These are chat archives for dropbox/pyston

12th
Jan 2016
Marius Wachtler
@undingen
Jan 12 2016 10:31
sorry at the moment we don't have concrete plans for adding python3 support
:-(
Endre Karlson
@ekarlso
Jan 12 2016 11:20
@undingen do you guys use eventlet or something for concurrency ?
Marius Wachtler
@undingen
Jan 12 2016 11:32
I'm not sure but I read that it is using greenlet and I if I remember correctly this lib needs changes to get it to run with pyston which nobody made yet. So currently we probably can't run it.
Endre Karlson
@ekarlso
Jan 12 2016 11:34
@undingen what py stack u guys use then ?
Marius Wachtler
@undingen
Jan 12 2016 11:36
sorry no idea
Endre Karlson
@ekarlso
Jan 12 2016 11:47
@undingen u work on python there but dunno the code ? :smile:
Sun
@Daetalus
Jan 12 2016 11:54
I think this is because they working on a compiler/interpreter it self. Most focus on C++. For example. You can ask the Python user about what Python stack they use. But you ask the question to CPython developers is a bit weird.
Endre Karlson
@ekarlso
Jan 12 2016 11:59
@Daetalus ok :)
An Long
@aisk
Jan 12 2016 14:43
The thread_forking.py test failed some times on my machine but not often 😥
Marius Wachtler
@undingen
Jan 12 2016 14:52
:-(, what error are you getting?
An Long
@aisk
Jan 12 2016 15:00
5:                    thread_forking.py    FAILED (bad output)
5: --- /tmp/expected_L_QOrT    2016-01-12 14:39:31.363068199 +0000
5: +++ /tmp/received_BhbUrn    2016-01-12 14:39:31.363068199 +0000
5: @@ -1,4 +1,3 @@
5:  done
5:  done
5:  done
5: -done
Sun
@Daetalus
Jan 12 2016 15:01
I think maybe you run Pyston tests on a IDE command line?
An Long
@aisk
Jan 12 2016 15:02
in a vagrant terminal
rerun the test case, it disapeared
Marius Wachtler
@undingen
Jan 12 2016 15:19
this looks like a quite serious threading issue or at least I can't think of a reason why it can finish while only printing "done" 3 times..
An Long
@aisk
Jan 12 2016 15:23
I think this PR can be reviewed again dropbox/pyston#983 , the make check passed on my machine ..
Marius Wachtler
@undingen
Jan 12 2016 15:25
looks good, but I will wait until the tests finished
I would start with making the changes to get cpythons ast test working. Or would you like to make this changes?
An Long
@aisk
Jan 12 2016 15:29
Now the ast test is almost working, just except the syntax check in parse ast phase, I think we should remove them and just check it in cfg phase.
I just removed them, every thing works fine, except the try_continue test failed while using the -X param(I think it’s using cpython’s parser)
Marius Wachtler
@undingen
Jan 12 2016 15:33
yes that's what I'm talking about. This requires to move the error checking into the CFG phase. But you will encounter a problem: the cfg is currently generated lazily on the first call to the function. We must change this and make sure that if an error is encountered while generating the cfg we pass it correctly to the caller...
Are you still interested in doing this? :-D
An Long
@aisk
Jan 12 2016 15:34
Yes I’m glad to do this 😁
Marius Wachtler
@undingen
Jan 12 2016 15:34
cool!
An Long
@aisk
Jan 12 2016 15:35
I think this time I will do this more quickly as I can, because I don’t want to blocking other's work again 😅
And I think we should make pyston not to generate the cfg lazily?
Will this cause performance reduce in some situation?
Marius Wachtler
@undingen
Jan 12 2016 15:40
exactly. so you will have to make sure we call computeCFG at the same location where cpython would call PyAST_Compile
It will reduce the perf when generating a lot of functions and never calling them. I think this could change the import perf
If this is really changing the perf a lot we may need to fall back to the lazy approach but add a special syntax check pass instead
An Long
@aisk
Jan 12 2016 15:44
Got it 👌
I think it’s caused by the bad networking in travis
Marius Wachtler
@undingen
Jan 12 2016 15:48
Yes I think it's a flaky test. I sadly can't restart it at the moment because the button stopped working today :-D
An Long
@aisk
Jan 12 2016 15:49
ok, I’ve pushed a new empty commit
Marius Wachtler
@undingen
Jan 12 2016 16:19
@aisk I added a modifed cpython ast test in dropbox/pyston#1048 to get better testing until we can support the whole test
Sun
@Daetalus
Jan 12 2016 18:45
@rudi-c Hi! Would you mind to find time to finish dropbox/pyston#1027 and dropbox/pyston#1028 ? Just the Pyston changes, I can handle the NumPy patching. And I am happy to take over if you are busy.