Parrot 5.1.0 "Zombie Parrot" | parrot.org/ | Log: irclog.perlgeek.de/parrot | #parrotsketch meeting Tuesday 19:30 UTC
Set by moderator on 20 February 2013.
00:11 mtk joined 00:54 ctult_ joined
cotto ~~ 02:35
02:49 kid51 joined
dalek rrot: 28129d6 | rurban++ | config/gen/makefiles/root.in:
[GH #933] Fix wrong blib rpath in installable_pdump

Thanks to he32++ for detecting this		 03:11
04:41 preflex_ joined 06:31 Liz joined 06:32 woolfy joined 06:43 Liz joined, woolfy_ joined 07:03 woolfy joined, Liz joined 07:05 Mike-PerlRecruiter_ joined 07:50 zby_home joined
Coke (re: ops2c) - my take is that if we want a higher level language for this than PIR, use winxed. 10:06
(but don't bother with it now because there's so much other crap to do) 10:20
moritz wtf, using Socket.send, I get "Method 'seek' not implemented for type Socket" 11:15
/* If we are writing to a r/w handle, advance the pointer in the 11:19
associated read-buffer since we're overwriting those characters. */
Parrot_io_buffer_advance_position(interp, read_buffer, bytes_written);
maybe that's what causes it?
hm, no 11:20
11:34 not_gerd joined
not_gerd o/ 11:34
moritz: the culprit is io_sync_buffers_for_write() 11:35
that's the wrong thing to do for IO objects like sockets where input and output are decoupled
moritz so is that an architectural issue? 11:43
not_gerd yes - the semantics are wrong 11:44
it blows up here because sockets actually can't support random access 11:45
pipes probably have the same issue 11:46
quick fix would be to return from the io_sync_buffers_* functions if the handle is of type Socket 11:55
a proper fix would need a new Handle method, overwritten in Socket
something like the isatty() method 11:57
or a sync method that is a noop in case of Socket PMCs 11:58
moritz or maybe a "seekable" method
or that, yes
not_gerd well, seekable and coupling of input/output channels is imo orthogonal 11:59
no idea if that distinction matters in practice, but it would be 'the right thing' to do 12:00
moritz I've opened github.com/parrot/parrot/issues/938 for this issue
12:52 mtk joined 13:08 kid51 joined
dalek kudo/nom: c255f1d | (Tobias Leich)++ | src/ (2 files):
interpolation of e.g. regexes in arrays in regexes
 13:51
14:21 Psyche^ joined 14:25 PacoAir joined 14:50 kid51_ joined 14:53 eternaleye left 14:54 eternaleye joined
dalek p: 21a6b2c | (Arne Skjærholt)++ | src/ops/nqp_dyncall.ops:
Fix braino in nqp_dyncall.ops.
 15:02
15:04 kid51 joined
dalek p: f03f15a | moritz++ | / (3 files):
bump VERSIOn to 2013.02.1

turns out the native call support was mostly broken in 2013.02		 15:15
kudo/nom: 952d407 | moritz++ | / (2 files):
bump VERSION and NQP_REVISION to 2013.02.1

this gets us non-broken native call suport		 15:18
15:27 PacoAir_ joined 15:37 kid51_ joined 15:44 PacoAir joined
dalek kudo/nom: ff2ceee | (Tobias Leich)++ | src/core/Cursor.pm:
force regexes in array to match at pos 0 only
 15:55
cotto ~~ 16:53
16:59 xcombelle joined 17:09 kid51_ joined
rurban seeking on a socket? this makes no sense 17:22
cotto That seems to be the point. 17:29
rurban: are you working on a patch? 17:32
rurban no. the error is correct 17:33
if someone wants to seek on a socket, he has to do the buffering by himself
17:34 kid51 joined
rurban socket.seek violates every unix principle 17:34
cotto moritz ran into it while using socket.send, which is a more traditional use case 17:35
rurban I implemented sockopts on a branch
cotto socket.seek is obviously wrong
Hmm. The other 3 position-related socket vtable slots are noops. What about just making seek an noop? 17:56
*a noop
benabik Where does writing to a socket end up calling seek? 18:05
cotto io_sync_buffers_for_write 18:07
18:07 not_gerd joined
not_gerd o/ 18:07
cotto hio not_gerd
I was hoping you'd drop by. 18:08
not_gerd ordinary read/write can call io_sync_buffers_for_read/write which in turn call seek
benabik _read doesn't call seek
not_gerd (which of course doesn't work for sockets/pipes)
it's still wrong to sync input and output if they are decoupled 18:09
we're trying to sync channels which shouldn't be sync'd
that one of these functions calls seek on IO objects which don't support the operation is only a side effect 18:10
cotto I'm playing with adding a PIO_VF_SEEKABLE flag
benabik I don't think playing with seek is the solution.
I think we want a flag marking that the I/O is tied together.
And set it to false for pipes and sockets
not_gerd my idea was making the sync functions polymorphic (eg by adding them to IO_VTABLE) and just make them a noop for sockets and pipes 18:11
(which I already mentioned in a comment to the ticket)
cotto 3 out of 4 position-related vtable slots are already noops for socket
benabik cotto: I agree with not_gerd. Problem is higher. Shouldn't be syncing the buffers in the first place. 18:12
cotto benabik: sure. The questions is which level should it be fixed at.
benabik not_gerd: I think a simple yes/no flag is better. I can't think of any other behaviors we want. 18:13
cotto wrong fix, but seems to work if we can't come up with something better: gist.github.com/cotto/5020729
benabik Wait...
18:13 zby_home joined
benabik Why does the socket have a write buffer? 18:14
It doesn't set PIO_VF_DEFAULT_WRITE_BUF
cotto not_gerd: do you have the tuits to put together a patch that makes sync a proper polymorphic function? 18:21
not_gerd I can look into that, but probably not today 18:22
cotto ok
benabik Condo: I'm worried that making seek a no op will just hide getting the data wrong because of the offers. 18:25
*buffers
*cotto
not_gerd benabik: the idea is to add vtable entries for the syn functions and make these noops, not seek 18:26
a flag PIO_VF_SYNC_BUFFERS or somesuch as you suggested would also work
condo Now look what you've done. 18:27
cotto not_gerd & benabik: I like that better. moritz++ says that the fix isn't urgent so it won't hurt anything if it doesn't happen today. 18:29
benabik Still curious how the socket ends up with a write buffer. 18:31
not_gerd too
cotto looks like the buffers come from the Socket PMC's inheritance from Handle 18:47
benabik cotto: It should have a write_buffer attribute, but it shouldn't get initialized. 18:50
Or rather initialized to not null.
cotto It is initialized to null. 18:51
not_gerd cotto: and as long as PIO_VF_DEFAULT_WRITE_BUF is not set, it should remain so 18:55
benabik Oh. sync_for_write is conditioned on there being a _read_ buffer.
Nvm.
not_gerd made the same thinko 18:56
19:02 contingencyplan joined
not_gerd what should the flag be called: SYNC_BUFFERS, SYNC_RW or SYNC_IO 19:02
(likes SYNC_IO best) 19:03
cotto Votes will be accepted in the form of a working patch. 19:06
19:08 Mike-PerlRecruiter_ joined
not_gerd cotto: you already did the path, just replace SEEKABLE with SYNC_IO 19:11
^patch
cotto not_gerd: you don't think that sync should be abstracted into its own VTABLE slot?
not_gerd cotto: I think userhandle bypasses the vtable completely, ie only filehandle needs synchronization 19:13
adding vtable slots which will never be used probably doesn't make much sense
cotto not_gerd: not stringhandle too? 19:14
not_gerd stringhandle should get it as well, yes
I would be surprised if stringhandles are buffered, though 19:15
so the codepath will probably never be triggered
but your patch already does that
cotto yes 19:16
not_gerd just rename the flag and change the comment to something like "Input and output channels need to be kept in sync, which affects buffering" and you should be done
cotto not_gerd: just verifying w/ moritz
19:24 benabik joined
dalek rrot/smarter-io-sync: 10d8851 | cotto++ | / (4 files):
[io] only sync buffers for IO types where it makes sense
 19:27
moritz cotto: patching file src/io/utilities.c 19:28
patch unexpectedly ends in middle of line
Hunk #2 succeeded at 548 with fuzz 1.
cotto moritz: try the branch 19:29
It's the 21st century. I should be doing something better than gisting diffs. 19:30
moritz thanks, trying... 19:37
(slow machine, will take some time to rebuild all of parrot/nqp/rakudo/panda/Net::IRC 19:38
cotto You can also cheat and just install an updated Parrot.
moritz no, I can't 19:45
because my rakudo used parrot 4.10
and that has different installation paths 19:46
cotto ok 19:47
moritz cotto: that branch seems to work 20:20
cotto Is that what ilbot6 is using?
moritz yes
cotto nice
I'll give NQP and Rakudo's tests a spin and merge if it looks like nothing broke. 20:21
not_gerd cotto++ 20:27
bye, #parrot
20:27 not_gerd left
dalek p: b752c14 | jnthn++ | src/QAST/Operations.nqp:
Fix thinkos.
 20:38
p: 8d7d9be | jnthn++ | src/stage0/ (9 files):
Bootstrap update thanks to thinkos.
p: 8d1f7f6 | jnthn++ | src/ (4 files):
Replace some more pir:: with nqp::.
21:08 benabik joined 21:19 Khisanth joined
dalek kudo/nom: cfee628 | moritz++ | src/core/Mu.pm:
switch type name gistification from Type() to (Type)
 21:26
21:55 perlite_ joined 22:47 PacoAir joined
dalek p: 30d5020 | jnthn++ | src/ops/nqp.ops:
Avoid returning a C NULL.
 22:54
p: cd22fa9 | jnthn++ | src/NQP/Actions.pm:
Resolve a bunch of package names at compile time.

Cuts down on code output rather significantly in certain cases, of note in QAST::Compiler.
23:54 kid51 joined

Powered by Cro and the Raku® Programming Language.

Please report any issues / comments / feature requests as an issue on App::Raku::Log.

Thank you!