IRC Logs

Log of #ghostscript at

 <<<Back 1 day (to 2013/07/18)2013/07/19 
mvrhel_laptop whew. ok initial commit of windows phone work is up in my repos in a branch called winphone. lots of work to do still but it is coming along06:40.48 
  ray_laptop: up late writing emails to japan?06:41.36 
  head hurts from this windows phone stuff. calling it a night06:42.01 
kens chrisl ping07:31.13 
  Hmm, was it soemthing I said ?07:31.41 
  chrisl ping07:50.10 
chrisl kens: pong07:52.41 
kens chrisl could you try the file and setup described in bug 694423 please ?07:53.10 
  Also, what builds an executable called 'gsc' ? On my Linux it builds 'gs'07:53.30 
chrisl make so07:53.40 
kens OK I'll try that too thanks07:53.49 
  Currently I'm not able to reproduce the problem.07:54.45 
chrisl Given how specific it is, it might need 9.0707:55.31 
kens He says 'master'07:55.39 
  And teh GS outptu is clearly from teh 9.08 code07:55.55 
  I have noted in my reply (not applied yet) that he didn't state the SHA of teh commit he's using, but I'm using the absolte latest 7f764e507:56.40 
  Hmm, make so failed07:56.47 
chrisl Oh, yeh, I thought I'd read 9.07 somewhere.....07:56.59 
kens 'cannot find -lgs'07:57.03 
chrisl The file runs fine for me07:58.27 
kens OK well I'll note that in my reply. We can't fix it if we can't reproduce it.07:58.44 
  Probably we'll need his SHA at the very least and also need to know if he'sbuilding 64-bit I guess07:59.03 
  Thanks for trying it.07:59.56 
chrisl kens: sorry, I used the wrong command line - it does crash for me using the shared lib08:00.11 
kens Oh, then can you tell me why 'make so' fails for me08:00.26 
chrisl No, not really. You might need to "clean" and rerun autogen.sh08:01.01 
kens I'll try that when the VM finishes shutting down so I can relaunch it ;-)08:01.19 
chrisl But a monolithic build seg faults, too, so.....08:01.52 
kens Oh, nt for me08:01.59 
chrisl In cmd_put_list_op()08:02.57 
kens I'll just try a clean build08:03.01 
  So that's clist ? Makes it Ray's problem I guess08:03.18 
  OK make so worked this time08:05.19 
chrisl Then you need to do LD_PRELOAD=./ ./gsc08:06.50 
  Assuming you run it from the sobin directory08:07.05 
kens I was going to yes, why do I need to do that ?08:07.18 
kens is running just make at the moment08:07.28 
  well teh momolithic build still works for me08:08.48 
chrisl Normally, the dynamic linker will use its library search path to find the shared lib, so if you just run "./gsc" it will find the "system" one - usually /usr/lib08:08.55 
kens Ah, OK08:09.04 
chrisl Using the LD_PRELOAD= thing tells the linker to preload a specific library file, so it's already in memory and ready to link08:09.39 
kens shared object build works for me too.08:10.22 
  What version of gcc are you using ?08:10.29 
chrisl 4.7.208:10.38 
kens OK I'm using an older version 4.5.1-408:10.51 
chrisl It could also (or alternatively) be glibc version dependent08:11.42 
kens Yes, but I guess I dont' care, its not my problem :-)08:12.01 
  THanks for looking at it, I'm reassigning it to Ray on the basis that it seg faults in clist.08:12.23 
chrisl kens: mention in the bug thread that I can reproduce it, so if Ray struggles, he can assign it to me to take a look at08:13.00 
kens chrisl, I already did :-)08:13.11 
  clist doesn't seem to be a component, so it'll have to be graphics library.08:13.32 
chrisl Hmm, "graphics library" seems like too bug of a catch-all!08:14.41 
vtorri chrisl: something like that for the stdout cb seems reasonnable ? 08:14.44 
