IRC Logs

Log of #ghostscript at irc.freenode.net.

Search:
 <<<Back 1 day (to 2013/04/15)2013/04/16 
mvrhel_laptop ick in adding in the monitoring for color in the cm I see an MT issue that could have come up when we are doing the no color management case and the replace color stuff04:58.47 
  ok. all the changing of the color management procs are in place now for color monitoring. now to add in the monitoring05:19.37 
  ugh some issues to think about when the profile is identiy05:31.16 
  when the link is the identity05:31.26 
  we need to monitor but we have special paths to avoid cm05:31.43 
  I guess I will need to turn of the is_identity flag, forcing color management, and then deal with the data conversions that may exist 05:32.31 
  but not do any color managment05:32.43 
  sigh. it looks like the link is going to need to have some information about the source color space packed into it. 05:53.05 
  oh actually I think it already is in there with the buffer description05:55.15 
  nice VS crash..05:59.16 
  had not had that in a while05:59.33 
  henrys: so under $200 for me to fly to the open printing meeting so not too bad06:41.40 
  good night all06:53.06 
kens chrisl ping08:44.39 
chrisl kens: pong08:46.59 
kens got a build error, not sure why. WHen I do male clean I get '../common/ugcc_top.mak:68 @FAPIUFST_MAK@: No such file or directory08:47.51 
  THis is Linux08:47.57 
  I think I did a git pull -- rebase so I did a /08:48.21 
  ./confgure first. GS builds OK08:48.27 
chrisl You need to run autogen.sh08:48.37 
kens ah, OK thanks08:48.42 
kens can't keep configure and autogen straight08:49.04 
chrisl In both ghostpdl (for pcl and xps) and ghostpdl/gs (for gs)08:49.08 
kens OK make clean works now thanks08:49.47 
chrisl kens: the best thing to do, I find, is (assuming you have no important non-tracked files), after you do git pull, then do "git clean -x -f -d" and that will remove *everything* that didn't come from git. Then there's only once script left that you can run.08:50.52 
kens :-)08:51.08 
  I'll try and bear that in mind08:51.14 
chrisl Or, in the near future: make maintainer-clean will also work08:51.39 
kens Some new target you're working on ?08:51.57 
chrisl The target is already there in gs, but it's incomplete, and I'm adding it to ghostpdl for consistency08:52.37 
kens makes sense08:52.49 
chrisl Basically, "make clean" removes the files created by "make", "make distclean" gets you back to what's in a distributed archive, and "make maintainer-clean" gets you back to what comes out of git.08:54.00 
kens OK sounds useful08:54.16 
chrisl Like I say, all there, but bit rotted08:54.36 
  kens: I've got a public domain implementation of sprintf() integrated here (no localization!). Unsurprisingly it causes quite a few diffs in pdfwrite......08:57.20 
kens Oh great....08:57.35 
  How bad is it ?08:57.58 
chrisl So far, all pixel level differences08:58.07 
kens OKso probably precision then08:58.35 
chrisl Yeh, the floating point rounding is very slightly different, which isn't surprising08:59.05 
kens No, that's not at all surprising, unfortunate, but....08:59.22 
chrisl I may look into whether upping the precision would help, but I suspect not - it's a quite difference implementation to the glibc/bsd libc versions09:00.41 
kens Well we probably get slightly different results on Windows as well, so if its just pixels I doubt its worth worrying about09:01.28 
chrisl I actually grabbed it from SQLite: it says it's a public domain version from the '80's which they need because (shock horror!) they need a version without localization!09:02.56 
kens :-)09:03.05 
chrisl Given that, I find it even more astonishing that glibc doesn't provide a simple way to ignore the locale for a given call into the lib :-(09:04.08 
kens Yes, looks like its a problem a lot of people have struggled with.09:04.33 
  Hmm, I cannot reproduce the cluster results on my Linux system09:04.49 
  I suppose that means trying it on a node09:05.42 
chrisl Or run it under valgrind first?09:06.23 
kens Hmm, no this may be a GSproblem.09:06.35 
  I was looking at the resulting PDF file with Acrobat/MuPDF and they show the correct result, GS complains about an error09:07.02 
  What's the display device in Linux ?09:07.33 
chrisl x1109:07.41 
kens aha, thanks09:07.45 
chrisl or x11alpha09:07.55 
kens don't work, I guess i DON'T HAVE X INSTALLED09:08.07 
chrisl Probably don't have the x libs.09:08.23 
kens Its OK, it gives tghe same result under Windows09:08.37 
  So I can at least debug it now09:08.44 
chrisl What's the error?09:08.49 
kens Just 'Error reading a content stream'09:09.02 
  ie 'something is wrong'09:09.08 
  Its OK, this is new output from PXL+pdfwrite, so I'm happy to see an error09:09.30 
chrisl Ah, well, that's not so bad09:09.59 
kens and the GS output dfoes look like the cluster output, so I'msure that's the problem. I shouldn't have trusted Acrobat....09:10.01 
  Right, its a limitcheck in setdash09:11.01 
  Which is the problemI was wrestling with yesterday09:11.15 
  Looks like GS has an artifical limit on setdash parameters09:11.34 
  I guess I need to read the POOLRM some more09:11.45 
  PLRM09:11.49 
  Ah, 11 elements in a dash is listed as a 'memory limit' in PostScript. In level 1....09:16.21 
chrisl I've got to go out for a while....09:19.52 
tor8 chrisl: the magic invocation to printf & co is %.9g to get all 32-bits losslessly formatted for round-trips. maybe that could help test the public domain version of sprintf you got with the glibc one09:32.35 
Robin_Watts tor8: Did you see my comments yesterday?09:41.18 
  On robin/master there are various reviews.09:41.38 
  http://git.ghostscript.com/?p=user/robin/mupdf.git;a=commitdiff;h=cca5a8508f628035c41c969edfffdd9c9a3b574809:41.48 
  You may have looked at an earlier version of that one before.09:42.11 
tor8 ah yes, the don't-copy-verts patch?09:42.37 
Robin_Watts yeah.09:43.09 
tor8 you also said there was no measurable benefit, in which case I'd prefer the simpler to understand code09:43.27 
Robin_Watts So I then tried a version which further reduced the copying which had no measurable benefit... ok.09:43.52 
tor8 my hunch would be that the copying is on the stack in hot cache09:45.32 
Robin_Watts I do wonder about switching to fixed point for this stuff.09:45.56 
tor8 fixed point will be slower on non-arm today, and possibly slower on arm tomorrow as well09:46.47 
Robin_Watts All the patch decomposition could be done in fixed point, and it won't be slower on anything!09:46.59 
tor8 the triangle rasterization is all done in fixed point already?09:47.25 
Robin_Watts The drawing routines end up wanting results in fp (we convert everything by multiplying by 65536 to put stuff into the gel.09:47.42 
  Looks like x and the colors all end up fixed point, but y stays fp.09:57.07 
  s/fp/floating/09:57.15 
tor8 y is integerizet10:03.42 
Robin_Watts tor8: is it?10:03.53 
tor8 y in the gel is an int (which subscanline)10:04.15 
Robin_Watts I'm looking in draw/draw_mesh.c in fz_paint_triangle.10:04.55 
tor8 sorry, which scanline, not subscanline10:05.11 
Robin_Watts oh, I see, sorry.10:05.14 
  so my contention is that we could convert the values to int and/or fixedpoint earlier.10:05.52 
  and do the recursive splitting on the fixed point versions.10:06.29 
tor8 we could fairly easily clip in fixed point, but then we'd need to have y as fixed point as well10:06.34 
Robin_Watts tor8: going from fixedpoint to int is a cheap operation.10:07.01 
tor8 yeah, just pointing out where it needs to be done lest we forget10:07.19 
Robin_Watts So, there are 2 commits ready to go on robin/master then.10:08.29 
Robin_Watts runs for a bit.10:09.37 
tor8 Robin_Watts: those look good10:09.47 
Robin_Watts Ta.10:11.50 
  tor8: I'm going to abandon this fixed pointing for now.13:57.37 
  The tensor patch stuff works OK, but the other mesh types, like the radial ones use huge values that will clip badly.13:59.13 
henrys kens:just a few things:update the makefile with new includes, the comment explaining the code < 0 return can be omitted - I really hate changing the parsing code but I guess we're stuck with that as things stand. BTW great job getting the knack of PXL code so quickly.14:15.33 
kens henrys, no problem, I will do thosae things. Can't say I understand the parser though :-)14:16.03 
  I oculd have changed the underlying code instead of the parser, but it would have meant *much* more extensive changes14:16.42 
  For XPS I did it wihtout touching the parser, buit it only needed changes in 2 places.14:16.59 
henrys kens:so pcl is next?14:23.58 
kens Yes, that's the plan. DId you get a chance to look at the diffs ?14:23.59 
henrys I've only looked at XL diffs.14:24.05 
  kens:did you send others?14:24.29 
kens It was the XL ones I meant, I'm a bit concerned about the missing lines in C424.bin and the other test suite files like it14:24.57 
  The rendering of the 'G' looks to be resolution dependent (its ok at 75 dpi) andthe 'grid effect' I'm not too concerned about, but missing content worries me14:25.48 
sojic gs -sDEVICE=pdfwrite -o page_003-t.pdf -c "[/CropBox [0.00 0.00 623.62 807.87] /PAGES pdfmark" -f page_003.pdf -c quit // causing rotate pdf.14:27.03 
henrys kens:I assume the missing lines were resolution dependencies 14:27.15 
kens henrys thsoe are missing at both 75 and 600 dpi14:27.30 
sojic http://butterfly.venikom.com/pdf/sourcePdf/VOGUE_/123/distribute/3/page_003.pdf - original14:27.32 
  http://butterfly.venikom.com/pdf/sourcePdf/VOGUE_/123/distribute/3/page_003-t.pdf - rotated14:27.36 
kens sojic -dAutoRotatePages=false14:27.45 
sojic Thank. I'll try right now14:27.56 
kens come back if that doesn't work14:27.59 
henrys kens:arg I guess I should simplify it for you.14:28.13 
  kens:you are sure pdfwrite is being run at 600 on the cluster?14:28.54 
kens henrys, according to the logs marcos runs pdfwrite at 75 and 600 then runs GS at 75 and 600 on the resulting PDF files14:29.18 
  Many other files do show resolution dependent differences14:29.38 
henrys kens:well I'll break down the file after the meetings today but I wouldn't hold up a patch for that.14:30.03 
sojic kens : gs -sDEVICE=pdfwrite -o page_003-t.pdf -c "[/CropBox [0.00 0.00 807.87 623.62] /PAGES pdfmark" -dAutoRotatePages=false -f page_003.pdf -c quit // Same problem14:30.04 
kens What particularly puzzles me about the missing lines is that the other lines on the same page don't have a problem, and I assumed they were constructed the same way14:30.20 
henrys kens:have you looked at both resolutions non pdfwrite?14:30.55 
sojic Fixed14:31.01 
kens sojic put the -dAutoRotatePages *before* the pdfmark (starting at -c) , order matters14:31.03 
sojic Yes14:31.06 
Robin_Watts ooh, factor of 4 speedup! I win!14:31.08 
sojic Figured out14:31.11 
  Thanks14:31.13 
kens NP14:31.15 
  henrys I did check pcl6.exe at 600, the lines are present. I thought I tried 75, I'll do it agian to be sure14:31.59 
Robin_Watts tor8: ping14:32.50 
kens henrys, yes the lines are there at 75 dpi as well14:33.16 
  THe 'G' fill is not clipped at 75 dpi though, so that is obviously resolution dependent (one reason why I was less worried about it)14:34.00 
henrys hmm maybe touched vs centered rendering but I'll break it down and we'll see for sure.14:35.14 
Robin_Watts tor8: 2 reviews on robin/master for you.14:35.40 
kens OK thanks, like I said, the fact that the other lines exhibit (and the lines were present in the old pattern handling) is a slight worry14:35.41 
kens wonders if this missing line might be a ROP-related thing14:42.07 
henrys kens:now that I'm thinking about this high level patterns should break a number of things in PXL with pdfwrite - any non trivial raster op between source and pattern would have been previously handled when rendering to the pattern cache.14:42.22 
  yes exactly what I was just writing...14:42.31 
kens :-)14:42.42 
  Mostly the files seem to be OK14:42.53 
tor8 Robin_Watts: okay (re fixed pointing gradients)14:44.07 
  Robin_Watts: 2 patches LGTM14:44.16 
Robin_Watts tor8: We *could* change the code to use floating point for x/y and ints for colors.14:47.27 
  but I can't be bothered at the moment.14:47.47 
henrys kens, Robin_Watts: what is the upper limit on compares for bmpcmp, are we looking at everything? I am expecting to see a particular rop test from the customer_files directory but I don't see any files from there.14:49.49 
kens 100014:49.57 
Robin_Watts 1000 jobs.14:49.57 
kens do you know hte name henrys ? THre were not that many files that showed diffs14:50.30 
henrys Feature request a list of files as html links could be the first page of bmpcmp output.14:50.35 
Robin_Watts If kens is generating more diffs than that, the best approach is to retest with -filter set to get you a subset (say all the ppmraw 600dpi ones).14:50.44 
  Then bmpcmp that subset which will hopefully be less than 1000.14:50.54 
kens There were not 1000 diffs14:50.55 
  197 jobs showed differences14:51.26 
Robin_Watts henrys: You want an easy way to find what page a given diff is on?14:51.45 
henrys yes but you have at least 5 diffs per job right?14:51.54 
Robin_Watts I'd rather put the list at the end of the bmpcmp output, but it's doable, certainly.14:52.07 
kens AIUI its the number of jobs, not the number of diffs14:52.31 
Robin_Watts It is the number of jobs, yes.14:52.46 
kens Sp lots of diffs per job just gets you lots of pages of output14:53.08 
henrys Robin_Watts: I meant the first thing we should see on the web is a list links that say like 150: tests_private/xl/pcl6cet/c427.bin.pdf.ppmraw.600.0 (600dpi) and we can click on that and look at that output.14:53.36 
kens NB each resolution or 'device' is a job14:53.37 
Robin_Watts henrys: Right, and I'm saying I'd rather have that list on the last page.14:54.29 
henrys kens:I was hoping to see all_rops.pxl 14:54.30 
kens one moment henrys14:54.38 
henrys Robin_Watts: that's fine for me too.14:54.54 
kens THat file doesn't come up in the list of diffs14:55.08 
  Either its not different, or (possibly) Marcos doesn't test it14:55.30 
  I guess maybe its not tested, because it soudns like it wouldn't work with pdfwrite anyway ;-)14:56.51 
henrys kens:well it would exhibit the type of fallout I'm expecting from the pattern cache not handling the rop.14:57.37 
kens Yes, I understand14:57.48 
henrys anyway we are 3 minutes of a meeting and I'm completely unready14:57.57 
kens coffeee quick! :-)14:58.19 
Robin_Watts is boiling the kettle for tea already :)14:58.44 
henrys kens:we'll ask marcos what the coverage is I don't feel like log excavating.15:00.19 
kens OK no problem henrys, I'll sort out the other thigns now and put it to one side15:00.44 
henrys paulgardiner: I can't wait to have signature at the show - I'll demo our support contract ;-)15:01.15 
paulgardiner :-) How many do you think you can catch?15:02.27 
mvrhel_laptop sorry I am late15:03.22 
Robin_Watts henrys: which show?15:03.53 
henrys chicago15:04.10 
Robin_Watts ah, ok.15:04.26 
paulgardiner henrys: this is assuming I can figure out how to use certs from the OS within openssl, which isn't so far going well. :-(15:04.50 
henrys paulgardiner: you know the pdfcreator guy might have researched this already and we could pay him something if he had something interesting.15:06.09 
  http://www.pdfforge.org15:07.02 
  you can also look at the code but be careful not to infringe.15:08.29 
paulgardiner Ok, that may well help.15:08.46 
henrys Robin_Watts: good to hear from a mupdf customer.15:09.32 
paulgardiner The verify call I'm using takes as an argument a store of root certificates. Possibly I need to enumerate the OS store and place them in the openssl store.15:09.42 
Robin_Watts henrys: Christophe?15:09.52 
henrys right15:10.10 
Robin_Watts That company talks to me a fair amount, normally text messaging via skype.15:10.17 
  I'm trying to steer them towards copying support using bugzilla.15:10.32 
henrys so it looks like paulgardiner will be in signature hell for a bit longer - anything to discuss schedule wise?15:11.32 
Robin_Watts I've been distracted away from image extraction. I hope to get back to that soon.15:12.38 
paulgardiner Difficult to know at the moment. There's several things that could turn out to be awkward.15:12.44 
mvrhel_laptop henrys: I am going to be slowing down a bit on the windows stuff this week due to ghostscript work, but I am going to pull out all my mupdf calls and wrap them up so that they can be called from the other languages so that is it a true winRT set up15:12.56 
  the viewer has a couple glitches but I have links, contents, search all working15:13.25 
Robin_Watts The same customer is interested in text reordering etc, so that might cause us to revisit/rework some of the existing 'done' stuff on the text extraction roadmap.15:13.40 
henrys mvrhel_laptop: okay sounds good. Have you posted all the viewer code for feedback?15:13.50 
  mvrhel_laptop: I wish we could keep you going on this but the gs project is uniquely yours.15:14.32 
Robin_Watts mvrhel_laptop: Did you look at reflow? ISTR you saying that you were going to look at that.15:14.36 
mvrhel_laptop Robin_Watts: yes. just typing about reflow15:14.48 
Robin_Watts marcosw: Has the cluster jammed?15:14.58 
marcosw Robin_Watts: let me check15:15.12 
mvrhel_laptop I have it working, but the html webview object blows in windows 8 ui and intercepts all the manipulations making zooming difficult15:15.28 
  right now, I have it display at a rich text output and you can scale that. I need to look a bit more at what my options are15:16.10 
Robin_Watts mvrhel_laptop: We could use onscreen icons for zoom +/-, but it's a shame that the UI is different.15:16.35 
mvrhel_laptop Robin_Watts: yes15:16.51 
henrys tor8:ping15:16.57 
mvrhel_laptop anyway, my big thing will be to pull out the mupdf stuff and add in a couple simple sample apps in other languages15:17.30 
  that is pull it out into a dll that can be shared amongst the languages15:17.52 
Robin_Watts mvrhel_laptop: Presumably, someone has already done that for the thing that David used to write the article ?15:17.58 
tor8 henrys: yes?15:18.27 
henrys how is localization?15:18.53 
tor8 mvrhel_laptop: did you see the link that robin pasted the other day, the blog post on msdn?15:18.55 
  henrys: I've got the string resources in the android app all cleaned up. just need to set up the oneskyapp account and figure out how that works.15:19.21 
  do we have a specific list of languages we want to support?15:19.43 
Robin_Watts All of them!15:19.58 
tor8 that'll put a dent in my wallet...15:20.08 
Robin_Watts http://email.oneskyapp.com/wf/click?upn=R-2B1hm6oTz-2FyAhd1Dtw8ZZkhELWqcMtWHUIwXk-2Bm2L4ITZsoNHiAV6LyrTLGQs4FU8G-2BfycZsWcTfHxktuAkU7cTt7Q2NsEAxiIbovh2ejUQxaB9MJKRlzQ9UJMqPXHP9vBKZISNcLOUR63hoXutGr8omhEEfhRJOW66xukANGLLeIheU3JHB9nXFyOM0W9ezjKwefYDLmnkSvY6eD3igPTJzyX8T1QLmit5rzrFpUpVMSFEYaP2FQgRFcWsJd879YEuszRmkh8KZkI9PSmDUlg-3D-3D_6xWBv-2B8To-2Bx7VnI0-2BJY-2FHF6hzuUwn6Y34xjriiq51ybcJeyOew-2B8Qe5qY15:20.10 
  6XCkU8tOy5o7ch6MIUh8643UCPeoVjpnqeju1VD82htsxBVzByWXPcg-2F4SKUYmQi6f4XfjhMeko40HHypxKnlFb0ga5cuAoP0cx-2FGsceSynE-2FnwHfzFnYGq0AW93DNgzfloxbvzub-2FBn4rrtNdMginEekPcbQ-3D-3D15:20.12 
henrys we might as well support all15:20.18 
Robin_Watts It's like $400, right ? I have a US $ card if that helps.15:20.45 
henrys Robin_Watts: overflows my irc clients line buffer15:20.45 
mvrhel_laptop tor8: yes. I was looking at the code/project. they guy did a nice job on it15:21.07 
tor8 the dollar is pretty worthless compared to the swedish currency, and I don't have a visa attached to my USD account15:21.14 
Robin_Watts http://tinyurl.com/cjlzfxq15:21.25 
marcosw Robin_Watts: I'm not sure what the cluster is doing. I could just restart it but would rather understand what's going on.15:21.43 
Robin_Watts mvrhel_laptop: Is it worth us just paying him $$$ and using that?15:21.49 
  Cost of mvrhel vs cost of paying out etc.15:22.03 
mvrhel_laptop Robin_Watts: well, we could do that. 15:22.19 
  what do you think henrys?15:22.48 
Robin_Watts mvrhel_laptop: I can imagine there might be arguments both ways (he might not have done it in quite a general enough way etc).15:23.02 
mvrhel_laptop Robin_Watts: there will still be work to do. He does not have a viewer here, just wrapped up the calls needed into mupdf so render pages15:23.50 
  s/so/to/15:23.56 
henrys mvrhel_laptop: I am fine with it but you should look at it carefully, if you know over time your are going to do it differently it doesn't really make sense. CLA of course.15:24.08 
mvrhel_laptop henrys: yes, I think it will not be too hard for me to take the few calls that I have into mupdf, wrap them up into a new class that can then be shared amongst the different languages15:24.47 
Robin_Watts mvrhel_laptop: Right. Were you planning to just expose the calls you need for your viewer? Or all the calls?15:24.55 
mvrhel_laptop to me this makes more sense, since these are the things that I know we need for a viewer15:25.06 
  just the viewer ones for a demonstration15:25.19 
Robin_Watts mvrhel_laptop: Do you handle forms etc?15:25.29 
mvrhel_laptop Robin_Watts: I have no special stuff for form handling15:25.54 
Robin_Watts ok.15:26.08 
henrys mvrhel_laptop: also once you have something workable in git you can point him to it and ask if he's interested then post bountiable enhancements, right?15:26.11 
mvrhel_laptop I have reflow, text search, links, content display15:26.14 
  henrys: true15:26.30 
  let me at least make an effort at it as I will feel like I wasted a bunch of time otherwise15:26.54 
henrys mvrhel_laptop: seems like you'd have better control over direction that way.15:26.57 
  anything else we should discuss here? 3 minutes to go.15:27.56 
  tier 3 is 208.58 right?15:28.29 
Robin_Watts it is.15:29.09 
  It's $467.72 for all 3 tiers for the app description.15:29.29 
  Plus some other as yet unknown sum for the resource strings.15:29.47 
henrys oh I see I though lower tiers were subsets.15:30.07 
  s/though/thought15:30.14 
  peculiar how they are split up whey would swedish be a different tier than norway?15:31.04 
Robin_Watts henrys: number of people that speak that language I guess.15:31.29 
henrys then chinese would be tier 115:31.50 
tor8 henrys: twice as many swedes as norwegians15:32.02 
Robin_Watts ...and have smartphones.15:32.02 
  4.7M swedes with smartphones. 2M Norweigies.15:32.26 
  Hindi has 38Million, but then they all speak english too anyway.15:33.13 
tor8 odd that they have spanish (spain) and latin american spanish as different... I'd have guessed the difference there was like between US and UK english15:33.48 
Robin_Watts tor8: It's different when spoken (very different accents)15:34.17 
  but yes, it's all a bit arbitrary I suspect.15:34.37 
henrys anyway let's just finish that - it really could matter sales wise and it takes little effort.15:34.45 
Robin_Watts We can cherry pick individual languages.15:34.51 
tor8 I'll set up the account and prep the files then maybe robin can whip out his USD visa?15:35.08 
Robin_Watts tor8: Sure.15:35.15 
tor8 and check the price diffs between the tiers to see if it's actually worth spending or not for the flavours of same langugae15:35.43 
  Robin_Watts: tor/android has one commit that I see hasn't been merged yet15:36.26 
Robin_Watts latin american spanish = $1615:36.36 
marcosw Robin_Watts: figured out in what state the cluster is stuck and will add code to automatically recover from this condition the next time it happens. the cluster would have recovered 2 hours after your job started, but the new check will cut that to five minutes. however, I'm still not sure how it got into this state. 3 of the nodes never received the command to start the run but they cluster master is convinced it sent the comma15:36.45 
  I could believe if one node had this happen due to some internet glitch, but it's odd that 3 nodes are in this condition.15:36.46 
Robin_Watts tor8: Looking now.15:37.25 
  Looks fine. Will push. 1 on robin/master for you.15:39.19 
henrys wow 5k running race today in a foot of snow, races don't get canceled here for anything.15:39.43 
Robin_Watts henrys: Pays to not be leading that one I guess.15:40.24 
  marcosw: odd.15:40.57 
mvrhel_laptop henrys: so I am going to be out for a good part of the day. I have to chaperone a field trip at my daughters class 15:41.02 
henrys mvrhel_laptop: okay15:41.23 
  where you off to?15:41.33 
Robin_Watts The cardboard box factory again?15:41.45 
tor8 Robin_Watts: yours also looks good15:41.58 
Robin_Watts tor8: Thanks. It caused diffs in the last test. trying to figure out why :(15:42.24 
henrys I remember chaperoning to the butterfly pavilion and I kept stepping on the damn things and all the kids would yell at me.15:42.48 
Robin_Watts henrys: And 1) you were never asked back, and 2) it's all the kids remember about that day, right?15:43.19 
paulgardiner so a thoroughly good result really.15:44.09 
henrys yes I volunteered for all sorts of stuff and in hindsight … well I don't know what I'd do.15:44.24 
tor8 henrys, Robin_Watts selecting absolutely everything on the UI strings resource for android runs up at $55215:44.41 
  and that's just 52 strings15:44.48 
Robin_Watts tor8: That includes tiers 1-3 ? or is that in addition?15:45.05 
tor8 Robin_Watts: there's a different ui for picking once logged in and doing a proper "project" in oneskyapp15:45.54 
mvrhel_laptop henrys: hehe. we are off to a boat15:46.23 
tor8 picking the "the european biggies" and "the roaring east asians" tiers is $132 and gets us basically tier 1 and 215:46.24 
mvrhel_laptop http://www.argosycruises.com/15:46.24 
  I pick only the fun field trips to help out on15:46.48 
marcosw bbitft9m (be back in time for the 9:00 meeting)15:47.22 
Robin_Watts mvrhel_laptop: kids on a boat. nausea, vomiting and drowning. Is it too late to back out?15:47.29 
mvrhel_laptop ha. I had not thought of that15:47.39 
henrys mvrhel_laptop: so the mission is to keep them out of the water.15:48.18 
ray_laptop and stay clear when they throw up their fruit loops and milk15:48.56 
Robin_Watts mvrhel_laptop: So, on the bus on the way there, pick the roudiest ones, and appoint them "Vomit Monitors" as a punishment.15:49.38 
mvrhel_laptop good idea15:50.00 
ray_laptop my wife said there was just an earthquake, but I didn't feel anything 15:52.41 
  I'm only .25 mi away. Must have been pretty small15:53.10 
  nothing shows on the usgs map. Probably some construction local to our house15:57.43 
henrys mvrhel_laptop: last meeting I was going to ask about the NRE project but I think that's on the way and 689057 which was felt worthy of fixing at the meeting.15:58.08 
Robin_Watts ray_laptop: will get home to find a tree embedded in his roof.15:58.10 
chrisl Robin_Watts: or a roof embedded in his tree........15:58.49 
Robin_Watts There's no place like home. There's no place like home...15:59.10 
mvrhel_laptop henrys: yes. I made some good progress on this project yesterday15:59.31 
chrisl thinks Robin_Watts would not suit ruby slippers15:59.48 
mvrhel_laptop hoping I can get it clean up middle of the week and then pass off to ray_laptop for him to finish up15:59.54 
henrys marcosw:kens and I wanted to verify you run all the tests with pdfwrite - he made a change that should have triggered failures but didn't.16:00.33 
ray_laptop mvrhel_laptop: sounds fine. I should have the #10 issue soon as well.16:01.03 
henrys actually we don't know for sure it should have16:01.04 
marcosw henrys: yes, all the usual test files are run with both pdfwrite and ps2write.16:01.18 
  I can check the logs, was it a clusterpush or commit regression run?16:01.48 
henrys marcosw:a local push by kens associated with ~regression/ken bmpcmp's16:02.42 
kens My guess is it was wrong before and worngnow.16:03.16 
  I will get the file and look16:03.22 
henrys I find it odd thought that not a single file from the customer_tests directory shows up in your bmpcmp16:04.57 
  kens:given the errors I do see.16:05.22 
  alexcher:anything for the meeting?16:05.49 
kens Umm, possibly, I have the file now, give me a moment16:05.52 
alexcher marcosw: did you do any new testing for the fuzzing bugs?16:05.53 
marcosw henrys: ? The first file in <http://www.ghostscript.com/~regression/ken/> is "tests_private/customer_tests/A1TEST_CHART.hp2.pdf.ppmraw.600.0 "16:06.21 
  alexcher: no, I just moved the issues from the two bugs to the new, individual bugs.16:06.49 
henrys marcosw:ah okay I am not sure how I did that started on page 1.16:07.15 
alexcher marcosw: OK, I'll close the bugs that have been fixed already.16:08.29 
kens henrys looks to me like none of the reops are applied, sothe result is always incorrect16:08.37 
  sonow its the same incorrect16:08.44 
  Need to do a rebuild to check though16:08.58 
marcosw alexcher: thanks. I should have done that but haven't had time.16:09.07 
henrys kens:okay so I should apply your patch and find the missing lines part in isolation correct?16:09.45 
kens henrys I can apply the patch iof you like, the missing lines thing could just be rops, but I would like to see for sure so it would be nice to check it out before comitting16:10.24 
henrys chrisl, ray_laptop : anything for the meeting?16:10.27 
marcosw chrisl: The customer who reported a problem with ufst some weeks ago but didn't give a version has come back with UFST 6.2 and GhostPCL 9.05. However I can't get that combination to compile. Okay if I assign the bug to you to have a look at? 16:10.32 
henrys kens:that's fine I'll do it just after the meeting.16:10.55 
Robin_Watts marcosw: 6.2? I thought they said 6.3 ?16:11.07 
kens henrys did a rebuild, checked old PDF vs new PDF< thery are identical. Wrong, but identically wrong16:11.10 
henrys kens:apply the patch from your email that is.16:11.12 
chrisl marcosw: as far as I'm aware, we only support PCL withe UFST that we supply, and we've only ever supplied 5.x16:11.16 
marcosw kens: right, 6.316:11.25 
kens me ? 6.3 ? confused...16:11.49 
Robin_Watts kens: marcosw meant me :)16:11.59 
henrys hopefully ray is still on terra firma heard they had a 7.8 in iran.16:12.14 
marcosw kens: I meant Robin_Watts, all you brits look alike16:12.16 
kens :-)16:12.21 
mvrhel_laptop hehe16:12.27 
chrisl henrys: I did want to mention that I'll be committing the tiff configure changes soon, so everyone that uses the configure build will need to "make maintainer-clean" and autogen.sh16:12.37 
Robin_Watts It's our matching bowler hats, umbrellas and moustaches.16:12.43 
ray_laptop henrys: nothing for the meeting from me (I already mentioned cust #801's enhancement is in progress)16:12.43 
marcosw pasty with bad teeth :-)16:12.45 
kens henrys I didn't send you quite all the patches as diffs, there's a change also needed in pdfwrite, I was happy to review that one myself though :-)16:12.54 
  Also I need to do the makefile changes and remove the comment16:13.22 
  So I'm happy to commit it tomorrow instead.16:13.37 
  Unless you would prefer to do so.16:13.44 
henrys kens:at this rate I'd say just commit it and we'll have an open bug.16:13.46 
marcosw could be worse, at least you aren't fat germans in lederhosen.16:13.50 
Robin_Watts was in the middle of preparing a german bashing joke :)16:14.13 
kens henrys I do need to do those changes anyway, so I'm happy to wait and look at a reduced problem tomorrow16:14.18 
marcosw henrys: what is our policy regarding ufst 6.3? do I tell the customer it's not supported by us? We have discussed supporting ufst 6.x16:15.16 
henrys kens:I am concerned I'll be able to find the problem without the completed patch set. Do I have everything I need to get the missing lines?16:15.19 
kens henrys I htink so yes, the change in pdfwrite prevents a seg fault in C318.bin, so don't run that one16:15.51 
henrys kens:okay16:16.00 
kens I believe all the other files run to completion with what you have16:16.13 
Robin_Watts henrys: kens changes are shown at the end of the cluster test report, aren't they?16:16.31 
kens yes16:16.37 
chrisl marcosw: as we don't have UFST 6.3, and UFST needs our custom patches to work, I would double check they don't actually mean 5.3.....16:16.46 
kens but because I wasn't using a totaslly up to date source, there are some other there too Robin_Watts16:16.59 
marcosw chrisl: will do.16:17.08 
kens ones that chris had pushed16:17.09 
henrys marcosw:what chrisl said, oh this is the folks mixing and matching fcos?16:17.26 
kens henrys looking at the all_rops file, the pdfwrite output is just totally wrong, but at least its consistent16:17.37 
marcosw henrys: yes.16:17.53 
chrisl Why would they not use the FCOs supplied?16:18.16 
henrys marcosw:so I think the first answer should be we don't know anything about compatibility beyond the stock fco's that come with a particular version and the first question should be how did they think to do something so daft?16:20.16 
chrisl It makes me very wary since the FCOs are shipped with the UFST releases16:21.04 
henrys anything else meeting wise I can't think of anything but I didn't try very hard this week ;-)16:24.33 
alexcher marcosw: Sometimes debug build of gs generates warnings about GC and memory problems. How difficult is to screen the whole test suite for such warnings?16:25.38 
marcosw alexcher: I run the debug build as part of the weekly regression tests (on Tuesday, so it's running as we speak). I should be able to just write a simple script to check the log files. 16:26.52 
  should I be using a particular -Z option to see the issues?16:27.18 
  we run with -Z@ currently.16:28.03 
Robin_Watts I had a random thought earlier today, when I wanted to find which test files used a type 6 shading - I put exit(1) in the code to make it die and looked for the ones that died in my clusterpush.16:28.12 
alexcher marcosw: I have gases when all Valgrind warnings are fixed but gs warnings persist. I didn't use any -Z options.16:28.26 
chrisl alexcher: I think a bunch of those that relate to strings are when we try to gc scan string constants - so we get a "not in chunk" warning.16:28.43 
Robin_Watts I was wondering about maybe having the cluster automatically grep the logs etc for lines starting with a given string, and collating them into a log file reachable from the dashboard.16:29.04 
chrisl Robin_Watts: that would be useful - I ended up adding code that would segfault for that purpose16:29.36 
Robin_Watts So people could printf("CLUSTER: whatever\n"); and cluster test and automatically get a list of what files emitted that.16:29.49 
alexcher chrisl: such strings should be in the foreigh memory, souldn't they?16:30.03 
chrisl alexcher: the problem is we don't know that because "string allocations" (*not* string refs!) don't have memory management headers - we reach them through enum_ptrs calls.16:31.21 
henrys I've always wanted to try cross referencing code coverage lines with test files. you'd have some sort of html output of the code click on a line and see which files exercise that line.16:32.21 
chrisl alexcher: I'm not suggesting that's all the problems, though, we probably do need to investigate the warnings16:32.32 
alexcher chrisl marcosw: but little is done without a tracking number.16:34.05 
chrisl alexcher: yes, I am in agreement with you - a list of files, and a bug would be a good thing to have16:34.58 
  alexcher: I've only ever looked into any of them in-passing when working on another problem16:35.34 
marcosw Robin_Watts: I can do that pretty easily, can you add some output and do a clusterpush so that I can test the code when I add it later today?16:36.05 
Robin_Watts marcosw: I will.16:36.22 
marcosw henrys: I'm surprised such a thing doesn't already exists. I should be able to hack something together...16:38.35 
henrys marcosw:I don't think it can be just brute force - you are going to need some way to keep the number of files under control.16:39.19 
chrisl marcosw: is the clusterpush.pl in ghostpdl/gs/toolbin/localcluster the "golden" version, or does it reside elsewhere, too?16:40.37 
marcosw chrisl: that's the "golden" version.16:41.07 
chrisl Cool, thanks16:41.16 
Robin_Watts henrys: The number of lines more than the number of files, I suspect.16:41.17 
henrys well if you click on one line of the code is that going to show a list of 10,000 files, then the next line the same 10,000 files on another page?16:42.19 
  presumably the same files.16:42.41 
Robin_Watts henrys: Right. You'd want to identify lines in the source that were called 'together'.16:43.43 
marcosw henrys: I'm guessing lines that are covered by 10,000 files aren't interesting; we could just report the first 10 matches and a ...16:43.44 
henrys marcosw:well I also wanted to say give me all the test that use this line.16:44.45 
marcosw Robin_Watts: yeah, I was going to do that. Generate a md5sum of the files that affect each line and treat groups of lines that have the same md5sum as a set.16:44.47 
henrys Robin_Watts:exactly.16:45.09 
Robin_Watts marcosw: What tool would you use to generate the coverage ?16:45.13 
marcosw henrys: even if there are 10,000 of them? 16:45.19 
  Robin_Watts: gcov. Or is there something more better?16:45.40 
Robin_Watts marcosw: No, I just wasn't aware of gcov :)16:45.54 
  You've just halved the problem size :)16:46.12 
marcosw I think it's gcov (it's some gcc related coverage tool).16:46.28 
henrys marcosw:If you go to the trouble of implementing "together" I don't think it would be a problem to report all files.16:47.48 
marcosw the biggest problem I foresee in this coverage tools is that every time a source file is edited the results are invalidated. And as I recall coverage testing isn't fast, so regenerating the results every commit isn't possible.16:48.01 
mvrhel_laptop bbiaw . 16:49.02 
henrys marcosw:I think that is okay for the use cases we are imagining. once a week it runs and the database is available online somewhere.16:50.07 
  we discussed gcov cluster stuff before that would certainly help, merging gcov is suppose to be easy ...16:50.56 
marcosw henrys: I will probably have to involve the cluster machines, as I recall coverage analysis on a single machine takes many days.16:51.48 
kens OK have to go make dinner goodnight all16:51.56 
marcosw Robin_Watts: is there a way to have the memento memory squeeze feature to stop at some point? I.e. specify a starting and an ending allocation count? 16:55.26 
Robin_Watts marcosw: Not currently.16:55.41 
henrys marcosw:you also have to identify all the parameters with the test file which makes it quite nasty. pbmraw uses different lines than ppmraw etc.16:55.47 
Robin_Watts henrys: You just collect coverage data based on the filename_options not just the filename.16:57.06 
marcosw henrys: what Robin_Watts said16:57.21 
Robin_Watts i.e. tests_private__pdf__foo__bar__baz__200_0_ppmraw or whatever it is.16:57.39 
  marcosw: Is there a particular reason you want that ?16:58.01 
  The initial allocations are probably the same for many files (the postscript setup is identical for any given device call).16:59.21 
  The latter ones are where it's interesting.16:59.28 
henrys yes I understand that, I was just thinking of the ballooning number of files.16:59.37 
marcosw Robin_Watts: makes it easier to run memory squeezing on idle cluster nodes. Right now even a simple job such as tiger.eps runs for many hours and other than killing the job there isn't any way to stop it.17:00.23 
Robin_Watts marcosw: Can you not suspend processes under unix ?17:01.25 
  kill -STOP pid then kill -CONT pid ?17:02.06 
Robin_Watts wonders if the pid is a moving target with all the forking...)17:02.40 
  or... nice the process ?17:03.04 
henrys marcosw:something else I forgot from the meeting: you were talking about more machines for miles' office - should we be getting heavier duty hardware and having those nodes do more. It sounded like we were getting into diminishing returns problems from communication.17:03.05 
  marcosw:you've been purchasing fairly low end stuff node wise.17:03.57 
marcosw Robin_Watts: nice doesn't really work, I run some jobs at nice level 19 and they still slow down the normal cluster jobs (which run at nice level 10).17:04.04 
Robin_Watts marcosw: OK. so -stop, -cont ?17:04.21 
  I can change memento so that the 'main' app doesn't run up the pid table if that's wrong currently.17:04.49 
  (i.e. have the one that forks be the one that dies)17:05.04 
  right. it's always the child that dies, so if you -stop the parent (the one that you initially started), that will stop the whole process (within as long as it takes the child to die).17:06.35 
marcosw Robin_Watts: I understand how to do this; I wasn't looking for advice or suggestions. I just asked if the ability already existing in memsqueezing. I can work around the lack of such a feature,.17:10.25 
Robin_Watts marcosw: right, sorry. There isn't code in memento to do that, though I could add it if required. I can't see why we'd be better off with code in Memento than using -stop/-cont though.17:11.27 
marcosw Robin_Watts: because stopping/continuing jobs isn't currently an ability of the cluster code and before I added it I wanted to check if it was needed. It's not better or worse than adding a feature to memento.17:17.31 
Robin_Watts restarting memento from a given point takes time though (we rerun through the first n allocations).17:18.06 
  and presumably if you wanted to start/stop memento using 'MEMENTO_SQUEEZEAT'/'MEMENTO_SQUEEZETO' you'd have to decide those limits before starting.17:18.54 
marcosw henrys: I've been buying 6 and 8 core AMD machines, I'm not sure I'd call these low end. They are cheap, but that's probably because they have really low end graphics and not much expansion capabilities (i.e. few PCI slows and usually only two memory slots).17:18.58 
  henrys: I did look at higher end machines (i.e. multiple 6 core xeons, but they cost thousand and I'm not sure how much faster they would be).17:19.52 
henrys marcosw:I'm not sure either I just looked here http://www.cpubenchmark.net/high_end_cpus.html and noticed my henrysx6 chip is pretty far down the list.17:22.30 
  and the other cluster machines are of similar performance17:25.45 
marcosw one of the latest nodes uses a AMD FX-8150 Eight-Core, which gets a score of 7,747, which isn't as good as the Intel Xeon E5-4650 @ 2.70GHz which get a score of ~14,000 but otoh, that chip costs $3,899.99 (odd how cpu manufacturers use $0.99 pricing, like it's an impulse purchase in the supermarket check out lane).17:26.11 
  what cpu does your henrysx6 node use?17:26.38 
henrys phenom 1055T17:27.27 
marcosw we could look into upgrading the cpu. It's a AM3 socket cpu and there aren't many faster AM3 chips, but if you mother board supports AM3+ chips there are a lot of options. 17:32.27 
Robin_Watts aws nodes that are down are now hidden on the dashboard.18:08.14 
  henrys: And bmpcmp's have an index on the last page.18:08.50 
henrys thanks Robin_Watts 18:09.17 
Robin_Watts np.18:09.33 
  marcosw: I'm fiddling with the log searching stuff now, to avoid us both doing it.18:21.02 
  marcosw: Do the .log.gz files get both stdout and stderr?18:56.49 
  oh, he's not here.18:56.59 
henrys paulgardiner: still about?19:13.04 
Robin_Watts I think he'll be gone. He has a regular tuesday night activity.19:13.52 
henrys Robin_Watts:okay no problem, if you see him before I do tell him to buy a ticket to miami - he'll be coming regularly now so I assume you or the other UK'ers will try and hook up with him for travel.19:16.39 
Robin_Watts will do. Cool.19:16.58 
  henrys: Yeah, we can share Arnie to and from the airport :)19:17.19 
  hi marcosw 19:17.48 
  Just to say, I'm playing with the log searching stuff now.19:18.00 
marcosw Robin_Watts: uh oh. What's fell over now?19:18.00 
Robin_Watts Do .log.gz files contain both stderr and stdout ?19:18.13 
marcosw I think so, let me check.19:18.39 
  it turns no, just stdout, but trivial to chagne.19:19.51 
Robin_Watts ah, ok. Where does stderr go? hyperspace?19:20.08 
  or is it in another file?19:20.14 
marcosw hold on, it's more complicated...19:20.38 
  never mind, it's not complicated. The .log.gz files contain both stdout and stderr. I thought perhaps they only included stderr and stdout for the pdfwrite/ps2write portion of those jobs, but it should be everything.19:22.46 
Robin_Watts OK, I have the searching stuff almost working. Just need to sort out a webby front end for it. will do that tomorrow.19:28.29 
  OK, new search links are active on the dashboard. Let me know what people think.23:48.08 
 Forward 1 day (to 2013/04/17)>>> 
ghostscript.com
Search: