09:13 <cfbolz> mjacob: that’s fine

09:17 <cfbolz> mjacob: commit looks good

09:17 <cfbolz> mjacob: are you using Revdb?

09:19 <mjacob> cfbolz: let's put it that way: if i debug python code and have to do more than just looking around, i always use revdb if possible

09:19 <mjacob> i hate stepping through code with a traditional debugger, trying to find the source of some problem

09:20 <cfbolz> mjacob: awesome

09:21 <cfbolz> mjacob: does it work with cffi extensions?

09:22 <mjacob> cfbolz: yes, unless they use something which doesn't work with revdb

09:22 <mjacob> like https://bpa.st/NMWQ

09:23 <mjacob> everytime i upgrade revdb, i hit a problem in _cffi_ssl :(

09:24 <mjacob> but i've thrown away the old binary of revdb supporting python 3.5, so i need to find a fix

09:25 <cfbolz> mjacob: ssl is basically equally annoyed

09:25 <cfbolz> Annoying to all of us

10:08 <cfbolz> mjacob: you should make a video explaining how you use it, it's still really unknown

10:09 <mjacob> cfbolz: armin had made one years ago, and i don't see why anyone would watch a video i made :)

10:09 <cfbolz> because you actually use it ;-)

10:10 <mjacob> i planned giving a talk at the local python user group in leipzig, so i was not there (even as a visitor) since the start of the pandemic

10:10 <mjacob> s/so/but/

10:13 <cfbolz> mjacob: that sounds great

10:51 <ctismer> cfbolz: On object identity, https://doc.pypy.org/en/latest/cpython_differences.html#object-identity-of-primitive-values-is-and-id says `x + 1 is x + 1`.

10:51 <ctismer> On the other hand, small numbers in C seem not to preserve identity.

10:52 <ctismer> cfbolz https://www.irccloud.com/pastebin/v8MDKyMu/feature_select.cpp

10:52 <cfbolz> ctismer: yes, you cannot use pointer equality in C

10:53 <cfbolz> I mean you can

10:53 <cfbolz> But it'll be weird for ints

10:53 <cfbolz> And str, float, bytes

10:54 <cfbolz> Cpython will get a macro for that soon, I think

10:54 <cfbolz> (need to ask vstinner)

10:57 <ctismer> cfbolz: so there is no small integer that preserves identity? Ok, have to undo this shortcut. At the moment it doesn't matter either because selectable features cannot be enabled (need no be able to exchange the whole __dict__ of a type)

10:58 <cfbolz> ctismer: no, small integers aren't cached

10:59 <cfbolz> It interferes with the JIT unboxing of ints

12:24 <agronholm> hey guys, I found something strange with pypy3.8 that doesn't happen with pypy3.7 (7.3.7): https://github.com/agronholm/typeguard/runs/4198401860?check_suite_focus=true

12:25 <agronholm> basically there's a conditional dependency for typing_extensions on python_version < '3.9' which does not get installed on pypy3.8 but does for pypy3.7

12:25 <agronholm> at it correctly gets installed on cpython 3.8

12:27 <agronholm> this is with latest pip installed

12:34 <agronholm> looks like my install requirement specification was wrong; it's a mystery how it worked at all

12:35 <agronholm> could be that typing_extensions was brought in by a transitive dependency

15:27 <mattip> it seems our tests of app_main have been broken for a while on py3.7, py3.8

15:28 <mattip> because the untranslated tests are run with a cpython3.7 or cpython3.8, which are only available on the buildbot after an update to manylinux2014

15:30 <mattip> so now I discovered that CPython inserts '' into sys.path[0], where we insert the cwd when interactive

15:33 <mattip> or when run with stdin

15:34 <arigo> mjacob: about revdb: great to see someone using it. If you do, I'm up for fixing any new issue that invariably show up over time

15:35 <mattip> I am surprised we didn't get an issue about differences when running interactively about doing something like

15:36 <mattip> import os; os.chdir('dir/with/script'); import script

15:36 <mattip> which doesn't work on pypy but does work on CPython

15:37 <arigo> mjacob: specifically, about the "cannot be rendered as a plain raw address" error on revdb, I think I had a plan at some point in the past to fix it in general; with some motivation I could give it another try

17:54 <cfbolz> we really need to work on startup and warmup at some point again

17:54 <cfbolz> just unmarshalling is 3x slower in pypy

18:24 <cfbolz> ugh, why is there a new failing tests in the marshal module?

18:24 <cfbolz> mattip: any clue?

18:25 <cfbolz> http://buildbot.pypy.org/summary/longrepr?testname=AppTestMarshalMore.%28%29.test_unmarshal_ascii&builder=own-win-x86-64&build=554&mod=pypy.module.marshal.test.test_marshalimpl

18:33 <cfbolz> pffffff, it only happens when you delete pycs before running tests

18:40 <cfbolz> some surrogate problem

19:12 <mattip> is it only windows?

19:16 <mattip> only windows, and only py3.7

19:55 <cfbolz> mattip: no, it fails on my Linux laptop too

19:56 <mattip> weird, it is not failing on the buildbots afaict

19:57 <mattip> if I go to http://buildbot.pypy.org/summary which is a summary of all the branches and search for test_unmarshal_ascii, only the windows failure shows up

19:59 <cfbolz> Pretty weird

20:02 <mattip> maybe a bug in the host python2 that is running the test

20:03 <mattip> for windows that is Python 2.7.18[pypy-7.3.6-alpha], pytest-2.9.2, py-1.4.29, pluggy-0.3.1

20:05 <mattip> which I guess is the same pypy as for translation, hg hash 46fb2ebdc3ed, Jul 21 2021

20:24 <cfbolz> mattip: I've been running it on pypy2 even

20:24 <cfbolz> it's really about surrogates I think

20:29 <cfbolz> mattip: are the buildbot checkouts completely cleaned out between runs? pycs removed?

20:54 <cfbolz> mattip: I'll need your help with this, I am extreeeemely mystified

22:07 <mattip> cfbolz: both you and the windows buildbot are running the tests with pypy2 pytest.py ...

22:08 <cfbolz> right

22:08 <mattip> where all the other buildbots are running cpython2 ptest.py ...

22:08 <cfbolz> by now it looks like a JIT bug, honestly

22:08 <mattip> ahh, with jit --off it passes?

22:09 <cfbolz> yep

22:09 <cfbolz> I should sleep too :-/