: http://paste.pocoo.org/show/355640/17:07
arigato(plus a few colors)17:07
arigatoantocuni: yes17:07
lambacckcreating a get_total_memory  for win32 won't help because it is not about the amount of memory I have. I am on 64 bit system and have more memory available than is actually available in a windows 32bit address space.17:07
arigatoget_total_memory() is supposed to return a number capped by the addressable space17:08
juneau001 (~juneau@131.225.24.162) joined #pypy.17:08
ronnyarigato: looks a lot like glog17:09
arigatoit should default to 2GB on Windows anyway17:09
arigatoronny: yes17:09
ronnyarigato: custom template?17:09
lambacckThere is a limit defined there for windows, but the actual addressable limit appears to be less than 2G17:09
arigatoronny: yes, and a Python script to show the lines on the left17:10
ronnyarigato: why not use the glog extension for those lines?17:10
lambacckthe address space is 2G, but there are things that eat into the 2G on windows.17:10
arigatoronny: dunno17:10
ronnyarigato: as far as i can see hg glog -v should be pretty semilar17:11
arigatoronny: the hg question "why don't you use extension xyz" is usually answered "because I didn't know it existed at that time"17:11
antocuniarigato: I have a patch that removes the hardcoded path to hgtree.py17:11
ronnyarigato: sometimes it makes sense to aks if there is something like X17:13
ronnyhmm17:13
ronnydamn, bbl17:13
lucian_ (~lucian@78-86-217-168.zone2.bethere.co.uk) left irc: Remote host closed the connection17:19
Alex_Gaynor (~alex@76-206-244-32.lightspeed.cicril.sbcglobal.net) joined #pypy.17:20
lambacckshould I be running the cpyext tests with cpython or pypy?17:21
antocunilambacck: as you prefer, but we usually use cpython17:21
antocuni(we should start using pypy, probably :-))17:22
lambacckI am getting the pypy dieing if I run with the pypy interpreter. 17:22
lambacckthe process goes away with windows complaining that there is a problem with the process, but not providing further information :(17:22
antocunilambacck: a segfault, maybe?17:23
stakkars_ (~tismer@72-254-60-112.client.stsn.net) joined #pypy.17:23
antocuniwell, actually I don't know, maybe cpyext tests are not meant to run under pypy17:23
lambacckthat would be my guess, but it is not telling me. I would have to go figure out how to turn off the windows auto notify microsoft of program errors thing.17:24
lambacckokay it says exception code c000000517:25
lambacckDoes not seem to be giving me a stack trace.17:26
mcdonc (~mcdonc@ip72-209-213-54.dc.dc.cox.net) left irc: Remote host closed the connection17:27
amaury_exception code c0000005 is a segfault17:30
amaury_lambacck: how do you run tests exactly?17:31
kenaan12etrepum default 11eec5b51676ae 15/pypy/rpython/memory/gc/: fix up get_total_memory on darwin to clip at addressable_size17:32
arigato (~arigo@63.133.138.10) left irc: Ping timeout: 276 seconds17:34
lambacckOkay cpyhon 2.7 dies in test_eval.py, the error is a BEX error which as far as I can tell is related to data execution prevention.17:35
mcdonc (~mcdonc@ip72-209-213-54.dc.dc.cox.net) joined #pypy.17:39
cwillupypy's evil plot is at last revealed:  the execution of data as code!17:39
amaury_lambacck: ah, you probably segfault around the fopen and other FILE* functions17:40
amaury_I'm sure you are not using VS9.0 :-)17:40
etrepum (~bob@c-67-180-34-190.hsd1.ca.comcast.net) left irc: Quit: etrepum17:41
