<kenaan>
mjacob default 1a274c144a73 /rpython/rlib/rgc.py: Raise ValueError in nonmoving_raw_ptr_for_resizable_list() if running on RevDB. On RevDB, raw and GC'ed memory mu...
<mjacob>
i'm also going to be late... let's do the break day today? ;)
<tos9>
mattip / mjacob: also looking now... not sure I understand the hackmd notes yet or how to read buildbot :)
mattip has quit [Remote host closed the connection]
<mjacob>
tos9: the app-level test is easy
<mjacob>
do you remember what you did in 51b646896e2?
<tos9>
I think one of those (the segfault) is probably easy to fix, there's a conditional missing from Compressor's copy that's in Decompressor
<mjacob>
which conditional do you mean?
<tos9>
mjacob: Decompressor's copy does if not self.stream: raise zlib_error(space, "decompressor object already flushed") before it tries to call inflateCopy
<tos9>
(which was something armin noticed we needed to do)
<tos9>
But compressor doesn't do that
<tos9>
what's the equivalent of git show so I can see what that revision is
<mjacob>
instead in pypy we raw-allocate the structure and store a pointer in the pypy object
<tos9>
Ahh ok
<tos9>
mjacob: so in that case the conditional just shouldn't be in Decompressor and should be in Compressor?
<tos9>
Probably need to have tests for calling copy after flush and then that will fall out
<mjacob>
yes, an app-level test would be a good start
ronan has joined #pypy
* tos9
nods
<mjacob>
yes, just moving the conditional to Compressor would work
<mjacob>
but note that cpython does something different here: it also calls deflateEnd(), but the pointer to the zstream stays constant
<mjacob>
then, if calling copy(), zlib complains that the zstream internal state is messed up
<mjacob>
it doesn't really matter, though
<mjacob>
tos9: can you add a failing tests as a commit and push? then i can play around as well.
<tos9>
mjacob: yes, trying to make sure I follow the last thing you said as well -- meaning, you get a better error message (one that comes out of zlib directly) on CPython?
<mjacob>
tos9: just different ;)
<mjacob>
the check that you added in the wrong place is actually more informative
<tos9>
oh I see
<tos9>
you get nonsense like ValueError: Inconsistent stream state
<tos9>
yeah ok our error is better :)
<mjacob>
ah, but i think it should be valueerror instaed of zlib_error
<tos9>
yes seems like it
<mjacob>
no clue, why cpython raises both in different places
JohnWitulski has joined #pypy
<tos9>
mjacob: pushed the new test
<kenaan>
Julian Berman zlib-copying-redux bf5838a3edac /pypy/module/zlib/test/test_zlib.py: Add a test for copying flushed zlib compressobjs.
<kenaan>
andrewjlawrence default f8548d5c059a /rpython/: Fixed nan sign on windows
<tos9>
mjacob: (which I think this commit seems to fix now ^)
<tos9>
well not that one but bitbucket is slow as hell
<kenaan>
Julian Berman zlib-copying-redux 6c3c1a0558e6 /pypy/module/zlib/interp_zlib.py: Move our check for stream being nullptr to the right place. It's compressobj that sets this to ...
<mjacob>
tos9: looks good
<mjacob>
tos9: about the interp2app failure: do you understand how the interp2app thing - introduced in 51b646896e26 - works?
mattip has joined #pypy
<tos9>
mjacob: no not particularly (well I understand I think what it means to do), but I think I understand that I need a `if cls.runappdirect: pytest.skip("these tests cannot work with -A")` ?
<mjacob>
not quite
<mjacob>
skipping the single test is fine
<mjacob>
no need to skip all tests with runappdirect
<mjacob>
two tests should be skipped, actually
<tos9>
right that's what I'm trying to figure out, how to just skip those two
<tos9>
mjacob: I'm sitting with carl now too
<tos9>
(though now we're gonna do planning)
<tos9>
mjacob: so if you have a suggestion I'm all ears
<tos9>
do the 2 we want to skip need to be extracted to their own AppTest class?
<tos9>
or is there some method-level skip
<mjacob>
"if self.runappdirect: skip("can't run with -A")" will do
<tos9>
ah, within the body?
<tos9>
yeah ok
nunatak has joined #pypy
<kenaan>
Julian Berman zlib-copying-redux ab1d3b141472 /pypy/module/zlib/test/test_zlib.py: Skip the tests that need to rpython-munge when we run app-level.
<tos9>
mjacob: seem OK ^?
<mjacob>
tos9: seems good to go
<tos9>
mjacob: ok, gonna let antocuni teach me how to kick the buildbot
<tos9>
mjacob: (Thanks!)
<mjacob>
tos9: no need to kick the buildbot for a fix this small
<mjacob>
(however, you probably should have done it before merging the original branch)
<mjacob>
before merging, don't forget the whatsapp-snippet and closing the branch: ``hg ci --close -m "Close to-be-merged branch."``
<tos9>
whatsapp-snippet :)
<mjacob>
WTF, what's going on with me
<mjacob>
whatsnew snippet of course ;)
<tos9>
mjacob: you're not here
<mjacob>
not yet, will be there in the afternoon
<tos9>
mjacob: how do I interpret c5c3ad13d149
<tos9>
Did I break translation because stream became nullable?
<tos9>
I thought that was OK in RPython
<mjacob>
you can't mix None and raw pointers
<mjacob>
you could have used rzlib.null_stream instead of None
nunatak has quit [Quit: Leaving]
<tos9>
a ha...
<kenaan>
Julian Berman zlib-copying-redux 9b1847feb4f2 /pypy/doc/whatsnew-head.rst: Add the whatsnew snippet.
<mjacob>
but i think that constructing the zstream in the factory function is better anyway
* tos9
nods
<tos9>
mjacob: do I need to go back and close zlib-copying too? I didn't do that yesterday
<mjacob>
someone will do the house cleaning for you at some point
<mjacob>
so don't worry to much about the old branch?
<mjacob>
didn't mean to type the '?'
<kenaan>
Julian Berman zlib-copying-redux 32412a37b57f /: Close to-be-merged branch.
<kenaan>
Julian Berman default fee9d13fa77c /pypy/module/zlib/test/test_zlib.py: This appears to be necessary to actually have runappdirect available for app tests.
<kenaan>
antocuni default 4c920e4d6d68 /pypy/tool/release/repackage.sh: commit the tweaked version of the script which I used for the release
<mjacob>
fijal: did you leave already?
<mjacob>
fijal: keep me updated, i'll arrive around 19:00 and prefer the vegetarian group
<fijal>
mjacob: ok, we did not leave
marky1991 has joined #pypy
<kenaan>
antocuni pypy.org[extradoc] 3d49c68495b0 /source/: (antocuni, mattip): update the website for the new release; only the source, the HTML will be regener...