| <<<Back 1 day (to 2014/08/18) | 2014/08/19 |
kens | Bugzilla seems... unhappy | 08:54.54 |
pedro_mac | kens: seems ok here - is it still broken for you? | 09:03.02 |
kens | Not broken just excruciatingly slow. But its intermittent | 09:03.26 |
pedro_mac | ah, ok | 09:03.42 |
| performance seems relatively normal for me just now | 09:03.59 |
kens | It comes and goes for me, and its not my end because other web pages are sprightly enough | 09:04.16 |
pedro_mac | nods | 09:04.26 |
kens | It jkust took several minutes to load the page for a specific bug | 09:04.37 |
henrys | canât sleep - way to get after the security pests kens | 09:33.25 |
kens | :-) I'm happy to have security issues raised, but these people need to think a bit | 09:34.03 |
kens | tried to remain polite | 09:34.49 |
| more coffee...... | 09:35.09 |
Robin_Watts_ | morning tor8. | 09:46.22 |
| zeniko has some patches. | 09:46.27 |
tor8 | Robin_Watts_: morning. okay, I'll look them over. | 09:46.38 |
chrisl | I have to go play chauffeur - back in a couple of hours..... | 10:04.30 |
kens | have fun :-) | 10:04.38 |
| Welcome back 'Parker' :-) | 13:13.09 |
chrisl | Well, parents have their own car back now, duties done for now...... | 13:13.45 |
henrys | oh my, the day of the meeting is upon us and Iâve not a thing to say. Iâll think of something. | 14:16.38 |
rayjj | morning, all | 14:29.50 |
kens | Morning | 14:29.55 |
rayjj | kens: see -- you found something to say ;-) | 14:30.10 |
henrys | so fred is now consulting with us to port michaelâs gsview UI to QT as dicussed, he making progress. Do you want me to get him started in GIT and a personal repo. Given the short term nature of his work I donât know if the boot up time is worth it. Thoughts? | 14:30.17 |
kens | THat was henrys I think | 14:30.24 |
Robin_Watts | fred? | 14:30.34 |
henrys | yes it is a friend of miles who needs a temporary job while he waits for startup money. So we put him to work on the QT port. | 14:31.19 |
Robin_Watts | I would have thought that it was pretty crucial for him to be using git. | 14:31.30 |
| but it doesn't need to be a git repo on our server. | 14:31.45 |
henrys | I think you were here for that meeting but you were on vacation. Donât remember. | 14:31.45 |
Robin_Watts | I missed that entirely. Sounds good though. | 14:31.57 |
kens | Was that a Skype meeting ? I didn't see it | 14:32.00 |
chrisl | I agree with Robin_Watts, I think we want it done in git | 14:32.15 |
kens | But I'm with Robin, he'll need Git access, for MuPDF if nothing else | 14:32.21 |
rayjj | for the longest time, peeves locally wouldn't display -- I thought the display card had gone, but after the weekend power cycle it's working again. ;-/ | 14:33.27 |
Robin_Watts | Lots of people have used git, but assume that git == github. | 14:33.29 |
henrys | okay Iâll talk to him about git. Itâs no big deal just to create an account for him, I think that will be okay if he isnât a gitphobe that is. | 14:33.55 |
chrisl | rayjj: possibly the card overheated? | 14:33.58 |
henrys | rayjj: how is the C company port coming? | 14:34.32 |
rayjj | henrys: coming along. Reading *copious* amounts of documents, and trying their "reference code" | 14:35.09 |
Robin_Watts | henrys: My point is that he may be quite comfortable using github, but less so using our server. That need not be a problem though, as we're open source, right? | 14:35.19 |
henrys | Robin_Watts: thatâs a good point, Iâll ask him. | 14:35.54 |
rayjj | as long as he can format patches with comments and send to (one of) us to be committed after review, that should work | 14:36.22 |