lambacckIt is forcing me to use VC 2008 lite. 17:41
lambacckI also have full VS201017:41
lambacckthe build process that is17:41
apoirier (apoirier@sakura.nagare.org) left #pypy ("Leaving...").17:42
lambacckThe build appears to be picking VC from Visual C++ express edition 2008 even if I run vcvarsall.bat from vcvarsall.bat from VS201017:44
lambacckamaury_: so is there any specific thing I can do to not have an error?17:46
amaury_lambacck: did you try to *not* run vcvarsall.bat?17:47
lambacckDon't remember. 17:47
amaury_it's normal that the tests suite tries to pick the same compiler as the one used to compile CPython17:47
amaury_some tests open file descriptors from CPython, and pass them to pypy17:48
amaury_it fails/crashes if CPython and pypy don't use the same msvcrt17:48
exarkun(File descriptors are okay, it's FILE* that cause problems)17:49
lambacckso which crash is this related to? cpython crash or pypy crash?17:49
amaury_exarkun: no, file descriptors are local to an instance of msvcrt17:50
amaury_which pypy did you use?17:50
exarkunamaury_: Oh, awesome.17:50
exarkunOf course.  Windows likes handles instead.17:51
amaury_yes17:51
amaury_real Windows developers don't use fd or FILE*17:52
ronnyuhm guys, wouldnt it make sense to throw away the flowgraph/lltype/whatever data once it can start compiling?17:52
ronny(cause it got me into swappy issues)17:52
lambacckI was using local built pypy 42788:92752920722217:52
ronnypypy translation ram usage + 2 gcc = swappy17:52
lambacckI was also using cpython 2.7 to run cpyext tests17:53
lambacckboth crashed (but at different points and for different reasons)17:53
amaury_if you used cpython2.7 to build your pypy, chances are that it uses VC2008 as well17:54
lambacckI am pretty sure VC2008 is being used for everything17:55
amaury_yes17:55
amaury_on my machine it's different, I only have VS2005 and VS201017:55
amaury_so these cpyext tests consistently crash for me with CPython2.717:56
amaury_I had to recompile CPython with one of these compilers to let the tests pass17:56
lambacckokay so I should rebuild cpython vs2010?17:57
lambacckor I should try to force vs2008 for pypy17:58
amaury_we still have no explanation for the crashes17:59
amaury_everything uses VS2008, and should work17:59
lucian_ (~lucian@78-86-217-168.zone2.bethere.co.uk) joined #pypy.18:00
cco3 (~conley@c-69-181-140-72.hsd1.ca.comcast.net) left irc: Ping timeout: 246 seconds18:00
amaury_lambacck: can you run the tests again with the -v option?18:00
lambacckI am not sure which compiler was used for dependency dlls18:01
amaury_you should open a new console without vcvarsall.bat18:01
lambacckI am going to rebuild deps with vc2008 as well to be sure that everything is correct18:02
lambacckie openssl, gzip, etc18:02
amaury_ah, right18:02
lambacckls18:02
lambaccksorry wrong window18:02
amaury_"dir"18:02
amaury_on windows18:02
lambacckI can't train myself out of using ls, so I just have cygwin in my path and then I can lapse into my unix ways :)18:03
amaury_well, be careful18:04
amaury_link.exe is supposed to call the linker18:04
lambacckI don't think I have any build tools installed in cygwin18:05
lambacckokay link is in my path, but when vcvars all is run link goes to VS18:06
amaury_right18:07
amaury_btw, the pypy compilation tools also use vcvars to update the environment of spawned commands18:08
fprimex (~fprimex@brent-macbook.sc.fsu.edu) joined #pypy.18:09
