Camelia, the Perl 6 bug

IRC log for #darcs, 2012-12-22

| Channels | #darcs index | Today | | Search | Google Search | Plain-Text | summary

All times shown according to UTC.

Time Nick Message
02:44 intripoon_ joined #darcs
02:53 schlaftier joined #darcs
04:54 mizu_no_oto joined #darcs
07:08 gpiero left #darcs
11:09 schlaftier joined #darcs
11:56 owst joined #darcs
13:42 schlaftier joined #darcs
13:49 favonia joined #darcs
13:51 donri joined #darcs
14:28 amgarchIn9 joined #darcs
14:46 owst Heffalump: did you raise a GHC ticket about the decoding seg-fault? I can't find it if so.
14:52 ManateeLazyCat joined #darcs
15:37 Heffalump owst: not yet, will do so today
15:37 Heffalump do you need the example code?
15:46 owst joined #darcs
15:47 owst No, I was just interested to see any comments :-)
15:47 owst Also, stupid linux wireless driver and home router!
15:47 owst ieee80211 phy0: brcms_c_prec_enq_head: No where to go, prec == 4
15:47 owst \o/
16:01 owst joined #darcs
16:05 alexsuraci_ joined #darcs
16:07 Heffalump owst: I think the "append junk then remove it" solution is looking plausible
16:07 Heffalump but I'm really not comfortable with it..
16:08 owst Have you determined what the minimum non-seg-faulty length of input is?
16:08 Heffalump 4 seems enough
16:09 owst 4 bytes?
16:09 Heffalump I wanted an entire UTF32 character to be on the safe side, so I didn't try for shorter
16:09 Heffalump yes
16:09 owst Right
16:09 owst How are you thinking of avoiding the "make an invalid sequence valid" issue?
16:10 owst I really should read up on UTF-* and understand what I'm talking about :-)
16:11 mndrix joined #darcs
16:19 mizu_no_oto joined #darcs
17:20 mizu_no_oto joined #darcs
17:48 Heffalump what do you mean?
17:53 owst Yesterday you were worried that appending junk might make a invalid sequence become valid?
17:58 Heffalump ah, I see
17:58 Heffalump my current plan is to count the characters you get from decoding the junk
17:59 Heffalump and then chop that off the end of the result of decoding (input ++ junk)
17:59 Heffalump chop that many characters, that is
17:59 Heffalump but - it looks like this problem may only affect UTF8 - and if so there are alternative decoders we can use in that case instead
17:59 Heffalump so I think the scope and reasons for the crash need to be characterised before picking a fix
18:00 Heffalump so I shall report the bug and see if that helps gain insight
18:00 owst Wasn't it a problem with attempting to decode iso-8859?
18:00 owst Yeah, sounds like a good plan]
18:03 Heffalump I don't think iso-8859 would have a problem, because it's not multi-byte
18:03 owst Oh. Then why did that error show up when reading the iso-8859 encoded inventory?
18:05 Heffalump because it's trying to read it as UTF8
18:05 Heffalump that's why it finds invalid sequences
18:06 Heffalump at least I think so - I don't fully understand how we decide what encoding to use for on-disk structures
18:06 owst ...so it does have a problem? :-)
18:06 Heffalump and how that interacts with pre-unicoding stuff
18:06 * owst is confused
18:06 owst right
18:06 owst do you mean at some point, darcs was "unicoded"?
18:07 Heffalump oh, you mean "attempting to decode as UTF8 where the input stream of bytes happens to actually be in ISO-8859"?
18:07 owst Ah, yes, that's what I mean. I assumed that's what the error message implied it was trying to do
18:11 Heffalump well, any invalid sequence is a problem - but the most likely source of invalid sequences is things that are actually encoded as something else
18:12 owst Indeed.
18:12 owst So was the answer that you don't know how darcs determines what something is encoded on disk as?
18:13 Heffalump I don't fully understand it. There's a comment in the source (in Darcs.Patch.Info) along the lines of trying UTF8 first
18:15 Heffalump I did actually extract the encoding/decoding code from Haskeline a while ago as a possible solution to the GHC 7.6 problem, but it seemed like a pretty messy solution
18:15 Heffalump perhaps I should resurrect that for the short-term though as it would provide a better guarantee of no behaviour changes
18:22 Heffalump (it basically involves having yet another set of iconv bindings on linux and to the win32 apis
18:32 amgarchIn9 joined #darcs
19:00 owst joined #darcs
19:50 gpiero joined #darcs
20:07 Heffalump http://hackage.haskell.org/trac/ghc/ticket/7522
20:11 owst joined #darcs
20:13 Heffalump owst: http://hackage.haskell.org/trac/ghc/ticket/7522
20:15 owst Cool
20:28 favonia joined #darcs
21:13 favonia joined #darcs
21:41 gpiero left #darcs
22:08 sm happy post-solstice all
22:09 sm JamesJRH: no, I haven't worked on it lately aside from ops
22:38 owst joined #darcs
22:42 schlaftier joined #darcs
23:22 owst joined #darcs
23:45 mizu_no_oto joined #darcs

| Channels | #darcs index | Today | | Search | Google Search | Plain-Text | summary