IRC Logs

Log of #ghostscript at irc.freenode.net.

Search:
 <<<Back 1 day (to 2013/01/27)2013/01/28 
ray_laptop Bug 693590 looked sort of obvious, but something with psdcmyk cropped up in regression testing. May have been there and masked by the other bugs. Testing to see if the bug reported all the cases, or just a few00:04.05 
mvrhel_laptop henrys: oh I forgot about that01:00.19 
  I will take a look at it late tonight or tomorrow01:00.31 
kens chrisl I've got a fix for the bookmarks problem in test now.08:32.30 
chrisl kens: Cool - but no rush, given the blockers, the critical and the stuff reported over night :-(08:33.30 
kens Yes, I noticed Marcos had been busy again :-(08:33.49 
chrisl And I'm trying to decide whether I want to test the build on XP with VS2010......08:34.55 
kens I think that's a 'no'....08:35.05 
chrisl Well, I realised I have an XP machine here - my own laptop has XP on it, it just hasn't been booted into XP ~1 year08:35.56 
kens I really don't think it's worth it....08:36.12 
chrisl You're probably right08:36.29 
  And I thought I was saving overall effort by remembering the stroked font fix you did - that'll learn me!08:37.09 
kens I was imp[ressed you'd remembered it!08:37.27 
chrisl It was out of the ordinary enough to stick in my mind08:38.04 
kens chrisl do you remember if we own gsprint now ?08:47.47 
chrisl I believe we do, but I'm not 100% sure08:48.15 
kens Well its another free user batch printing 'many' PDF files, and presumably not paying us or Russel.08:50.14 
  So I don't think I have much sympathy08:50.27 
chrisl You mean many PDFs with the same instance?08:52.31 
kens Yeah I think so, using gsprint08:52.42 
  69359308:53.07 
chrisl Yeh, I just saw it - I'd have thought that risky from a print queue perspective, anyway08:53.50 
kens Well yes, thousands of print jobs and 'something goes wrong'. I mean,. this is Windows.......08:54.20 
chrisl :-)08:55.25 
kens OK fix for outlines commited (also fixes articles, named destinations and page labels)08:57.59 
  Now to look at the next one from Marcos08:59.17 
  Oh joy a pattern in an SMask09:00.38 
  Now, is Marcos using the original file or the simplified version :-(09:01.15 
chrisl I must admit, it is confusing that we (allegedly) get an infinite loop in gc_objects_clear_marks() since that's traversing memory we've already scanned once.09:05.52 
kens I don't think its true09:06.15 
chrisl Which part?09:06.33 
kens THe file takes several seconds to open in Acrobat, and to render (at 72 dpi) in Ghostscript. When I use pdfwrite it goes slowly and produces lots of warnings09:06.54 
  I think its just very slow (on page 2)09:07.02 
  Oh look, its a Cairo-produced file09:07.24 
chrisl That would make some sense if pdfwrite is, for some reason, flattening the transparency?09:07.42 
kens So we can expect insane nesting, stupid transparency and a file which is hideously over-complicated09:07.53 
chrisl Nice!09:08.07 
kens Presuambly this is why its also slow on Acrobat09:08.10 
  OK extracting page 2 gets me the same lengthy flow of 'Error reading a content stream' messages09:10.16 
  Its possible the PDF interpreter is retrying the form endlessly I suppose, though I'm doubtful.09:11.26 
chrisl A self referencing SMask group?09:12.09 
kens And the uncompressed PDF (just page 2) is > 8Mb09:12.26 
  Oh yeah, patterns which call forms which use patterns, which call forms, whic use patterns.....09:13.31 
  All using transparency of course.09:13.45 
chrisl If Acrobat takes a long time to render it, that's a *very* clear sign you've done something stoopid09:14.33 
kens Well, its a page full of coupons for McD's, here's the content stream for the page:09:15.13 
  stream09:15.13 
  q09:15.14 
  Q09:15.14 
  endstream09:15.14 
  Grrr09:15.21 
  "stream09:15.30 
  "q09:15.30 
  "/x692 Do09:15.30 
  "Q09:15.30 
  "endstream09:15.30 
chrisl Nice and simple then..... ;-)09:15.51 
kens And object 692 sets a pattern and fills a rectangle with it....09:16.12 
  AH, of course, x692 is in fact object 1720, that makes sense09:16.48 
  And that form content stream draws 2 XObjects09:17.10 
  THe first of those draws 6 XObjects....09:18.05 
  Oh, and a bunch of other ones later09:18.24 
  hundreds of them09:18.42 
  Classic. The resources dictionary for the form lists objects /s101 to /s155 as pointing to the identical object '1724 0 R' then /s162 to /s325 all point to object 1726, and so on. SO it needlessly defines dozens of duplicate objetcs. Presumably in order to slow down the interpreter....09:20.56 