panni_ (hannes@ip-178-203-73-7.unitymediagroup.de) left irc: Read error: Connection reset by peer18:09
amaury_lambacck: it seems that "pypy.exe test_all.py module/cpyext" does not work18:10
amaury_this has nothing to do with compilers &co18:10
amaury_but probably with different kind of function types18:11
lambacckokay. but I am still crashing out of cpython with the data execution protection problem18:11
amaury_in which test?18:11
lambaccktest_eval.py18:12
amaury_which function?18:12
ronnytranslated :)18:12
seanbrant_ (~seanbrant@c-98-206-179-115.hsd1.il.comcast.net) joined #pypy.18:13
amaury_and: can you open the just-in-time debugger, and check whether there are several versions of the msvcrt runtime loaded in the running program?18:13
ronnyman, my laptop is a pain, translation took 3.54 hours18:13
lambacckI don't know which function18:14
lambacckall I am getting are dots18:14
amaury_you have to run tests with the -v option18:14
amaury_or: count printed dots for test_eval.py18:14
seanbrant_im trying to pip install mysql-python on a mac in a pypy virtualenv 18:14
seanbrant_getting a compiler error18:15
lambacckrunning tests again. I had 3 dots18:15
amaury_so the failure was in test_run_file()18:16
amaury_(the 4th function in the file)18:17
seanbrant_erorr http://pastebin.com/mcH7N8fc18:17
amaury_seanbrant_: is your virtualenv recent enough?18:18
amaury_I know there is a specific version for pypy support18:18
seanbrant_amaury_: just created it18:18
seanbrant_just installed pypy with homebrew18:18
seanbrant_ill upgrade it and see what happens18:18
lambaccktest_run_file is indeed the culprit.18:19
lambacckhow do I tell how many mscvrts are loaded?18:19
amaury_attach the debugger when it crashes18:20
amaury_to quickly run one test, you can add "-k test_run_file"18:20
lambacckdebugger is attached18:20
amaury_lambacck: in the debugger: Debug/Windows/Modules18:21
amaury_IIRC18:21
lambacckhmm, don't seem to have that.18:22
amaury_shortcut is "Ctrl-Alt-U"18:22
lambacckthats got it18:23
lambacckI have msvcrt.dll and msvcrt90.dll18:23
juneau001 (~juneau@131.225.24.162) left irc: Quit: juneau00118:24
amaury_should be OK18:24
lambacckokay. working on rebuilding everything with vs2008.18:25
Action: amaury_ -> dinner18:26
amaury_see you18:26
lambacckbye, thanks for the assistance18:26
amaury_well, it's not othen that I can share with someone on pypy-win3218:28
amaury_often18:28
amaury_ (5bc3480e@gateway/web/freenode/ip.91.195.72.14) left #pypy.18:28
seanbrant_ (~seanbrant@c-98-206-179-115.hsd1.il.comcast.net) left irc: Quit: seanbrant_18:39
lambacckdependencies built and now trying to build pypy with VS2008 selected with vcvarsall.bat18:40
fijal (~fijal@71.33.237.102) joined #pypy.18:43
fijalAlex_Gaynor: so I clicked the buildbot so we had two benchmark runs18:51
Alex_Gaynorah18:51
fijal<Alex_Gaynor> fijal: what's "abort bad loop"?18:51
fijalI don't know, but I know how to find it18:51
fijalgrep pyjitpl.py for "bad loop"18:51
fijal<Alex_Gaynor> fijal: what are the semantics of int_divide(i0, i1) where i1 is 0?18:52
fijalI don't remember18:52
fijalbut I think there is int_divide_overflow18:52
fijaloperation or something18:52
fijalor zerodivision18:52
fijalthat should raise18:53
fijaland int_divide should not get 018:53
fijalin other words it's undefined18:53
Alex_Gaynorso would constant folding int_div(i2, i2) -> 1 always be legal?18:53
Alex_GaynorI see ABORT_LOOP at L1814 of pyjitpl18:54
Alex_Gaynorthe comment isn't super uesful18:54
fijalit *might* be the case that zero division checking got moved to codewriter18:54
fijalin which case there should be guard just before18:54
fijalthat checks if stuff is zero18:54
panni_ (hannes@ip-178-203-73-7.unitymediagroup.de) joined #pypy.18:54
fijalyou can easily reproduce it with a test18:54
fijalthat runs the jit and checks for ZeroDivisionError18:55
fijalthere should be one already even18:55
fijalbad loop means compilation somehow failed18:56
fijallike the loop which was supposed to close didn't close18:56
fijalor something18:56
Alex_GaynorI get 2 of these running 1 file of django tests :/18:56
fijalyou can get errors if tracing goes on some bad path18:56
fijalor you ended up at the wrong inline level18:56
fijalI think those are a bit mysterious :)18:56
fijal2 are not too bad18:56
Alex_Gaynorok, there's also a bunch of too long traces18:57
htoothrotamaury_, do you have a buildbot slave config for machines that were doing the windows builds?18:57
Alex_Gaynorbut we knew about those before18:57
htoothrotor anyone, for that matter18:57
Nick change: lucian_ -> lucian19:00
daniloaf (~daniloaf@150.165.63.86) joined #pypy.19:00
fijalAlex_Gaynor: do you *really* want to know?19:04
Alex_Gaynorfijal: know what?19:04
fijalwhat's a bad loop19:04
Alex_GaynorI suppose yes19:04
fijalgrep for InvalidLoop19:05
fijaleverything that raises that19:05
Alex_Gaynorok19:07
daniloaf (~daniloaf@150.165.63.86) left irc: Quit: Saindo19:07
panni_ (hannes@ip-178-203-73-7.unitymediagroup.de) left irc: Read error: Connection reset by peer19:09
alfredodeza (~alfredode@c-24-131-47-167.hsd1.ga.comcast.net) joined #pypy.19:16
panni_ (hannes@ip-178-203-73-7.unitymediagroup.de) joined #pypy.19:21
__name__ (~name@sburn/devel/name) left irc: Remote host closed the connection19:25
dmalcolm (~david@c-24-61-12-82.hsd1.ma.comcast.net) left irc: Quit: Leaving19:36
alfredodeza (~alfredode@c-24-131-47-167.hsd1.ga.comcast.net) left irc: Remote host closed the connection19:36
fijalok19:37
fijalI need to implement SSE before running benchmarks19:37
fijalcrap19:37
stakkars_ (~tismer@72-254-60-112.client.stsn.net) left irc: Quit: schnarch19:44
fijal (~fijal@71.33.237.102) left irc: Ping timeout: 276 seconds19:49
AdemanI thought we already emitted SSE code?19:52
Alex_GaynorAdeman: I assume fijal is talking about vectorized SSE for numpy19:53
Shinato (~John@osbk-4db15c0b.pool.mediaWays.net) left irc: Ping timeout: 248 seconds19:59
Shinato (~John@osbk-4d08b7b9.pool.mediaWays.net) joined #pypy.19:59
AdemanI assume so too, I just wonder how he's going to do it. Unroll two or four iterations at a time then merge float_*() ops?20:00
Alex_Gaynorno idea20:01
fenrrir (~fenrrir@187.40.154.178) left irc: Quit: Saindo20:06
daniloaf (~daniloaf@189.71.125.186) joined #pypy.20:11
lucian (~lucian@78-86-217-168.zone2.bethere.co.uk) left irc: Ping timeout: 250 seconds20:15
jacob22__ (~jacob@63.133.138.10) left irc: Read error: Operation timed out20:16
lucian (~lucian@78-86-217-168.zone2.bethere.co.uk) joined #pypy.20:17
mvt (~mvantelli@53530442.cm-6-4a.dynamic.ziggo.nl) joined #pypy.20:19
antocuni (~antocuni@host17-122-dynamic.0-79-r.retail.telecomitalia.it) left irc: Ping timeout: 252 seconds20:25
Moku (~John@osbk-4d08b7b9.pool.mediaWays.net) joined #pypy.20:25
Shinato (~John@osbk-4d08b7b9.pool.mediaWays.net) left irc: Ping timeout: 255 seconds20:25
jacob22__ (~jacob@63.133.138.10) joined #pypy.20:26
cco3 (~conley@c-69-181-140-72.hsd1.ca.comcast.net) joined #pypy.20:38
stakkars_ (~tismer@75.101.111.84) joined #pypy.20:40
pjenveyfijal or Alex_Gaynor or someone with push access, I have a quickie patch here https://bitbucket.org/pjenvey/pypy20:49
Alex_Gaynorpjenvey: test please :)20:50
pjenveyyea, it's barely even worth one, just a dumb error message20:50
pjenveybut I'll make one if you really want one20:50
Alex_Gaynoryeah I'll live without at est20:50
Alex_Gaynorpjenvey: can you send a pull request?20:51
pjenveyAlex_Gaynor - done20:52
Alex_Gaynorwas I cc'd on it (or however it works)20:52
pjenveyI think it just alerted user 'pypy' but I'm pretty sure you'd get a notification  too20:53
hpk_maybe we should make an email alias for pull requests that goes to the mailing list 20:54
Alex_Gaynorpjenvey: whatever, I'll merge this the old fashioned way20:57
Alex_Gaynorah crap, my cwd is full of stuff, can anyone else merge this?20:57
kenaan12pjenvey default 11362d11da6174 15/pypy/objspace/std/objecttype.py: fix thinko in _abstract_method_error's error message21:02
Ademanhopefully that worked21:02
lizardo (~lizardo@189.2.128.130) left irc: Quit: Leaving21:02
Ademanhg makes this stuff so easy :-)21:03
pjenveythanks guys21:06
mdeval (~miguel@134.76.235.21) left irc: Quit: leaving21:06
Ademannp, thanks for the patch21:06
fprimex (~fprimex@brent-macbook.sc.fsu.edu) left irc: Quit: http://www.fprimex.com21:10
DanielHolth (~dholth@2002:62b4:2270:1:224:8cff:fe26:8c67) left irc: Quit: Ex-Chat21:12
Rhy0lite (~dje@nat/ibm/x-zrossdkqjnzpexxt) left irc: Quit: Leaving21:18
fijal (~fijal@c-67-166-33-96.hsd1.co.comcast.net) joined #pypy.21:37
gtaylor (~gtaylor@99-5-124-9.lightspeed.gnvlsc.sbcglobal.net) left irc: Quit: Ex-Chat21:40
Shanita (~John@osbk-4d08722a.pool.mediaWays.net) joined #pypy.21:41
Moku (~John@osbk-4d08b7b9.pool.mediaWays.net) left irc: Ping timeout: 248 seconds21:43
Varriount (~Varriount@pool-96-255-210-183.washdc.fios.verizon.net) left irc: Read error: Connection reset by peer21:44
bobbyz_ (~bobbyz@12.131.26.130) joined #pypy.21:44
zrbecker (~zrbecker@ip98-164-227-211.oc.oc.cox.net) joined #pypy.21:44
Nick change: Varraway -> Varriount21:44
etrepum (~bob@accessnat4.mochimedia.net) joined #pypy.21:50
kkris (~kris@93-82-42-99.adsl.highway.telekom.at) left irc: Ping timeout: 255 seconds21:51
mcdonc (~mcdonc@ip72-209-213-54.dc.dc.cox.net) left irc: Ping timeout: 264 seconds21:56
mcdonc (~mcdonc@ip72-209-213-54.dc.dc.cox.net) joined #pypy.21:59
fijalronny: ping22:03
Ademanfijal: so I extended my optimizeopt to handle the obvious case of int_sub(iX, ConstInt()) and now my optimizeopt has the potential to emit int_add(iX, ConstInt(-something)). Currently I emit int_sub(iX, ConstInt(something)) instead, this breaks two tests in test_jitdriver because it expects int_add(iX, ConstInt(-something)) instead, can I consider this wrong and alter the test accordingly?22:12
__name__ (~name@sburn/devel/name) joined #pypy.22:13
fijalAdeman: what's wrong with int_add(iX, ConstInt(-something))22:13
fijal?22:13
fijalbut in general, yes22:13
Ademanfijal: I don't really know what's "best" I just figured emitting int_sub(iX, ConstInt(something)) makes sense22:15
fijalyeah, it does22:15
Alex_Gaynoremitting int_sub makes it a bit easier for people reading traces I suppose22:15
Alex_Gaynorand it'll keep tests passing22:15
bobbyz_ (~bobbyz@12.131.26.130) left irc: Ping timeout: 246 seconds22:15
Ademanok, then I'll fix these two tests. Thanks22:17
Action: fijal thinks how to introduce SSE without having new boxes22:18
fijaladding new boxes is a mess :-[22:18
Alex_Gaynorfijal: you're talking about vectorized ops?22:20
Ademanwould it make any sense to keep a repository of unoptimized traces somewhere, and allow optimizeopts to be independently run over those traces to experiment with real world traces ?22:20
fijalAlex_Gaynor: yes22:20
Alex_GaynorAdeman: its' called the test suite :)22:20
fijalAdeman: no22:20
fijalAdeman: at least not more than test_optimizeopt works22:20
fijaland test_pypy_c22:21
Alex_Gaynortest_pypy_c sucks because you have to retranslate22:21
Ademanwell, for instance my optimization works, but I don't know if it'll ever help in a real trace22:21
fijalyeah22:21
fijalAdeman: that's a thing you never know :)22:21
fijalever - yes22:21
fijalsignificantly so - doubtful22:22
Ademanaside from my brainfuck interpreter (50%, pretty happy with that)22:22
fijalok I'm annoyed now22:23
fijalso guess what22:23
fijalyou need a pypy *default* to run jitviewer22:23
whitelynx (~whitelynx@li117-47.members.linode.com) left irc: Quit: Ex-Chat22:23
fijaland then you can't do svn up <other branch> without internet access22:23
Ademanbranches don't work?22:23
fijalbecause hey svn22:23
fijalno22:23
fijalbecause we keep changing interfaces22:23
fijalbetween a part that lives in pypy and part that does not22:24
Ademanhah, well that explains why jitviewer wasn't working for me a few weeks ago22:24
fijalAlex_Gaynor: remember "no vim installed" error?22:24
Alex_Gaynorfijal: yes22:24
fijalit's even worse if you have vim installed22:24
Alex_Gaynorwell "vim is not compiled with diffs enabled"22:24
fijalwhat was the fuckin option?22:24
Ademanvim -d  ?22:25
fijalno hg merge -something22:25
Ademanunless there's some weird nocompatibility problem22:25
Ademanoh22:25
Alex_Gaynorfijal: http://paste.pocoo.org/show/355793/22:26
Ademan--tool emacs ?22:26
Alex_Gaynorfijal: it's the merge= line22:26
fijalChecked out revision 80409.22:27
fijalabort: svn: '/home/fijal/src/pypy/lib_pypy' is not a working copy22:27
fijalI can't merge default now22:27
fijalanyway screw that22:28
fijalI have shopping to do22:28
Ademansvn is hard, let's go shopping22:29
fijalthat's hg22:29
fijalI know how to merge with svn22:29
Action: Alex_Gaynor doesn't22:29
kenneth_reitz (~kenneth_r@c-24-127-96-129.hsd1.va.comcast.net) left irc: Quit: Computer has gone to sleep.22:29
fijalAlex_Gaynor: do you feel like merging default into numpy-exp?22:30
Alex_Gaynorfijal: yeah I can do that22:30
fijalawesome :)22:31
fijalnot numpy-exp to default mind you!22:32
Alex_Gaynorfijal: hmm a few merge conflicts22:33
fijalI got them the first time22:33
fijalbut then I killed my terminal22:33
fijalbecause vim kept popping up22:33
Alex_Gaynorwant me to try to resolve them, or push with the issues?22:33
fijaland now I get only svn errors22:34
fijalthey're trivial no?22:34
fijallike policy merge22:34
Alex_Gaynormerging pypy/jit/metainterp/test/test_ztranslation.py failed!22:34
Alex_Gaynoris not obvious to me22:34
fijalto me neither22:34
Alex_Gaynorwell I think I got it22:35
Alex_Gaynorfijal: pushign22:39
kenaan12alex_gaynor numpy-exp 11f90ccf766b69 15/: Merged default into numpy-exp.22:39
Alex_Gaynorfijal: you should probably review numarray.py22:39
fijalwell22:40
fijalmy working copy is in an unfixable state22:40
fijalWTF?22:40
fijaland...22:40
fijalI just removed all my compiled pypys22:41
fijalawesome22:41
stakkars_ (~tismer@75.101.111.84) left irc: Quit: schnarch22:42
hpk_fijal:  somehow your working machine tends to up in this state, no matter what, doesn't it? 22:43
fijalno22:43
fijalmachine is fine22:43
fijalbut I tend to try too many things at once22:43
fijalwhich usually break tools that are unprepared22:43
fijallike mercurial22:44
fijalapparently if you crash your merge in the middle22:44
hpk_"unprepared" :)22:44
fijalnot by pressing C-c22:44
fijalyou're screwed22:44
lacthere is a way to get hg to not think that you want to resolve conflicts with vim22:44
hpk_i have ended up with borked checkouts as well 22:44
fijalsee?22:44
exarkunwhere the good software at22:45
fijalmoving testrunnner out would already be an improvement22:46
hpk_what do you mean? 22:46
fijalbecause right now you can't work without internet connection at all22:46
hpk_inlining testrunner? 22:46
fijalyeah22:46
fijalwell, removing the svn part22:46
hpk_ronny:  would be cool to do this soonish22:46
Alex_Gaynorlac: yes, it's in the thing I pasted22:47
fijalanyway22:47
fijaloff for the weekend see you22:47
fijal (~fijal@c-67-166-33-96.hsd1.co.comcast.net) left irc: Quit: Leaving22:47
ronnyhpk_: hmm, where should i put it?22:50
hpk_exactly where it is now22:50
hpk_just not as a subrepo 22:50
ronnyok22:51
hpk_and please see if a pull -u actually works 22:51
hpk_i.e. going from subrepo to non-subrepo does not cause problems for people naively updating22:51
ronnythere isnt a hg release with svn removal bug (i checked and ensured back when i had the issue with hackbeil)22:52
ronny(cause i run hg's default branch)22:52
hpk_so do you suppose one needs to manually remove the subdir before updating? 22:52
hpk_(not sure what you mean with your hackbeil/release commenT) 22:53
kkris (~kris@93-82-42-99.adsl.highway.telekom.at) joined #pypy.22:54
ronnyhpk_: im 100% sure that all hg releases with svn subrepo support handle the case well22:54
ronnycause i hit the breakage when messing with hackbeil22:54
hpk_ronny:  ah, fine :) 22:55
ronnywile running the development version of hg22:55
kkris (~kris@93-82-42-99.adsl.highway.telekom.at) left irc: Client Quit22:55
kenaan12RonnyPfannschmidt default 11b0feedf8d4d8 15/: inline testrunner23:02
hpk_nice23:09
Action: hpk_ turns the part testrunner of codespeak read-only 23:10
kenneth_reitz (~kenneth_r@c-24-127-96-129.hsd1.va.comcast.net) joined #pypy.23:10
ronnyhpk_: what do we do about the benchmark23:13
ronny+s23:13
ronnyhmm23:13
ronnylets see if there was any change since i converted them23:13
hpk_why not create a bb+pypy/benchmarks? 23:13
hpk_but let's maybe discuss this with fijal 23:13
mvt (~mvantelli@53530442.cm-6-4a.dynamic.ziggo.nl) left irc: Quit: Leaving23:13
ronnyhpk_: i kind of put it there already23:14
ronnybut its not used23:14
hpk_ah ok23:14
ronnylike i said - im kind of in limbo23:14
ronnyi wonder if i need to adabt how i handle communication23:14
hpk_moving forward piece by piece is fine, i may be off for some time from any time on, though23:15
bobbyz_ (~bobbyz@12.131.26.130) joined #pypy.23:15
ronnyhpk_: considering the timeframes more could have been done23:16
Kaskuka (~John@osbk-4d087afa.pool.mediaWays.net) joined #pypy.23:19
Shanita (~John@osbk-4d08722a.pool.mediaWays.net) left irc: Ping timeout: 248 seconds23:20
Nick change: Varriount -> Varraway23:22
hruske (~Gasper@188-230-156-183.dynamic.t-2.net) left irc: Ping timeout: 255 seconds23:23
Nick change: Varraway -> Varriount23:28
Shinato (~John@osbk-4d08728e.pool.mediaWays.net) joined #pypy.23:30
Action: lac -->dinner23:30
Kaskuka (~John@osbk-4d087afa.pool.mediaWays.net) left irc: Ping timeout: 255 seconds23:30
Action: ronny -> bed23:31
daniloaf (~daniloaf@189.71.125.186) left irc: Quit: Saindo23:32
stakkars_ (~tismer@72-254-85-128.client.stsn.net) joined #pypy.23:36
stakkars_howdy23:40
Alex_Gaynorafternoon23:42
piotrek_m (4dfd6b75@gateway/web/freenode/ip.77.253.107.117) joined #pypy.23:45
piotrek_mhi, any1 here?23:45
Alex_Gaynordepends what you nee23:46
Alex_Gaynord23:46
pmav99 (~pmav99@79.103.98.226) joined #pypy.23:46
piotrek_mI am runnig debian, downloaded binaries 1.41, but pypy runs slower than python23:46
piotrek_m10x times slower23:46
Alex_Gaynorcan you show us your benchmark?23:46
piotrek_mok 1 sec23:47
piotrek_mthis is simple django app: http://82.160.204.87:8000/p/all/?page=3 <- python 23:48
piotrek_mhttp://82.160.204.87:8001/p/all/?page=3 <- pypy23:48
Alex_GaynorSQLite DB I assume?23:48
piotrek_myes23:49
piotrek_mbut sqlite takes only 80 ms23:49
piotrek_mi tried to compile from source but I failed twice23:50
piotrek_monce with python 2.623:50
Ademanpiotrek_m: did you measure that using cProfile? because cProfile used to (still does I think) ruin our performance23:50
Alex_Gaynordebug toolbar doesn't invoke cProfile23:51
Alex_Gaynorit doesn't call sys._getframe I think23:51
Alex_Gaynoryeah it does, that's bad for our performance23:51
Alex_Gaynorpiotrek_m: can you disbale DDT, and benchmark it using ab or siege or something?23:51
piotrek_mok 1 sec23:52
Ademanthat's a pretty nifty feature, that debug toolbar23:52
Alex_Gaynoryeah DDT is slick23:52
piotrek_mok DDT disabled, still the same thing23:53
piotrek_mpypy 2* times slower23:54
piotrek_mi am runing x64 btw, does it matter?23:54
Alex_Gaynorshouldn't23:54
Ademanwait I thought it was 10x slower before, that's a big difference :-p23:54
Alex_Gaynorok I tihnk we can lay that on SQLite being really slow on pypy, since it's pure python using ctypes, anto is working on speeding it up tons though23:54
hpk_ (~hpk@HSI-KBW-109-193-198-175.hsi7.kabel-badenwuerttemberg.de) left irc: Ping timeout: 250 seconds23:54
Ademando we have an ETA for jitffi?23:55
Alex_Gaynorwhen its ready! :)23:55
piotrek_mwhat db backend does work well with pypy?23:55
AdemanAlex_Gaynor: any idea what's up with MySQLdb ? since greg was interested in that?23:56
Alex_GaynorAdeman: not a clue23:56
piotrek_mok so I disabled db. still same problem23:56
Ademanthere's always your mysqldb module though :-p23:56
piotrek_mhttp://82.160.204.87:8001/p/ & http://82.160.204.87:8000/p/23:56
Ademaner, psycopg23:57
piotrek_mthis does only return rendertoresponse23:57
piotrek_mno db stuff here23:57
piotrek_msame problem23:57
piotrek_mpypy is slower (about 2x)23:57
Alex_Gaynorhow many times did you access the page?23:58
piotrek_m10023:58
Alex_Gaynortry hitting it 5000 times :)23:58
piotrek_mok, benchmaring with ab23:59
Ademansome of the inner loops should have been jitted already though23:59
--- Sat Mar 19 201100:00

Generated by irclog2html.py 2.9.2 by Marius Gedminas - find it at mg.pov.lt!