chrisl | rayjj: what's the OS? Is it proprietary? | 14:36.41 |
henrys | kens: well at least support is light now. | 14:37.23 |
rayjj | chrisl: ThreadX | 14:37.30 |
henrys | marcosw returns next week? | 14:37.32 |
chrisl | rayjj: Ugh :-( | 14:37.43 |
kens | henrys, yes Marcos must have forgotten to send out his email. Though there's plenty of free user activity.... | 14:37.47 |
henrys | is there anything agenda-wise that can be ticked off? Other than that I didnât have much else for this meeting. Anybody else? | 14:39.39 |
kens | has nothing right now | 14:40.01 |
chrisl | henrys: any updates on the font front? | 14:40.21 |
henrys | chrisl: nope I guess theyâre working on them. | 14:40.58 |
| chrisl: oh you did see the lurateck release? | 14:41.08 |
| luratech sorry | 14:41.17 |
tor8 | henrys: Robin_Watts: he can do his work on github and we can pull from there, without any trouble at all | 14:41.24 |
Robin_Watts | tor8: Indeed. | 14:41.33 |
chrisl | henrys: I did. I'm thinking as it was seg faults, we should roll that into the next release - but I wanted to check with Marcos first | 14:41.41 |
henrys | tor8, Robin_Watts okay seems easier than setting up an account | 14:41.58 |
kens | chrisl when (approx) is the next release, September ? | 14:42.35 |
rayjj | cust 532 (Len) just called yesterday with a question, but not a real bug. the ATS CDX3E1EC_Save.pdf p2 works differently on different PDF consumers. mupdf matches Adobe, but gs HEAD looks "right" | 14:42.50 |
chrisl | kens: Yes, I haven't narrowed it down any further, but I'd *like* it done around mid-Sept | 14:43.08 |
kens | NP, should be possible to get Luratech into that I'd have thought | 14:43.27 |
| rayjj : where's the file ? | 14:43.47 |
chrisl | kens: I'd hope so, but it's mostly going to depend on regression testing it, hence wanting to check with Marcos | 14:44.06 |
rayjj | tests_private/pdf/PDF_1.7_ATS/CDX3E1EC_Save.pdf | 14:44.30 |
kens | goes to grab a copy | 14:44.43 |
rayjj | about 44Mb | 14:45.00 |
kens | O.O | 14:45.11 |
chrisl | Hrm, I don't have that file.... | 14:45.21 |
henrys | chrisl: I guess I could hardwire a local push to be luratech and see what comes back. | 14:45.36 |
kens | At 44Mb I'm not surprised, its probably why I don't have it either | 14:45.39 |
henrys | marcosw is a week out right? | 14:45.47 |
kens | Back on the 26th IIRC | 14:46.16 |
chrisl | henrys: we could do that. I've been wary about it since the security on the cluster dashboard is "simple" | 14:46.19 |
henrys | chrisl: good point | 14:46.36 |
rayjj | kens: yeah, I don't keep the ATS files locally either. The ATS is about 11Gb unless you want a local svn of it, then that doubles | 14:47.23 |
henrys | I have some other news which Iâll send to âtechâ now. | 14:49.38 |
kens | is duly intrigued.... | 14:49.55 |
mvrhel_laptop | sorry I am late. | 14:49.57 |
chrisl | rayjj: what aspect of the page was Len querying? | 14:50.48 |
kens | still waiting for the file to arrive :-( | 14:51.11 |
rayjj | chrisl: look at p2 with Acrobat vs. current gs -- it should be obvious | 14:51.13 |
Robin_Watts | we *could* enable https on casper. | 14:51.14 |
| and use that for the dashboard. | 14:51.23 |
chrisl | rayjj: that means firing up a VM...... | 14:51.37 |
Robin_Watts | It might enable us to enable viewgit for the SOT repo too. | 14:51.37 |
rayjj | chrisl: the text along the left edge is missing and the top of the sundial is mostly not painted (with mupdf and Acrobat) | 14:51.58 |
| chrisl: you need a VM for mupdf ? | 14:52.30 |
kens | GS looks right to me, Acrobat doesn't | 14:52.41 |
chrisl | rayjj: Acrobat | 14:52.44 |
rayjj | kens: exactly | 14:52.47 |
chrisl | Robin_Watts: I don't know how much extra grief we might get with https - it might be worth considering | 14:53.41 |
kens | It'll take quite some work to figure out which is correct for sure | 14:53.52 |
rayjj | kens: I'm bisecting now. 9.06 (their base) is partly wrong (the text is there, but the sundial is broken) | 14:54.13 |
henrys | the QT discussion is here http://ghostscript.com/irclogs/2014/08/12.html, I think everyone was about. | 14:54.14 |
kens | Oh yes, but I don't think you said who it was ? Maybe I just missed it | 14:54.41 |
| Interesting, another rip of our acquaintance throws an invalid access on the image | 14:55.32 |
chrisl | Holy crap, Acrobat is *so* wrong! | 14:55.33 |
kens | chrisl Well it certainly *looks* wrong | 14:55.45 |
rayjj | chrisl: mupdf (recent) matches Adobe | 14:56.11 |
chrisl | I can't imagine anyone would design a page to look like that | 14:56.11 |
| Rather smacks of softmasks | 14:56.55 |
kens | THe old version of that rip agrees with MuPDF and Ac robat | 14:57.09 |
| GS is looking to be in a minority ehre | 14:57.18 |
Robin_Watts | henrys: right, sorry, I saw the qt bit of that discussion, but missed the initial "there is someone that miles wants to employ temporarily..." | 14:58.15 |
henrys | Robin_Watts: yeah I think you were on Easter Island for that one? | 14:58.57 |
Robin_Watts | possibly, yes. | 14:59.03 |
| though I'm sad enough to have read the logs, mostly :) | 14:59.20 |
kens | I always skim the logs when I get back | 14:59.52 |
| chrisl rayjj extracting just page 2 causes the same behaviour, whihc is a start at reducing it | 15:00.09 |
chrisl | There's fairly extensive reliance on softmasks of some form in there | 15:00.42 |
kens | There certainly is, the entire page has *lots* of transparency in it | 15:00.59 |
rayjj | kens: except if you look at the other CDX3E1EC_* variants (_A_1B.pdf and _Writer) gs matches p2 | 15:01.08 |
kens | rayjj maybe, but that doesn't prove its correct. It mayh be what the author intended but still..... | 15:01.37 |
chrisl | kens: I don't know if it's using softmasks, or softmask groups | 15:01.51 |
kens | The text is buried under something in Acrobat, it appears ot be an image not text | 15:01.56 |
rayjj | I think that the original Corel Draw output through Adobe PDF Writer is probably what the creator intended | 15:02.37 |
| i.e., what HEAD gs produces | 15:02.55 |
mvrhel_laptop | where is this file? | 15:03.23 |
kens | Corel Draw writes *very* bad PDF files. THat one with teh multiple inks of the same name in DeviceN was from Corel | 15:03.23 |
| mvrhel_laptop : See logs above | 15:03.32 |
rayjj | mvrhel_laptop: tests_private/pdf/PDF_1.7_ATS/CDX3E1EC_Save.pdf (cut and paste from above) | 15:04.06 |
henrys | email sent weâll keep that off IRC. | 15:04.10 |
| mvrhel_laptop: oh youâre here. | 15:04.24 |
mvrhel_laptop | ok I see thanks. | 15:04.50 |
henrys | mvrhel_laptop: fred would like gsview to be a better desktop citizen, menuâs and help dialogs I think he mentioned I said have at it. But I didnât know if you wanted to talk to him also. | 15:05.18 |
kens | henrys re your email, is Joseph coming to CHicago, or is this going to be via telecoms ? | 15:06.02 |
mvrhel_laptop | henrys: I would love to get feedback from him | 15:06.09 |
henrys | kens heâs coming to chicago | 15:06.25 |
kens | RIght, thanks, welcome to the frequent flyers club jogux :-) | 15:06.44 |
jogux | kens: thanks :) | 15:06.57 |
mvrhel_laptop | henrys: if he wants to make changes directly to the windows version that is good too | 15:07.12 |
henrys | mvrhel_laptop: he does seem to have a handle on it, Iâll just let him go with it. | 15:07.51 |
mvrhel_laptop | henrys: that sounds great | 15:07.59 |
kens | rayjj : chrisl if I remove enough stuff form the PDF file, then it matches Acrobat et al. I think there's something wrong with GS on this one. | 15:08.53 |
kens | hopes rayjj wasn't expecting me to fix this one...... | 15:09.55 |
chrisl | So you think the page is supposed to look like Acrobat's rendering?? | 15:10.24 |
kens | chrisl I think its *supposed* to look like GS's rendering, but the page is incorrectly created by Corel and technically should render as per Acrobat | 15:11.10 |
chrisl | kens: the poppler rendering matches Ghostscript | 15:11.29 |
kens | Well transparency is notorious for these kinds of 'problems' | 15:11.51 |
| Let me send you a reduced file. | 15:11.58 |
| Still 3.4 Mb :-( | 15:12.26 |
chrisl | I'm not going to try PDF.js :-) | 15:12.27 |
kens | No, that's pointless its always wrong on transparency anyway :-) | 15:12.40 |
chrisl | And wrong on many other features, too....... | 15:13.00 |
kens | chrisl there appears to be a softmask in place which is clipping out the text, and much of the sundial. Clearly GS isn't rendering that softmask (or softamsk group) the same as Acrobat and others. | 15:15.32 |
henrys | chrisl, Robin_Watts I forgot to ask at the meeting can we get download stats from Go Daddy. They should offer that, right? | 15:15.37 |
chrisl | kens: I'll need to poke the innards of the to see what form the softmask takes...... | 15:16.43 |
kens | chrisl, yes, I just find it interesting that *removing* some content causes GS to stop rendering the text, it makes no difference to the other rendering engines. | 15:17.21 |
| So I suspect that we are closing a group when we shouldn't or something similar. But its going to take a *lot* fo work to figure it out. | 15:17.46 |
rayjj | kens: well, 9.06 matches Adobe | 15:18.02 |
Robin_Watts | henrys: See ~robin/DownloadsInformation.txt to get log in details. | 15:18.17 |
| Then you can log in and see bandwidth details etc. | 15:18.31 |
henrys | thanks | 15:18.32 |
kens | rayjj well, I'm inclined to think that's 'correct', but like I said, its going to mean a lot of working with the file to figure it out | 15:18.33 |
chrisl | rayjj: you're bisecting, aren't you? | 15:19.06 |
rayjj | kens: but we had so much that we've fixed with SMask from 9.06 on | 15:19.11 |
kens | rayjj I'm not saying that we aren't correct now, but MuPDF, Acrobat (X and XI) and Jaws all agree | 15:19.51 |
rayjj | chrisl: bisecting is proving painful. the version it checks out won't build (missing ./autogen.sh for example, and missing psi directory) | 15:19.55 |
kens | Missing psi ? That can't be right surely...... | 15:20.16 |
chrisl | It's where we merged the history from the jbig2dec repo | 15:20.38 |
| rayjj: do git bisect skip to find the next usable commit | 15:21.03 |
Robin_Watts | henrys: What stats are you after? | 15:21.30 |
rayjj | kens: chrisl: the repo 6ef4989 and 606ce11 won't build | 15:21.34 |
Robin_Watts | We run at about 1.6TB downloaded each month. | 15:21.50 |
chrisl | rayjj: like I said, you can "bisect skip" to find the next usable commit | 15:22.38 |
henrys | oh okay I did want to know that. I donât suppose they have logged bad/incomplete downloads. | 15:23.14 |
Robin_Watts | henrys: no, not sure they could. | 15:23.31 |
| what with caches etc, how can they know what's incomplete? | 15:23.48 |
rayjj | is there a way to force git to correctly update to a commit (during a bisect). | 15:23.58 |
Robin_Watts | git checkout ? | 15:24.10 |
rayjj | I can't imagine that we had so many commits with psi missing | 15:24.19 |
chrisl | rayjj: that's the history from jbig2dec and jbig2dec doesn't have a "psi" directory | 15:24.46 |
henrys | Robin_Watts: anything that failed because of a quota issue would be logged. | 15:24.47 |
Robin_Watts | bandwidth says: 845GB/infinity | 15:25.11 |
rayjj | git checkout gs/psi error: pathspec 'gs/psi' did not match any file(s) known to git :-( | 15:25.19 |
Robin_Watts | so, we have no download quota, AIUI. | 15:25.21 |
henrys | Robin_Watts: they seem to have been confused about that in the past and we had another report recently of a user not able to download. | 15:25.37 |
Robin_Watts | Yes, I saw the user complaining about not being able to download. | 15:26.30 |
chrisl | kens: it's a softmask group :-( | 15:26.44 |
kens | Yeah I thought it might be | 15:26.53 |
Robin_Watts | But it worked for chrisl and me, so I have to wonder if a cache between her and godaddy had a truncated version in. | 15:27.05 |
henrys | Robin_Watts: but moving that much data Iâd expect a few glitches so maybe no big deal. | 15:27.12 |
Robin_Watts | IME, godaddy have never turned downloads off because of a quota issue. | 15:27.42 |
rayjj | chrisl: but the other directories for gs aren't "disappeared" | 15:27.57 |
chrisl | rayjj: are they populated? | 15:28.19 |
Robin_Watts | I think when we complained about there being a problem before, they quibbled that we shouldn't really be using that much bandwidth, but they found and fixed a real problem and we were sorted again. | 15:28.22 |
| rayjj: git reset --hard ? | 15:28.32 |
| assuming you're using a vanilla version. | 15:28.42 |
chrisl | rayjj: as I said, I just use "bisect skip" until I reach a suitable commit | 15:29.15 |
rayjj | chrisl: do you know how to find the jbig2dec range (for git bisect skip <range>) or do I just have to keep skipping | 15:29.36 |
| this makes git bisect pretty unusable | 15:29.58 |
chrisl | rayjj: I don't know the range, I've never had to do it more that 3-4 times | 15:30.15 |
| rayjj: I might point out that we discussed this issue *at length* before tor8 did the history import, and this problem was made clear to *everyone* | 15:31.16 |
Robin_Watts | henrys: When is the SOT meeting these days? | 15:32.08 |
chrisl | kens: could this be related to the SMask being Luminosity rather than Alpha? | 15:32.57 |
kens | chrisl I haven't dug any further into the file I'm afraid | 15:33.12 |
chrisl | Nope, seems not...... | 15:34.07 |
| rayjj: if you're having trouble, I can run the bisect | 15:37.16 |
rayjj | chrisl: I put in a while loop: while [ ! -f gs/autogen.sh ] ; do git bisect skip ; done | 15:39.06 |
| it's still going :-( | 15:39.20 |
| ahh. just stopped at 7de89cf | 15:39.58 |
tor8 | rayjj: git log --all --oneline -- gs/autogen.sh if you're looking for a commit where a file was introduced | 15:42.29 |
rayjj | I just prefix that 'skip' loop before trying to cd gs ; etc.. and it works fine. | 15:43.02 |
| tor8: thanks. | 15:46.25 |
| 4 steps to go... | 15:47.07 |
marcosw_ | rayjj: for next time: "git bisect skip b536be3..7a82fc0" | 15:47.50 |
tor8 | rayjj: there's a mail from 13 sep, 2012 with the subject "git bisecting with jbig2dec" | 15:48.10 |
rayjj | kens: So the ColorConversionStrategy put_param is what's doing the damage -- it rc_decrements the old device_profile which frees the 'name', but there is still a dictionary with OutputICCProfile pointing to that string | 15:48.53 |
kens | rayjj I think it needs to do the rc_decement, but maybe Micahel knows | 15:49.27 |
rayjj | tor8: I guess I've just been lucky (and haven't done much bisecting) since then. | 15:49.28 |
tor8 | rayjj: chrisl: easiest solution (mentioned in that email) is to tell git bisect to only look at the files you suspect "git bisect start -- gs" to only look at changes in the 'gs' directory | 15:49.51 |
rayjj | kens: I am looking to find out how to get that dictionary to update correctly | 15:50.18 |
tor8 | rayjj: if it's only come up once in the two years since we did the merge, I'm happy that it didn't cause more trouble, and worried that the next time it happens we'll have forgotten what to do about it :) | 15:51.10 |
rayjj | kens: and trying to find which dictionary that is | 15:51.15 |
kens | rayjj I don't know which dictionary it would be, maybe something in the startup code. | 15:51.36 |
rayjj | the git bisect start -- gs will probably work best for me | 15:51.43 |
kens | I don't *think* pdfwrite maintains any pointers to the ICC profiles | 15:51.52 |
rayjj | kens: I think it is in the mess of gs_setpd.ps | 15:51.56 |
| kens: no, the device pointers are OK | 15:52.13 |
kens | Yes, the setup code was rather what I expected :-( | 15:52.36 |
rayjj | kens: and you are correct that we need the rc_decrement to avoid leaks (the name is in non_gc_memory) | 15:52.44 |
kens | tor8 you seem to have an number of compilefail cluster regression reports coming in | 15:53.28 |
| Starting with: | 15:53.44 |
| 64e6858 don't unnecessarily include jmemcust.h when SHARE_JPEG isn't defined | 15:53.44 |
tor8 | kens: yeah. I accidentally pushed a commit I shouldn't have (and then force-push backed it out) so that might cause some hickups :( | 15:54.09 |
rayjj | kens: the setpagedevice 'sticky' aspect is what's causing some of the problem -- it seems to be merging the old (invalid) pointer instead of using the current | 15:54.15 |
kens | tor8 I wondered about that, it looked like the same commits twice, NP | 15:54.29 |
tor8 | I slipped up when rebasing zenikos patches and messed up omitting the one I wanted to omit | 15:54.46 |
kens | is off for the night | 16:03.15 |
| GOodnight all | 16:03.19 |
chrisl | marcosw_: if we pull in the new luratech release real soon, does that give you enough time to regression test it before a mid-Sept release? | 16:06.05 |
rayjj | chrisl: (kens for the logs): the patch which allows gs to render page 2 as CD expects is a4e3e55 "Bug 693115: Fix Smask handling in the interpreter and prevent multiple rendering" | 16:11.00 |
| not too surprising given that you guys figured out it was an SMask issue | 16:11.28 |
| now I have to decide if I want to expose cust 532 to that change. There might be dependencies on other SMask related changes :-/ | 16:12.57 |
chrisl | rayjj: surely we need to fix the current code to match Acrobat? | 16:18.59 |
rayjj | chrisl: well... | 16:36.07 |
| chrisl: unless it is clearly something we are doing wrong, seen in a "sane" example, I think we are better off displaying what Corel intended (but couldn't get right) | 16:37.11 |
chrisl | rayjj: we generally strive to match Acrobat - even when Acrobat is clearly wrong by the spec, so...... | 16:38.09 |
marcosw_ | chrisl_away: sorry, was at dinner. yeah, the luratech code is only one day's work. | 17:41.04 |
henrys | Robin_Watts: we do have fix the gray detection⦠http://bugs.ghostscript.com/show_bug.cgi?id=695424, it would be better to keep you on what youâre doing and have tor do it, but I also got the impression you wanted to do it. let me know | 19:47.37 |
mvrhel_laptop | bbiaw | 23:01.17 |
| Forward 1 day (to 2014/08/20)>>> | |