kens chrisl, yes, but I couldn't see anything more specific, I did put 'clist' in the comment though08:15.11 
chrisl vtorri: except the final return should be "return len" - other than that, it looks okay08:15.47 
vtorri chrisl: and if malloc fails, i return O ?08:16.20 
chrisl vtorri: it depends - do you want to just try to carry on if malloc fails, or potentially throw an error?08:16.55 
vtorri ha08:17.07 
  carry on08:17.17 
chrisl kens: I was just wondering if we should break it down into a few more options08:17.30 
  vtorri: so, both returns should "return len"08:17.43 
vtorri ok, thank you08:17.51 
kens chrisl you mean the options list ? Its probably not worth it, users almost always get it wrong (which is hardly surprising)08:18.02 
chrisl Oh, you should probably free the string, too!08:18.02 
vtorri oups08:18.09 
  thanks :)08:18.14 
chrisl kens: true, I suppose. It was more for our benefit08:18.29 
  vtorri: no08:18.32 
kens fetches coffee08:18.46 
chrisl_r61 Broadband is down :-(10:38.55 
vtorri chrisl_r61: what is the "no" above about ?10:40.32 
chrisl_r61 vtorri: read the next line - I just mistyped10:40.51 
vtorri ho, ok :)10:41.12 
chrisl_r61 vtorri: too little coffee at that time of the morning!10:41.45 
vtorri :)10:41.53 
  Including an abandoned public library *with all the books still in it*11:53.29 
kens is surprised these buildings haven't been subjected to arson11:53.55 
  Maybe all the vandals have left too11:55.07 
sebras Robin_Watts: _all_ the books was a slight exaggeration, but there are still some left which is surprising enough.12:00.39 
  oh... there were two. east side have more left.12:01.55 
kens Yes, 2 ex libraries12:02.09 
  (ex libris ? :-)12:02.21 
Robin_Watts tor7: ping13:36.52 
tor7 robin_watts: hi13:37.03 
Robin_Watts Did you get any further with the progressive stuff?13:37.05 
  I am going on holiday for just over 2 weeks starting next saturday, so I'd like to try to get it in as early as possible so I don't get any last minute problems to solve.13:37.52 
tor7 looking at the diffs now13:38.23 
  the diff from yesterday looks good13:40.10 
  I didn't think you meant "hail mary" name literally :)13:40.43 
  but I like it.13:40.49 
Robin_Watts I'm open to better names.13:41.23 
  (choosing decent names is always an interesting problem.)13:41.59 
tor7 the only potential problem with hail mary is its cultural specificity13:42.44 
Robin_Watts It's an internal name, not part of the API.13:43.00 
tor7 indeed. so I've got no objections there.13:43.09 
  were it part of the public API I'd be more wary of keeping funny names out13:43.28 
  and it does what it says on the box13:43.48 
  I think we can push now, if you've run it through the tests13:44.11 
  I have no more objections.13:44.20 
Robin_Watts ok, thanks.13:44.25 
  kens: Are you here?14:09.59 
kens yep14:10.03 
Robin_Watts kens: We have just started a discussion on skype that you might want to be involved in.14:10.20 
  We're using skype as a text-message system at the moment.14:10.35 
  So it might be worth you installing skype and getting an account even without having a microphone yet.14:10.52 
kens I have already done so14:11.16 
  and I do have a mic14:11.23 
Robin_Watts Ah. What is your skype address?14:11.31 
kens Have no idea how to actually use Skype though14:11.35 
Robin_Watts kens: Have you run it and signed in?14:11.49 
kens Dunno about addresses, my name is :-)....14:11.50 
  I've had it a while....14:12.04 
Robin_Watts I've just sent a contact request.14:13.02 
kens :-) Me too14:13.10 
  So, I have you as a contact....14:14.10 
Robin_Watts So you should now see a group conversation between lots of people?14:14.14 
kens Nope14:14.18 
Robin_Watts Under the Recent column?14:14.29 
kens There's something there14:14.41 
  I see a group at any rate14:14.54 
  and you are in it, and green14:15.03 
Robin_Watts Click on the group, and the conversation should appear in the main bit of the window.14:15.21 
kens I see you sayign stuff about me and Chris14:15.26 
Robin_Watts Right, so you haven't got the conversation history.14:15.38 
  Let me mail you that.14:15.41 
kens OK14:15.44 
henrys will kens be able to see what was typed before he joined?14:16.57 
Robin_Watts no. but I just mailed him it.14:17.07 
kens Network threw a wobblie, hust a moment14:17.23 
  OK mail arrived, let em read it14:17.49 
  OK, read that14:18.58 
henrys moist.von.lipwig is in our group Skype group?14:19.07 
kens Me14:19.11 
  No idea hwo to join group14:19.22 
Robin_Watts kens: You are in the group.14:19.31 
kens Oh, OK it doesn't show me myself ;-)14:19.41 
Robin_Watts Skype => Edit Profile, you can set your visible name to something... less intriguing. :)14:20.16 
kens ROFL14:20.27 
  OK now I can't see the message window any more14:22.26 
henrys so kens is moist.von14:22.50 
Robin_Watts henrys: He is.14:22.56 
kens I've had the accoutn a while14:23.00 
  But never used it for anything14:23.10 
Robin_Watts since before the operation.14:23.10 
kens I used to have an even older one but it expired and vanished14:23.23 
henrys moving along … I don't see chris14:23.32 
Robin_Watts broken broadband.14:23.42 
henrys oh right14:23.59 
kens This UI is terrible, how on earth do I get rid of my prifle now ?14:24.01 
Robin_Watts kens: Click back on the group.14:24.13 
kens I dopn't see a back.14:24.29 
Robin_Watts sorry. "click on the group again"14:24.39 
kens OK got rid of it by clicking randomly14:24.48 
  It looks like my name is more identifiable now, at least it is here14:25.26 
Robin_Watts kens: yes, your name is showing normally now.14:25.52 
henrys I'm going to try and get him on IRC and hopefully this will go away. I don't mind Skype but I'd rather not go back and forth and IRC more suites an open source setup14:26.18 
Robin_Watts kens: You might want to send contact requests to each of the people in the conversation.14:26.33 
  as they are not automatically added to your contacts.14:26.49 
kens I noticed that, but then htey might try and call me :-)14:26.59 
henrys kens:I want to see you send your Skype name to Joann14:27.56 
kens I did14:29.22 
Robin_Watts Tell her "It's a family name".14:29.36 
kens Its from the Uberwald region14:29.53 
  fixed now chrisl ?14:53.15 
chrisl Fingers crossed......14:53.23 
  The status website still claims it will be 18:10hrs before it's sorted14:53.49 
  Cor blimey - I just built gs, gpcl6 and gxps all entirely within the Ghostscript build system..... that seems like progress :-)15:04.52 
Robin_Watts chrisl: STOP NOW.15:10.15 
  enjoy your weekend in the glow of success.15:10.24 
  do NOT test any of them :)15:10.31 
chrisl That is my plan - I'm not even tidying up the white space issues now!15:10.48 
  It's been a while since I finished the week on a high note........15:11.33 
kens Its amazing what losing braodband can do :@)15:11.51 
chrisl I resent that implication!15:12.35 
kens Well I was thinking of people not bothering you, but....15:13.05 
chrisl That may have helped, but then, I also couldn't vent, so the laptop was nearly put to the sledgehammer15:14.05 
kens Hmm, I 'almost' have Indexed separation spaces working....15:14.25 
chrisl So, looking at the implementation of gsave/grestore, it looks to me like we can rely on the gstate being passed into fillpage as still being valid (taking account of GC changes) when we get to output_page - that should allow us to make the pcl3 devices not depend on psi15:19.55 
vtorri robin_watts: ping17:46.44 
  robin_watts_mac: ping17:48.14 
Robin_Watts pong17:48.28 
vtorri Robin_Watts: in mucbz (1.2) in cbz_read_zip_dir(), i can see :17:49.36 
  fz_seek(file, 0, 2);17:49.46 
  should it be more readable to use SEEK_END instead of 2 ?17:50.11 
Robin_Watts That should be fz_seek(file, 0, SEEK_END) I suspect.17:50.14 
  Yes. I didn't write that :)17:50.24 
dogisfat I am still working on a trace driver and am having problems with gradients/fills. The default trace driver reports unknown for the color of a filled shape. I wish to dump the pattern/gradient to a file.17:53.02 
  Where would be the best place to look to understand how to add that functionality?17:53.17 
Robin_Watts dogisfat: This is gs, right?17:54.05 
dogisfat Yes17:54.17 
Robin_Watts Which device entry point ?17:54.32 
  Hi zeniko17:54.38 
dogisfat Robin_Watts: I am not sure what that means17:54.52 
zeniko Robin_Watts: I'm back (just for the weekend)17:54.58 
  Is that tomorrow that you're leaving for your holidays yourself?17:55.21 
Robin_Watts zeniko: next saturday. 17:55.31 
  i.e. 8 days.17:55.35 
zeniko We might have some more time next Friday, then17:56.29 
  I've tried running the cluster with a modified submodule17:56.47 
  and that didn't work at all17:56.54 
Robin_Watts dogisfat: You're implementing a device, right? That device has lots of functions that can be called.17:56.58 
  zeniko: Ah.17:57.08 
dogisfat Yes I am17:57.09 
zeniko so we'll have to come back to one of your two options as well (of which I'd prefer the "preparing the cluster run on casper" one)17:57.26 
Robin_Watts zeniko: I'm sure we can find a way around that. How familiar are you with unix ?17:57.40 
  dogisfat: Right, so which one of your functions are you in when it reports that it doesn't know the color ?17:58.03 
zeniko it's been a while, but I should be fine17:58.15 
Robin_Watts zeniko: OK, I've had people who have had no experience of unix (and were not even prepared to try) before.17:58.36 
  So... ssh in to casper.17:59.01 
dogisfat Robbin_Watts: *_fill_path17:59.19 
Robin_Watts and make yourself a copy of mupdf.git to work in. I have ~robin/sauce/mupdf.git for this purpose.17:59.42 
  mkdir sauce && git clone /home/git/mupdf.git mupdf.git or something like that.18:00.15 
zeniko Robin_Watts: done18:00.34 
Robin_Watts OK. Now change directory into mupdf.git18:00.55 
  Now do: /home/regression/cluster/gitbridge/clusterpush.pl18:02.32 
zeniko seems to have worked ("queueing")18:03.17 
Robin_Watts OK, so that's the magic invocation to send the current directory to the cluster for testing.18:03.47 
  so you can try your submodule fiddling there and try it again.18:04.09 
  Ah, this gives me an opportunity to demonstrate another feature of the cluster :)18:04.27 
  If you look on the dashboard you'll see that your job is running.18:04.40 
  If you do: /home/regression/cluster/gitbridge/ abort18:04.53 
  it should kill it (likewise "git cluster abort" from your windows machine)18:05.10 
  It takes anything up to 1 minute for the cluster to react to your commands.18:05.33 
zeniko so it does18:05.39 
Robin_Watts so if you queue and cancel and requeue, too fast the cluster can get confused.18:05.51 
zeniko what's the difference between and git cluster?18:06.31 
  i.e. why might the former work for submodule fiddling while the latter doesn't?18:06.46 
Robin_Watts git cluster does a 'git stash' and then transfers that across to a staging area on casper.18:07.51 
  It then calls clusterpush.pl18:08.11 
  which transfers the staging area to the cluster area using rsync.18:08.30 
  The transfer using git stash specifically doesn't move any of the submodule stuff across.18:09.02 
  rsync doesn't care, it just sends all the directories.18:09.13 
  So, we stand a chance that the version might work, where the git push version doesn't. Make sense ?18:09.39 
henrys zeniko:do you want to have a look at this one: or should I ask shelly, it is reproducible in mupdf18:09.42 
  362 and 363 as well.18:12.59 
zeniko Robin_Watts: makes sense, I'll give it a try and if for any reason it doesn't work, either, we can look further next Friday (or after your holidays)18:15.45 
Robin_Watts zeniko: OK. I am generally around here most weekends and evenings on and off, so feel free to ping me out of hours. If I'm here, I'll answer.18:16.36 
dogisfat Robin_Watts: I just realized I tagged you with the wrong spelling. the function is *_fill_path18:16.36 
Robin_Watts dogisfat: Sorry, got distracted with other things.18:16.48 
  Let me look.18:16.55 
zeniko henrys: I've been having difficulties reproducing these issues on my 32-bit machine18:17.10 
dogisfat Robin_Watts: No problem, I should be a second priority 18:17.18 
zeniko they might be 64-bit specific, so maybe shelly has better luck18:17.31 
Robin_Watts dogisfat: So, you're getting called, and what are you finding to be unset?18:17.53 
zeniko once I've set up cluster testing with submodules, I might be able to look into them after all, though18:17.56 
  so I'd say that same as with the other issues, let's leave them to whoever of us who gets to them first18:18.38 
henrys zeniko:I'll give them a try and if I don't see anything I'll ship them back to marcosw, generally if you can't reproduce something put a comment in bug tracker and I'll one of us will move it along18:19.25 
Robin_Watts dogisfat: I started to write a 'spy' device a while ago.18:19.56;a=commitdiff;h=cc6a350093009c885ef74d49b6749d7f5d22116018:19.58 
  The idea is that in a debugger while running gs, you can attach it to a device and it will trace the calls that go through it.18:20.36 
  It works, but I haven't implemented all the entrypoints yet.18:20.51 
  But the first function in it might be useful to you.18:21.08 
dogisfat Robin_Watts: Interesting18:21.15 
zeniko henrys: I'll go through the list and add the notes, then, once I've got some more time at hand (maybe after the holiday season)18:21.16 
dogisfat Robin_Watts: So that looks like the same code from the trace device. In my case I get **unknown** out18:22.18 
Robin_Watts dogisfat: Yeah, I stole a lot of the code :)18:23.20 
dogisfat Robin_Watts: I am in the same boat18:23.41 
  Robin_Watts: This is an example file I am working with 
Robin_Watts Can you run it in a debugger and catch it at the point at which it hits the 'unknown' case?18:25.14 
dogisfat Robin_Watts: Should I try and run it through the spy or trace? It may take me a bit to get spy working18:25.49 
Robin_Watts dogisfat: Through your device.18:25.57 
  gx_dc_type_pattern is a possibility.18:26.02 
  gx_dc_type_devn another18:26.17 
dogisfat Robin_Watts: This test file is better because it includes a gradient as well as a better. 
Robin_Watts I strongly suspect that it's a gx_dc_type_pattern18:27.12 
  or a gx_dc_type_pattern218:27.37 
dogisfat Robin_Watts: Let me implement that18:28.22 
  Robin_Watts: How do I access those types i.e. where are they defined?18:29.21 
Robin_Watts gsdcolor.h ?18:30.25 
  What platform are you working on ?18:30.38 
dogisfat OSX18:30.50 
Robin_Watts then grep is your friend :)18:31.00 
zeniko Robin_Watts: does the cluster always recompile everything before running?18:31.03 
Robin_Watts zeniko: Yes.18:31.10 
zeniko and comparefiles/*.pdf are all tested in a mupdf run as well?18:31.41 
Robin_Watts I think so. Let me check.18:31.51 
dogisfat Robin_Watts: Yeah, I realised the stupidity of my question18:32.02 
zeniko Thing is, I've just reduced the maximum number of color components down to 8 which should cause visible regressions in at least two files - but it didn't18:32.32 
marcosw zeniko: yes, the comparefiles/*.pdf are run through mupdf18:32.50 
Robin_Watts zeniko: marcosw wrote the cluster code (and I've just prodded at the edges of it). He's a better authority than me on stuff like this :)18:33.49 
zeniko marcosw: ah, I see that in my run some of the comparefiles have started producing errors (Error_reading_input_file) which however for my change doesn't seem to make sense, either18:34.13 
marcosw zeniko: let me look at the logs and see if the error messages help clarify what's going on.18:34.44 
Robin_Watts zeniko: So, time to demonstrate another useful feature...18:35.03 
  So, let's look at Bug691002.pdf18:35.29 
zeniko Robin_Watts: that's actually one of the files where I'd have expected differences18:35.59 
Robin_Watts Click on 'search' to the right of 'zeniko'18:36.04 
  and enter Bug691002.pdf18:36.14 
  This hunts through the (huge) log files we get from the cluster for the sections for files that match that name.18:36.46 
  And you can see that we're getting errors saying "too many components in colorspace"18:37.10 
  So mudraw exits with a non zero error code.18:37.23 
  so the cluster reports it as an error.18:37.32 
  so, I think that explains it?18:38.00 
marcosw zeniko: what he said. I always look at the log files on casper, but using the web interface is clearly the more user friendly and faster way of doing it.18:38.21 
zeniko That explains it indeed - so in order to get differences, I'd have to make mudraw return 0 despite cookie.errors18:38.35 
dogisfat Robin_Watts: Ok so it is a gx_dc_type_pattern18:38.39 
zeniko Robin_Watts: marcosw: Thanks for the help!18:38.59 
Robin_Watts You can get to the full files by clicking the '+' sign to the right of zeniko18:39.02 
marcosw Robin_Watts: you can a yesterday afternoon just after 3:00pm your time that generated an email that was too large to send. I'm trying to track down those sort of errors so that you at least get a partial log file. Do you recall what you did for that clusterpush?18:39.16 
Robin_Watts but the search facility is much easier.18:39.25 
marcosw and did you get back any email at all?18:39.40 
Robin_Watts marcosw: Was it a gs one?18:39.41 
  I suspect I pushed a version of the source with some per clipped pixel printfs :)18:40.08 
marcosw yes18:40.11 
Robin_Watts so my bad, sorry.18:40.25 
  zeniko: no worries.18:40.39 
marcosw No problem at all. I'll add code to cut the email to some reasonable number of bytes so that you at least get back something that indicates the cluster finished.18:41.13 
Robin_Watts dogisfat: And to answer your next question, I'm not sure what's in a gx_dc_type_pattern - you'll have to look at the code, sorry!18:41.25 
dogisfat Robin_Watts: I was already looking into it, thanks for the direction!18:42.21 
Robin_Watts we have several sorts of patterns in gs.18:42.47 
  sometimes they get rendered to tiles.18:42.54 
  sometimes the tiles have transparency.18:42.59 
  and sometimes we keep the patterns as clists.18:43.11 
  (clist = command lists)18:43.20 
dogisfat Robin_Watts: Thank you so much!18:49.03 
Robin_Watts no worries.18:49.37 
mvrhel_laptop bbiaw18:58.58 
henrys Robin_Watts:can somebody in the mupdf group look at 694311 comment 4 - those files really shouldn't be broken or at at least they should have separate bugs.21:01.05 
  if altona broke in regression with openjpeg2 it's a blocker for the release.21:01.49 
ray_laptop I don't understand marcosw's reversion to put C++ comments back in gscie.c and gxhintn.c21:03.39 
Robin_Watts henrys: I'll try to have a look.22:59.58 
 Forward 1 day (to 2013/07/20)>>>