  As near as I can see that uses 3 actual different objects, but defines a couple of hundred different resources for them.09:22.18 
  And that's just the ExtGStates09:22.46 
chrisl I can't believe cairo haven't been forced to fix this crap!09:23.28 
kens I'm guessing that this has tripped over the nesting level problem again, which is why its generating errors, and that is causing some amount of memory thrashing, but I don't think its actually in an infinte loop, just very slow.09:24.43 
  Just running it through ps2write to see what happens09:25.44 
  Surprise, surprise, it works fine09:26.11 
  Because all the stupid transparency just gets rendered09:26.47 
  Yes increasing the gstate stack depth gets rid of the form errors. However GS does seem to vanish off into a loop then.09:30.01 
  Oh, and then seg faults....09:30.19 
Robin_Watts paulgardiner: hey11:21.55 
paulgardiner hi11:22.07 
Robin_Watts Do you want to take a look at the commits on robin/master ?11:22.10 
paulgardiner Sure11:22.18 
Robin_Watts I spent time on saturday with Sebras, and I think we got it working in the end.11:22.27 
  http://ghostscript.com/~robin/MuPDF.apk is the latest apk. If tor/you agree with the reviews, I'll get that uploaded to google later.11:23.10 
paulgardiner Which ones need reviewing?11:23.12 
Robin_Watts Smart Motion and NullPointer exception.11:24.21 
  but I was thinking that you'd be best placed for the latter, and tor8 for the former.11:24.36 
  (not that I won't welcome comments on both of course)11:24.45 
paulgardiner Ah. I'd fetch only Robin, not origin too, so it looked like there were loads11:26.27 
tor8 paulgardiner: "git remote update" fetches all origins, and submodules11:27.42 
Robin_Watts oh, I missed that tor8 was here too.11:27.58 
  Morning tor8.11:28.03 
paulgardiner tor8: ah right. Thanks11:28.16 
tor8 Robin_Watts: just popped in the minute you started writing, I think, going by the logs. morning!11:28.21 
  I see you were busy this weekend.11:28.26 
Robin_Watts Yeah, the gmail attachment thing was a bit of a bitch.11:28.45 
  Or rather, I made a stupid mistake that cost me hours.11:28.53 
tor8 JNI is a bit of a disaster from what I recall of messing with it 15 years ago...11:29.27 
Robin_Watts I've also started a local branch on which I am experimenting with making mupdf handle rectangles/matrices by reference rather than by value.11:29.38 
tor8 Robin_Watts: I'm not sure there will be any realistic gains from that though (matrix/rect by ref)11:30.01 
Robin_Watts I am sure that working by reference will be faster on ARMs at least.11:30.05 
tor8 since almost every function that takes one modifies it so needs to make another copy anyway11:30.14 
Robin_Watts tor8: actually, no, I don't think that's the case.11:30.30 
tor8 ctm = fz_concat(localthingy, ctm) is in most functions that take a matrix11:30.47 
Robin_Watts indeed.11:31.04 
tor8 Robin_Watts: so make damn sure you set them as "const" or we'll have a bazillion little bugs11:31.17 
Robin_Watts so fz_pre_mul(ctm, localthingy)11:31.32 
tor8 and all functions that take a rect usually intersect it11:31.34 
Robin_Watts yeah, this is why I am trying it on a branch.11:31.52 
tor8 fz_concat(localctm, localthingy, ctm)11:31.56 
Robin_Watts tor8: That's exactly what I have at the moment :)11:32.14 
tor8 fz_concat(fz_matrix *out, const fz_matrix *a, const fz_matrix *b)11:32.22 
  Robin_Watts: re the bbox/rect/bounds naming. any preference?11:32.45 
Robin_Watts We may find that we want to pass matrices by value in the dev interface, but do all the fiddling with them within the functions by reference.11:33.09 
  tor8: I'm still struggling to see what we gain with the bbox -> rect thing.11:33.30 
tor8 Robin_Watts: no, let's not mix11:33.34 
  a naming convention (ctm in arguments, localctm in the modified local matrix) will be better11:33.55 
  Robin_Watts: rect/bbox was a mess in the text device and apps code11:34.24 
Robin_Watts With the new code we end up with rects that happen to be bboxes.11:34.33 
  I'd rather have them specifically named that way.11:34.51 
tor8 now that everything is cleaned up to one type consistently everywhere, we can reintroduce fz_irect in the drawing device11:35.14 
Robin_Watts rect -> bbox is also bad for performance (floats are bad, M'kay)11:35.23 
tor8 Robin_Watts: floats are good, on intel arch's11:35.36 
Robin_Watts tor8: floats are bad, on ARMs.11:35.44 
tor8 stupid arm!11:35.54 
Robin_Watts Thinking back to what one of our customers said...11:36.02 
paulgardiner Robin_Watts: coo nice. That has worked out neatly.11:36.03 
tor8 Robin_Watts: yeah. I'll make a stab at irect in the pixmap and draw device.11:36.46 
Robin_Watts ...they spent a lot of time optimising stuff to avoid FP where possible. And they tweaked all the matrix stuff as it made a big difference for them.11:36.51 
tor8 but first, I wonder if we can do anything about the two rounding functions11:36.58 
paulgardiner Robin_Watts: only thought was whether env and thiz would be better in a structure separate from globals, as it seems that where env and thiz are needed, no other part of globals is.11:36.59 
Robin_Watts paulgardiner: env and thiz being in globals; get_globals updates env and thiz every time.11:37.21 
tor8 Robin_Watts: did that customer use 32-bit fixed point or 64-bit?11:37.31 
Robin_Watts in case env and thiz change between calls.11:37.34 
  tor8: They still used floats, just were more efficient about it, AIUI.11:37.46 
tor8 Robin_Watts: I know we've gone to great care to avoid doubles everywhere11:38.12 
Robin_Watts I think there is scope for us being smarter with the matrix stuff.11:38.22 
  let me have the rest of the day to bash at this stuff, and we'll see how it ends up and take a view on it.11:38.36 
tor8 I don't oppose fz_matrix and fz_rect being passed by value, but if and only if we spread const and be strict about constness everywhere11:39.05 
paulgardiner Robin_Watts: oh yeah ok. Makes sense11:39.08 
  Robin_Watts: looks fine11:39.20 
Robin_Watts tor8: OK, sounds good. I'm trying to add consts.11:39.23 
  paulgardiner: Thanks.11:39.26 
  tor8: Do you want to try the new apk and see if it solves the 'scrolling a tiny amount' problems?11:39.42 
tor8 Robin_Watts: should probably constify most of the rest of the API where it makes sense too11:39.48 
  Robin_Watts: will do.11:39.55 
Robin_Watts tor8: sure.11:39.56 
  Thanks.11:39.58 
tor8 Robin_Watts: oh yeah, icons too, but I guess that can wait for yet another market update?11:41.39 
Robin_Watts IF the icons are ready to go, I can rebuild/upload that11:42.18 
tor8 Robin_Watts: haven't noticed any weird behaviour so far with the new apk so I'd reckon the smart scrolling works11:43.50 
  tapping on the bottom and on the right is meant to behave the same, yes?11:44.34 
Robin_Watts yes.11:45.14 
tor8 Robin_Watts: oh, when you do the matrix/rect by reference branch, where did you base it off? the rect revamp or before?11:45.21 
  (worrying about painful merges)11:45.28 
Robin_Watts before.11:45.34 
tor8 ugh.11:45.37 
Robin_Watts but I'll take the pain of the rebase.11:45.41 
tor8 Robin_Watts: do it in two parts, one for matrix and one for rect. for your own sanity.11:46.02 
Robin_Watts Too late :)11:46.16 
  (both with the commits and with my sanity :) )11:46.25 
tor8 rats!11:46.34 
Robin_Watts Fwibble.11:46.49 
tor8 right, so round_rect adds 0.001 to the values to allow very-near-integer values to round rather than just strictly floor/ceil truncating11:47.31 
  question is, whether that's worth worrying about11:47.49 
  for all internal rects for temporary scratch pixmaps and font glyphs it shouldn't matter11:48.22 
  the only place I see it could be relevant is for the page media box rounding to an integer size to render11:48.52 
Robin_Watts I can only imagine that that was added because we had a specific file that was rendering wrongly.11:49.29 
  possibly because something was working out as 0.9999999 etc.11:49.43 
  Moving from ints to floats will only worsen this kind of problem :(11:50.08 
tor8 Robin_Watts: no it won't. *nothing* changes rendering wise from that patch.11:50.25 
  we probably did the fuzzy because of the white borders that can appear on pages that have a solid fill that exactly matches the mediabox boundaries11:51.14 
Robin_Watts Ah, could be. I'd hope that git blame would show us?11:51.46 
tor8 agh, soooo many uninitialized warnings with release build :(11:52.40 
  huh, no diffs at all with any of the sumatra files if I just do the exact rounding everywhere.11:54.00 
  except the warning for /TR disappeared. did I wrongly simplify that if test?11:54.34 
Robin_Watts I stared at that for a while.11:57.31 
  I think you didn't simplify it, you fixed it.11:57.39 
  You changed a "!tr" to a "tr"11:57.53 
  but if you weren't aware of changing the behaviour we should double check.11:58.13 
tor8 it didn't make sense the way I tried to read it11:58.15 
  so I *think* I fixed it11:58.23 
  gcc threw a warning at it, which prompted me to look and it just didn't make sense11:58.44 
Robin_Watts Then I think we're good.11:59.10 
tor8 Robin_Watts: http://git.ghostscript.com/?p=mupdf.git;a=commit;h=5409d2f3ba310b59074ec590ea69ecd423653d8512:02.36 
  that's where you introduced the two separate ones12:02.47 
Robin_Watts Does that bug have an example file on?12:03.15 
tor8 nope.12:03.24 
  I think though, before we had fuzz with FLT_EPSILON always12:03.33 
sebras tor8: the uninitialized warnings are mine. I tried to point to a fix patch, but you never took it.12:03.42 
tor8 you wanted exact rounding for the temp buffers so introduced the split12:03.52 
Robin_Watts ok.12:05.14 
tor8 I believe we should be okay with exact rounding everywhere though12:08.41 
Robin_Watts exact being +.0001 ? :)12:10.36 
  yes, you may be right.12:10.42 
tor8 Robin_Watts: exact being just plain floor/ceil12:10.58 
Robin_Watts no, that's no good. see the problem the bug was for.12:11.17 
tor8 i.e. fz_bbox_covering_rectt everywhere12:11.20 
  "round slavishly to ensure we have a bbox that covers the rect" is what you added12:11.51 
  "round, allowing for slight calculation errors" is what I'm doubting we have a need for12:12.13 
Robin_Watts ok, sorry.12:12.27 
  If your tests show it's safe, I won't fight it.12:12.38 
tor8 (did you get my meaning taken as opposite)12:12.39 
Robin_Watts you may be right.12:12.43 
tor8 Robin_Watts: I'd be happier with more tests, but I get zero diffs on the sumatra files from master to float-rect-everywhere with slavish rounding12:13.30 
Robin_Watts tor8: clusterpush!12:13.44 
tor8 how do?12:14.01 
  it's ... been a while since I did that. /hangs-head-in-shame I should know this by now.12:14.23 
Robin_Watts If you've set it up, then 'git cluster' :)12:14.41 
tor8 I have not :) I've got the cluster ssh key installed but that's it12:15.01 
  is there an email I can search for?12:15.19 
Robin_Watts If not (and you're not on windows), then PATHTOGHOSTPCL/gs/toolbin/clusterpush.sh12:15.26 
  tor8: I did send an email around about clusterpushing on windows, but you're on your mac, right?12:15.58 
tor8 I'm on linux mostly, actually12:16.16 
Robin_Watts oh, even better.12:16.21 
tor8 Robin_Watts: I don't have that path12:17.19 
  there's a gs/toolbin/localcluster directory with a couple different scripts in it12:17.35 
Robin_Watts gs/toolbin/localcluster/clusterpush.sh, sorry.12:17.37 
  and it might be .pl not .sh12:17.49 
tor8 you mean gitpush.sh or clusterpush.pl?12:17.55 
Robin_Watts clusterpush.pl12:18.17 
  gitpush.sh is my script to allow git based pushes (created for windows users like me)12:18.39 
tor8 Robin_Watts: right.12:18.47 
  what happened to the separate cluster.git repository?12:18.57 
  or is that cluster.git stuff only on the server side?12:19.17 
Robin_Watts That's the server side.12:19.33 
  So, 30 differences.12:30.32 
  Now do: clusterpush.pl bmpcmp12:30.40 
tor8 Robin_Watts: right, so most can't be compared due to ending up a different size12:39.21 
Robin_Watts ok, so we do get stuff a different size. check those manually for white lines etc?12:39.46 
  Other ones look either neutral or progressions to me.12:41.40 
  (In the bmpcmp it's "Candidate, Reference, Diff" (or "New, Old, Diff"). Different to sane.12:42.19 
  and to compare them hold the ctrl button and waggle the mouse into/out of the left hand image to see them swap.12:42.42 
tor8 the javascript doesn't work12:44.09 
  number 23 looks worrying to me, the arrowhead turning blue at the tip12:44.23 
Robin_Watts What browser?12:44.52 
  Yes, that is bad :(12:45.05 
tor8 chrome on windows12:45.11 
Robin_Watts I'm using chrome on windows too.12:45.19 
tor8 the popup zoom works, but it never toggles12:45.26 
Robin_Watts and the javascript works for me :(12:45.28 
  Left hand image for toggling.12:45.44 
  and you need to hold ctrl.12:45.50 
  and move the mouse pointer into/out of the image.12:46.13 
tor8 ah. pushing ctrl while inside doesn't toggle12:46.16 
  you have to cross the image border with control held down12:46.25 
Robin_Watts yeah.12:46.31 
tor8 but it stays stuck after leaving, so now I don't know which is reference anymore :(12:46.56 
Robin_Watts Put the mouse int he middle image and I think it all resets.12:47.46 
tor8 with the mouse over the image, press control and leave the image frame12:47.49 
  and it gets stuck in the toggled state12:47.57 
Robin_Watts yes. If you want to suggest a fix for the javascript please feel free :)12:48.21 
tor8 um, no :)12:48.29 
Robin_Watts It's not perfect, but it's much better than not having it at all.12:48.49 
tor8 I'll learn to live with the way it works! better that than hacking JS in a browser.12:49.11 
Robin_Watts I think it all worked fine until people wanted it put on the ctrl key cos it was confusing otherwise.12:52.06 
tor8 yeah. I'd prefer it without the control key12:52.25 
  or just toggle *all* of them on the page when pressing and releasing control12:52.38 
plauto cannot compile gs from git13:33.49 
  ./autogen.sh && make so13:33.57 
  output here13:34.00 
  http://pastebin.com/EVqEZEVj13:34.07 
  fedora 17 64 bit with gcc 4.7.213:34.28 
  any help?13:34.31 
Robin_Watts plauto: Does a normal build work?13:36.07 
  (i.e. not a .so build)13:36.11 
plauto yes, it works fine13:36.18 
Robin_Watts so, one for chrisl maybe?13:38.13 
chrisl Let me check it.....13:38.41 
  plauto: it works for me on Ubuntu 64 bit 4.7.213:44.17 
  er that should be "with gcc 4.7.2"13:44.38 
kens I have a Feora installed in VM< let me see what gcc I have13:44.47 
Robin_Watts I wonder if crt1.o is being included twice for plauto13:45.18 
  oh, possibly both crt1.o and crti.o are included.13:45.49 
kens Oh I'm afraid the latest 64-bit Fedora I have installed is 1413:46.00 
Robin_Watts lunches13:46.28 
plauto kens: i get linking errors also on a centos 613:48.38 
  gcc 4.4.613:48.47 
  let me see if it's the very same error...13:48.58 
kens Umm, OK let me boot Fedora and see what gcc I have tehre,its probably old htough13:49.09 
  My gcc is only 4.5.113:52.09 
  I'll get a git update and see what happens13:52.21 
  OK updated master to HEAD13:54.21 
  plauto did you do ./autogen.sh in ghostpdl or gs ?13:54.40 
plauto in gs13:54.55 
  hope it's not a problem13:55.01 
kens ok running it now13:55.02 
  Not a problem, just that they are differeny, and I want to make sure I'm doing the same as you13:55.43 
kens twiddles fingers waiting for compilation to finish13:56.43 
  Yes, I get the same problem with gcc 4.5.113:58.41 
  I oculd try updating gcc, what do you think chrisl ?13:58.59 
chrisl kens: I don't think it's gcc version dependant. I'm downloading Fedora.....13:59.46 
kens OK if there's anything I can do let me know13:59.59 
chrisl Find a Fedora maintainer, and smack them?14:00.16 
kens :)14:00.24 
  Just quickly trying a regular make14:00.43 
  I'm 'fairly' sure this used to work14:00.51 
plauto chrisl, if it helps you, very similar problem on centos 6 too14:01.06 
kens I don't normally build an so14:01.07 
chrisl plauto: don't have centos either, and I really don't want to get it - it's packed full of obsolete packages14:01.40 
kens OK regular build is just fine, its the so build14:02.24 
Robin_Watts stops being distracted by mupdf and really goes to lunch.14:03.37 
chrisl I don't see how a different distribution can cause that kind of problem14:04.14 
kens I am also puzzled, but like I said, its quite clear here14:04.31 
  I've shutdown my VM, but if you can't reproduce it I can tart it up easily enough14:04.55 
chrisl It'll be a while, even the download is taking a while, it's a DVD image14:05.21 
Robin_Watts plauto: Dumb question, but have you cleaned before you built?14:31.40 
plauto yes, fresh git clone14:32.04 
Robin_Watts tor8: So, did you pass the smart motion fix review then?14:50.05 
tor8 Robin_Watts: yes.15:06.24 
Robin_Watts Thanks.15:06.31 
Robin_Watts reinstalls MuPDF from the app store, before uploading new apk, to test auto update.15:09.27 
henrys chrisl:thanks for bringing up the blockers I didn't even notice.15:27.30 
chrisl henrys: it was a surprise to me! Total coincidence I noticed when I did15:29.00 
sebras Robin_Watts: so you posted a new MuPDF to Play today?15:41.44 
Robin_Watts sebras: I did. Have you seen it yet?15:41.58 
  It's not showing on my phone yet.15:42.28 
  or on the web.15:42.45 
sebras Robin_Watts: so then you updated the version-number I assue?15:46.56 
Robin_Watts 1.1 (Build 6)15:47.07 
sebras Robin_Watts: how does the android version-number correlate with the normal mupdf one? 1-to-1 mapping?15:47.17 
  oh.. a build number I see.15:47.32 
Robin_Watts The apk's have an internal version number, that the website ensures always increases.15:47.34 
  so I just made that visible. Seemed the simplest way.15:47.47 
sebras ok. I'll test this on my work phone too later tonight.15:52.55 
  had the intentention of doing that sunday but I ended up defrosting my freezer instead. ;)15:53.26 
henrys chrisl:we should be in freeze now right?16:50.48 
chrisl henrys: not with blockers pending, and a critical we really want done, I don't think16:51.23 
henrys I was thinking we should have a freeze announcement and the blockers are obvious exceptions.16:52.06 
chrisl henrys: okay, I'll do it tomorrow morning, if that's okay - it's nearly 5pm here16:52.46 
henrys okay sounds good - I guess we can discuss it at the meeting.16:53.04 
mvrhel_laptop I will see if I can get my item fixed today16:53.09 
henrys also16:53.10 
ray_laptop chrisl: thanks for your comment re: the pdf14 clist optimization. I'll wait 'til after the release.16:56.14 
henrys mvrhel_laptop: sorry you and ray_laptop are on the hot seats.16:56.57 
chrisl ray_laptop: that's fine. I must admit that I'm guess a little at the risk factor, which is why I thought it might warrant discussion tomorrow.16:57.02 
ray_laptop henrys: and I am working on my bugs, although why they just turned up, I don't understand. We've had nightly runs with NumRenderingThreads=4 since I put in Robin's patch.16:57.24 
  I thought I had nailed bug 693590, but it only fixes the cases that were mentioned in the bug report (and about 900 others). The remaining 1,113 are in process16:59.24 
henrys so marcosw didn't report them right away?16:59.27 
ray_laptop henrys: I don't know. I just saw the bug on Sat17:00.09 
  I have to run my dog over to the vet. bbiab17:00.35 
Robin_Watts That sentence is a fabulous example of the importance of punctuation.17:01.47 
  I have to run my dog over. To the vet! bbiab17:02.22 
ray_laptop Robin_Watts: :-)17:02.22 
Robin_Watts ray_laptop: Hope doggy is OK.17:02.37 
ray_laptop Robin_Watts: actually just going for shots and grooming. He's fine17:02.59 
Robin_Watts ah, cool.17:03.16 
henrys and why AI is a bitch.17:03.18 
  I can't wait to see these ubuntu phones - emacs on the phone - cool!17:09.14 
marcosw henrys: the multi threaded rendering test only runs once a week. And there have been several commits in the last few weeks that have changed thousands of bitmaps, so I've been a bit behind at looking at bitmaps and filing bugs. 17:09.49 
  Having so many commits before a release can cause the regression testing analysis to fall behind. And of course just before a release is when regression testing is most important. 17:09.51 
alexcher Robin_Watts: What do you thing about #ifndef NO_MEMENTO for the bug 693284? It's included by default. Clients that define random names are on their own.17:10.14 
henrys marcosw: yeah understood.17:10.59 
Robin_Watts alexcher: JBIG_NO_MEMENTO ?17:12.01 
alexcher Robin_Watts: Fine for me.17:12.24 
Robin_Watts cool.17:12.57 
marcosw we should have a one week code freeze before release, to allow the regression testing to catch up :-)17:13.28 
henrys marcosw:well we'll probably freeze everything except blocker work tomorrow.17:15.28 
marcosw yes, but commits will continue, so we won't be testing the actual shipping 9.07. I think we discussed this 6 months ago… Also we tend to not freeze for an entire week.17:16.37 
chrisl marcosw: I create a release branch which only gets commits required for the release, so you can test that17:18.16 
marcosw chrisl: Historically I haven't been pointing the regression tests at the release branch, I will do it this time.17:19.59 
ray_laptop it should be easy enough for us to hold off "optional" commits for a week. If we are concerned about losing a local copy, just email the patch to yourself, or stick it up on casper17:20.29 
chrisl ray_laptop: the idea behind the release branch was to have an explicit "gate" for commits going into the release, to reduce the chance of accidentally committing during the "freeze".17:21.45 
  marcosw: the release branch will be called "gs907"17:22.15 
ray_laptop we all just work on a 'during_freeze' branch and then merge it to the 'master' after the freeze17:22.29 
kens I have to go goodnight all17:22.50 
ray_laptop so that marcos' testing can use the master17:22.55 
  bye, kens 17:22.59 
mvrhel_laptop darn 693418 does not have the problem with any gray device. looks like I have to fool with cups with -dcupsColorSpace=017:23.05 
  how to I get cups working in windoze again...17:23.17 
chrisl ray_laptop: that's how we used to do it, and people didn't like it, that's why I changed.....17:23.17 
ray_laptop I almost have the file downloaded for our yet to be numbered customer bug 69359217:24.27 
  oops wrong bug. should be bug 69357917:25.14 
henrys we need a no number number ;-)17:26.01 
ray_laptop I missed marcosw to ask if he was going to see if he can reproduce it.17:26.03 
  henrys: just email Joann or Miles and ask for a number17:26.32 
mvrhel_laptop aha. WITH_CUPS=117:27.37 
  hmm that did not work17:29.39 
chrisl mvrhel_laptop: you probably need to "clean" as well17:30.06 
mvrhel_laptop yes. doing that now..17:30.13 
  still failed.17:34.03 
  I have it set in the Nmake preprocessor setting17:34.17 
  where all the others are set (e.g. SSE2 etc)17:35.01 
chrisl Is it on the nmake command line?17:35.05 
mvrhel_laptop no. but I had thought it would go where the other options went17:36.23 
  let me stick it in the Build command line directly17:36.39 
chrisl WITH_CUPS is an nmake option, not a preprocessor directive17:37.22 
mvrhel_laptop ok 17:37.47 
henrys ray_laptop:I just sent a email to them, funny I thought I had already done that but didn't see an outgoing email so I guess I didn't.17:46.40 
ray_laptop I can't reproduce bug 693579. I guess Marcos will have to find out more info from the customer. Also looking at the file, I can't see any reason it should leak.17:47.55 
  marcos should get my comments since the bug is assigned to him17:48.20 
  back to bug '590 17:48.39 
mvrhel_laptop finally... running17:50.30 
  good grief. in my crummy linux in hyperview this is going to take all day just to get set up to view the damn files18:03.18 
  so looking at the output, I am puzzled as to what the issue is18:11.41 
  henrys: I am going to need someone else to see if they see any diffs with this18:12.37 
  I am running the win32 build, but on my 64 bit machine18:13.17 
  so this is bug 69341818:13.39 
  oh comment 1 mentions a segv18:14.44 
  I guess I should run that in linux to see18:14.52 
  I was looking at comment 018:15.09 
henrys mvrhel_laptop: hard to understand how your change would be related to that kind of crash.18:18.50 
mvrhel_laptop yes. I am trying to see if I can get the segv in linux now.18:19.08 
chrisl plauto: I have fedora up and running now, and I can see the problem. What's happening is that the link step is being called (at least) twice. The second time, it's being called as if it were linking a full executable (so if you do "file libgs.so.9.07" it lists it as an executable instead of a shared object). Obviously, when we then try to link with an object file to produce the intended executable(s) we get the multiple18:24.16 
  definitions error. I currently have *no* idea why this is happening, and I need to finish for the day......18:24.16 
mvrhel_laptop ?? is cups not included by default in the linux build18:35.18 
  henrys18:35.22 
  ?18:35.24 
  oh shoot I have to head out to the kids school now to see my daughters presentation. back in a while18:35.48 
alexcher marcosw: Is there a way to combine 'lowres' and 'gs' regression modes? 18:41.56 
ray_laptop mvrhel_laptop: I have a question (when you get back is OK). Please call me.18:43.13 
plauto chrisl: thanks18:46.03 
Robin_Watts alexcher: clusterpush.pl gs lowres ?18:49.53 
ray_laptop Robin_Watts: in mem_planar_fill_rectangle_hl_color the loop for each plane is setting up for plane_depth in the loop. Does it _really_ have to do that ?20:15.45 
Robin_Watts let me look.20:27.04 
  Ghostscripts planar stuff is kinda schizophrenic.20:27.56 
  Some of it allows for planes with different depths, some of it doesn't.20:28.11 
  This code does, hence it has to get the mdproto each time.20:28.37 
  You could potentially short circuit that if you detect that the bit depth is unchanged.20:29.37 
ray_laptop I can't imagine a device with ddifferent depths per plane20:29.39 
Robin_Watts but really, gdev_mem_device_for_bits is just an array load.20:30.09 
ray_laptop true20:30.23 
mvrhel_laptop ray_laptop: I am here now20:40.15 
  ok. what magic do I need to do to have cups built in linux now to check for this segv20:41.45 
Robin_Watts Do you have cups installed on your linux box ?20:42.26 
mvrhel_laptop no20:42.31 
  that there is probably the problem20:42.39 
Robin_Watts Then configure may avoid it.20:42.40 
mvrhel_laptop ok. let me start that process20:42.52 
  well, I do have some cups related stuff installed. 20:47.07 
  I am not sure what all I need20:47.12 
  what a pain20:47.20 
  I have the Common Unix Printing System installed already20:48.08 
Robin_Watts ./autogen.sh --with-local-cups ?20:48.19 
mvrhel_laptop ok that may do it20:48.56 
Robin_Watts [Force using the GS supplied cups code - only useful for debugging]20:49.04 
mvrhel_laptop ok that worked. thanks Robin_Watts 20:51.36 
Robin_Watts np.20:51.41 
mvrhel_laptop and the page rendered fine20:51.43 
Robin_Watts Now if you can tell me why I've lost all type3 glyphs in mupdf, we'll both be happy :)20:52.02 
mvrhel_laptop henrys: so I can't duplicate any of the issues in the bug20:54.03 
  marcosw: I am going to punt this one back to you20:54.15 
marcosw uh oh. which bug?20:54.35 
henrys one of those blocker ones20:55.30 
mvrhel_laptop 69341820:55.44 
  I just changed it to WORKSFORME20:55.50 
  1/2 a day spent screwing with cups20:56.11 
marcosw which operating system where you using?20:56.46 
mvrhel_laptop linux and windows20:56.53 
henrys I thought we had this set up so it is easily worked on in windows?20:56.57 
mvrhel_laptop well I had to find the magic value to set20:57.08 
  to include it20:57.13 
  when I use something like that once a year I forget20:58.03 
  and that all worked fine20:58.13 
  so I looked for the segv that was reported on linux with the bug20:58.28 
  and I could not get that to occur20:58.33 
Robin_Watts mvrhel_laptop: To build with Cups on windows, just select the "Debug-cups" configuration in visual studio.20:58.47 
mvrhel_laptop oh crap20:58.50 
  I see that the comment says debug build20:58.57 
  I was running a release build20:59.01 
  let me try the release build20:59.07 
  i mean the debug build20:59.11 
  Robin_Watts: aha!20:59.18 
  that is easy20:59.24 
Robin_Watts I thought you added that? :)20:59.34 
mvrhel_laptop chris told me to change the nmake command20:59.50 
Robin_Watts chris is an old school command liner :)21:00.05 
mvrhel_laptop anyway, let me try the debug version to see if it segvs21:00.22 
henrys now you can spend 2 1/2 days screwing with cups21:00.30 
mvrhel_laptop :)21:00.35 
  ok. the debug build ran fine too. whew21:05.38 
  marcosw: so maybe you can recheck?21:05.48 
marcosw mvrhel_laptop: yes, I'm doing that know. I'm running a clean build on peeves.21:06.29 
mvrhel_laptop ok. thanks. 21:06.36 
  meanwhile, I think I see how to fix my vmware hyper-v issue.....21:07.00 
  brb. restarting to see if I can turn off hyperv21:09.43 
  success. I am able to use vmware again21:15.42 
marcosw mvrhel_laptop: Bug 693418 appears to be fixed. I'm curious, so am doing to run a bisect.21:15.51 
mvrhel_laptop marcosw: that is good news to me21:16.01 
  marcosw: it is probably a fix that Robin_Watts did21:16.25 
marcosw darn that Robin_Watts, fixing other people's bugs.21:16.42 
mvrhel_laptop it the issue is from the planar stuff that I did21:16.45 
  that is my approach to bugs like that one. wait for someone else to fix their bug and mine gets fixed as a bonus21:17.42 
henrys down to 1 blocker - good news21:18.59 
mvrhel_laptop now, to set up my vmware ubuntu machine which I had unfortunately deleted when I was thinking I would do a full move over to hyperv21:19.07 
henrys we have to get this out the customers so they can test it ;-^21:19.34 
  Are we going to move to 2012 project files soon?21:21.14 
mvrhel_laptop henrys: Good question about 2012. I think I am the only one running 201221:22.59 
henrys I used it for a couple of fixes recently.21:24.09 
mvrhel_laptop marcosw: how did the bisect end?22:11.24 
marcosw not well, can't reproduce the seg fault at all. 22:11.55 
  I'm going to be at home soon, will look for the bitmap difference issue then.22:12.35 
sebras tor8: Robin_Watts_: I see that you forgot to run the text-selection feature in mupdf by your testing guy...23:07.00 
  Robin_Watts: tor8: I was using my work phone and managed to first get MuPDF into a zoomy state and eventually it appears to have crashed.23:34.55 
  smaller.mp4 at ~sebras/tmp and original.mp4 for the adventurous...23:46.35 
  I have tried to reproduce this on my sensation, but I fail to do so.23:48.56 
  so it might very well be something tied to the android version in use.23:49.09 
  on my sensation: 2.3.3 on my samsung: 2.3.6 apparently.23:49.49 
  my work here is done >;) good evening.23:50.42 
 Forward 1 day (to 2013/01/29)>>> 
ghostscript.com
Search: