| <<<Back 1 day (to 2011/11/14) | 2011/11/15 |
mvrhel | henrys: tomorrow I will start ripping out wts | 06:38.55 |
vdp | Robin_Watts: hi! just for the record... Taking hints from your conversation with tor8 from yesterday I decided to try an older commit. I've tried 28e88083d782eb and it indeed renders the pages as it should(and still faster than poppler) | 08:22.59 |
Robin_Watts | vdp: I committed a fix yesterday. | 10:06.13 |
| If you're using the latest git version, it should work fine. | 10:06.25 |
| Hey marcosw_ - did you get my mail last night? How's Germany? | 12:24.46 |
marcosw_ | Robin_Watts: germany is freaking cold (2 degrees for today's high). I haven't checked email since yesterday afternoon, will do so later today. | 13:36.40 |
Robin_Watts | A change from balmy california, eh? :) | 13:37.27 |
| vdp: Actually the fix is only in my repo, not in the main one as yet. | 13:39.09 |
vdp | Robin_Watts: yes that's what I was going to say :). Thanks for fixing it so quickly | 13:39.43 |
Robin_Watts | np. | 13:39.50 |
| tor8: ping? | 13:56.46 |
tor8 | pong | 13:56.51 |
Robin_Watts | In mupdf, should I ever see a call to fz_draw_begin_group... then fz_draw_clip_path, then fz_draw_end_group | 13:57.20 |
| without an intervening fz_draw_pop_clip ? | 13:57.30 |
tor8 | you shouldn't | 13:57.43 |
Robin_Watts | I am. Will look into why after lunch. | 13:57.56 |
| tor8: Dumb question... in pdf_show_path, why do we 'begin_group' if (dofill || dostroke)' ? | 14:50.27 |
tor8 | to add the soft mask if it's set at the gstate level | 14:52.07 |
Robin_Watts | ok. | 14:53.53 |
| tor8: I've got a file here that runs us out of clipping state stack. | 15:10.45 |
| needs 163, rather than 96 | 15:13.21 |
| Any thoughts on how we should proceed at that point? | 15:14.12 |
| I could throw an error and abort rendering. | 15:14.30 |
| Or we could move to using dynamically allocated stack, and growing it. | 15:14.50 |
| Or I could keep incrementing the stack depth, and make every rendering thing check bale if the stack is too large; so we'd render 'most' of the page, and just skip the stuff that's overly stacked. | 15:16.10 |
| Or we could keep rendering, and just ignore the clipping etc done by the overflowed stuff. | 15:19.27 |
| Dynamic stack wins for me, I reckon. | 15:20.24 |
tor8 | Robin_Watts: bail, not bale, unless you're talking about hay ;) | 15:33.56 |
| dynamic stack sounds best imo | 15:33.57 |
Robin_Watts | http://www.phrases.org.uk/meanings/bale-out-bail-out.html | 15:37.28 |
| In other parts of the English-speaking world, should you decide to record your heroic jump from an aircraft, you'd be advised to write it as 'bale out'. | 15:37.38 |
kens | will be sticking with bail, and colour. | 15:39.00 |
Robin_Watts | "other parts of the English speaking world" in the above context is "Non-US" | 15:39.34 |
| So it's bale and colour :) | 15:39.44 |
kens | suggests checking dictionary rather then online source | 15:40.56 |
henrys | Robin_Watts:are there many more msvc warnings in pcl and other areas to be fixed? | 15:41.25 |
Robin_Watts | henrys: No. | 15:41.46 |
| Well, there are loads in gdevcups. | 15:42.01 |
| There are lots of 'const casting' ones. | 15:42.07 |
kens | Don't do gdevcups! <shudder> | 15:42.24 |
Robin_Watts | and there are a few 'unary negation applied to unsigned type - result is still unsigned' ones. | 15:42.39 |
| But those are all then used with &, so the warning seems spurious and it's not trivial to fix. | 15:43.10 |
| so msvc is now as warning free as it's going to get, I think. | 15:43.39 |
| henrys: Would you rather I had done them some other way? | 15:44.15 |
henrys | We had a big argument about this amonst the staff years ago. | 15:44.38 |
Robin_Watts | The opposing sides of the argument being? | 15:45.14 |
henrys | making casts explicit is not a needed correction | 15:46.29 |
Robin_Watts | Oh, I'm of the opinion that explicit casting is always better. | 15:47.00 |
kens | True, but it shows its a deliberate cast, and eliminates the warning | 15:47.08 |
Robin_Watts | 1) Getting rid of a warning is always a plus, 2) it shows you've thought about the types involved, not incorrectly assumed you were dealing with ints rather than floats etc. | 15:47.54 |
| It's doubly important when reviewing code for bugs; it's easy to get sucked into believing that the author knew what he was doing and assuming types match up. | 15:49.07 |
| kens: I wasn't planning to attempt devcups (although, actually, it might not be a bad idea) | 15:49.45 |
| I was vaguely tempted by adding /wd4090 to the compiler flags (to silence the const warnings), but that feels a bit like giving up. | 15:50.15 |
henrys | you haven't thought about the types involved you've forced them to match by casting them in more or less a mechanical blind process. If the types should match the underlying data structures should change in many cases. | 15:50.36 |
Robin_Watts | henrys: I thought about the types a bit when fixing them. But I'd like to think that the original author had thought about the types when writing it - putting the casts in at that point would have been a note in the code that they had been thought about. | 15:53.38 |
| But if there *are* problems with those casts, and someone has to go hunting for a bug there now, I claim it will now be easier to spot such problems with the explicit casts being present. | 15:54.37 |
henrys | disagree - explicit casts just hide problems that would be found by static analysis tools or compilers. Anyway I am not going to continue the debate unless the code is going to riddled with explicit casts. It sounds like there aren't many problems left. | 16:04.46 |
| Note these problems don't show up in gcc -Wall so chrisl and I can't keep up with them regularly. | 16:08.04 |
Robin_Watts | Yes, I appreciate that. | 16:09.01 |
henrys | I fell somewhere in the middle on the debate leaning toward no explicit casts Peter was very much against the change and Igor wanted to do it. I think many places in gs Igor in fact did change, I don't recall other views on the staff. | 16:11.27 |
Robin_Watts | Ah. So I'm the new Igor. | 16:12.07 |
kens | Mk II | 16:12.37 |
| New and improved, new and improved | 16:12.45 |
henrys | well as I recall the msvc users were for and others against - I think if you have to look at the warnings you'll build up a case and if you don't you'll build a different case. | 16:13.30 |
kens | We do insist on folks fixing the gcc warnings. | 16:14.17 |
| Many of which don't show up in VS | 16:14.30 |
| I know they show up in the regression build, but it often means the VS users going back and fixing them afterwards | 16:14.50 |
Robin_Watts | Not fixing the warnings that occur in MSVC but fixing the ones that occur in gcc does seem like a double standard. | 16:15.15 |
kens | Well, except that the gcc users don't see the VS warnings | 16:15.40 |
Robin_Watts | kens: A neat trick: clusterpush with filter=WARNINGS and it'll do a build (but no tests), so you can check for new warnings very quickly. | 16:15.47 |
kens | Whereas the VS users do, albeit after the fact | 16:15.52 |
Robin_Watts | Yes, we need a windows cluster machine :) | 16:16.17 |
kens | Not disagreeing. | 16:16.26 |
| Not volunteering either | 16:16.34 |
Robin_Watts | Though... I'd expect a windows cluster machine to give differences from a unix one. | 16:16.50 |
henrys | I very much want that too, I could try setting up virtual box to do it but it would be very slow. | 16:17.03 |
kens | Even if all it did was a build that would be a plus | 16:17.22 |
Robin_Watts | yeah. | 16:17.30 |
| I'd be up for helping to set it up. | 16:17.59 |
henrys | Robin_Watts:why don't you just make your windows machine a node? | 16:18.26 |
| I am able to use the mac pro and do frequently. | 16:19.14 |
| the machine doesn't even need to be left on all the time. A run once in a while would be useful. | 16:21.17 |
Robin_Watts | IF people can cope with it disappearing from the cluster every now and then, I could try it. | 16:21.57 |
henrys | the big issue is how many platform specific differences are there? | 16:22.47 |
Robin_Watts | I have windows XP set up on VMware fusion on the macbook. and that sits there largely unused. That may be a better option. | 16:23.04 |
henrys | but that is probably something we want to know. | 16:23.05 |
kens | rendering diffs ? | 16:23.07 |
henrys | well there might be math library difference for example. | 16:23.43 |
kens | Bound to be really. | 16:23.53 |
henrys | which lead to minor diffs. | 16:23.55 |
Robin_Watts | henrys: bitmap_raster is set to a multiple of 8 on 64bit linux builds. | 16:24.00 |
| and 4 on 64bit windows. | 16:24.08 |
| hence the band height calculations are different. | 16:24.32 |
| different height bands => rendering differences. | 16:24.41 |
henrys | don't we want to use 8? | 16:25.46 |
| on windows 64? | 16:26.16 |
Robin_Watts | sizeof(long) == 4 on windows 64, and that's the determining factor, if my (dodgy) memory serves. | 16:26.44 |
| (Ray fixed various things in the clist last week w.r.t. 64bit stuff, but I don't think this was part of it) | 16:27.58 |
henrys | hmm bitmap raster should be a multiple of the alignment one might guess. | 16:28.08 |
Robin_Watts | henrys: indeed, but I *think* it comes from sizeof(long). (I reserve the right to be wrong of course) | 16:28.59 |
henrys | reading gxbitmap.h | 16:31.05 |
| yup ARCH_ALIGN_LONG_MOD - there must be a better way to do this. | 16:35.02 |
| we can have separate msvc checksums also, but that opens a can of worms. | 16:38.26 |
kens | <sigh> Spent the day setting up pdfwrite so that it doesn't execute the Photoshop duotone tint transform. | 16:39.55 |
| Now I get black output :-( | 16:39.59 |
Robin_Watts | I believe if we get the bitmap_rasters the same, we *should* get the renderings right. | 16:40.04 |
kens | Good grief: | 16:43.49 |
| http://www.firebox.com/product/4839/OVO-4-Home-Flight-Simulator?aff=512&awc=550_1321276571_3f20a1f129bc4cad31547de7092693bf%C3%82%C2%A4cy_conversion=1 | 16:43.49 |
Robin_Watts | "That's why we're stocking the..." | 16:44.22 |
| I don't believe they carry stock. | 16:44.37 |
kens | Especially since it takes 3 moths to be custom built ;-) | 16:44.53 |
Robin_Watts | http://www.firebox.com/product/4791/Retro-Invader-Couch?via=top10 | 16:46.51 |
| I do like that. | 16:46.53 |
kens | Back is too low | 16:47.07 |
tor8 | henrys: I need to go very soon. did you have anything for me for the meeting? | 16:48.19 |
henrys | tor8:how is the schedule? | 16:48.49 |
| ios? | 16:48.59 |
tor8 | henrys: on track. search input ui done, search backend done, just need to show the results | 16:49.12 |
henrys | okay great | 16:49.24 |
kens | I wonder why Miranda sometiems crashes when auto-completing Robin_Watts | 16:49.39 |
Haris | Hello people, guys, folks, everyone, all | 16:49.45 |
Robin_Watts | Haris: Hi | 16:50.14 |
henrys | tor8:we normally start in 10 minutes right? It's fine to miss but I'm wondering if there is some timezone confusion going on. | 16:50.15 |
tor8 | henrys: no, I just realized I have to be somewhere else tonight | 16:50.58 |
| timezones have synced back up :) | 16:51.08 |
henrys | okay | 16:51.21 |
| kens:so is this the new cool shopping place in the uk? I see a treadmill for dogs, interesting. | 16:53.22 |
Haris | I'm testing by running ps2pdf on shell, which executes ghostscript, which is returning an error. Need insight into this error. Please advise -> http://pastie.org/2867939 | 16:53.55 |
kens | firebox has loads of weird stuff henrys | 16:54.27 |
chrisl | Haris: I doubt the log file is Postscript, more likely plain ASCII | 16:54.59 |
kens | Haris, you executed the instruction '****' which is not defined. Hence 'undefined in ****' | 16:55.06 |
Robin_Watts | firebox = online gift shopping for geeks, pretty much. | 16:55.10 |
Haris | It was meant to be a test | 16:55.48 |
| kens: What did I miss there ? | 16:55.59 |
kens | Haris I have no idea, I haven't seen your input file | 16:56.12 |
Robin_Watts | Haris: ps2pdf takes postscript in, and spits pdf out. | 16:56.22 |
Haris | ah | 16:56.27 |
kens | Robin_Watts : Also PDF in | 16:56.32 |
Haris | let me search a sample ps file | 16:56.33 |
Robin_Watts | If you feed in something that isn't postscript, it'll barf on it (or pdf, as kens says) | 16:56.52 |
Haris | got it | 16:57.05 |
Robin_Watts | fetches caffiene. | 16:57.49 |
Haris | well, I gave it a .ps file and it created the .pdf just fine. no errors | 16:58.29 |
kens | The clue is in the name ;-) | 16:58.48 |
| Although its slightly misleading because it handles PDF as well | 16:58.59 |
| meeting ? | 17:00.20 |
henrys | yes | 17:00.26 |
mvrhel | good morning | 17:00.30 |
Robin_Watts | morning | 17:00.36 |
kens | Morning mvrhel | 17:00.38 |
henrys | hi mvrhel | 17:00.39 |
ray_laptop | I'm doing a detailed response to Paul Goldberg since it relates to bug 692618 and their use of tiffsep and more generally, separations | 17:01.10 |
henrys | alexcher posted a bug about a ps cet regression. | 17:01.12 |
| alexcher do we have a bisect for that? | 17:01.34 |
| when did it start? | 17:01.38 |
| ray_laptop:okay | 17:01.54 |
| I guess marcosw_ won't be around for this one. | 17:02.02 |
kens | Its 6pm in Germany | 17:02.28 |
alexcher | henrys: I don't know when did it start. I saw SEGV's in the regression for a while. | 17:02.28 |
henrys | mvrhel, Robin_Watts:so planar is getting close? | 17:02.29 |
mvrhel | henrys: I think so | 17:02.36 |
| we are ready to do some performance testing | 17:02.48 |
Robin_Watts | henrys: It's all committed. Marcos is doing a test for us. | 17:02.49 |
henrys | alexcher:assign it to marcos for history check. | 17:03.03 |
| Robin_Watts, mvrhel: be interesting to see the final planar vs. chunky results. | 17:03.38 |
Robin_Watts | performance or diffs? | 17:03.54 |
henrys | performance | 17:04.09 |
alexcher | henrys: A patch is attached to the bug. The bug is. probably, related to multiple threads. | 17:04.10 |
Robin_Watts | The last set of performance figures I saw suggested that on average we were about the same planar and chunky. | 17:04.32 |
| but the variance was large. | 17:04.42 |
henrys | alexcher: I saw the patch that is hardly satisfying if we want to understand the underlying issue. | 17:04.59 |
| Robin_Watts:I thought we had some hand waving reason that halftoning would be faster with planar but ... | 17:06.23 |
mvrhel | henrys: so I am going to start the process of removing wts | 17:06.23 |
| I am hoping that images with planar will be faster | 17:06.47 |
henrys | mvrhel:great | 17:06.48 |
Robin_Watts | henrys: image based halftoning should be much faster with mvrhels stuff. | 17:06.57 |
mvrhel | if not, we will need to see why] | 17:07.00 |
Robin_Watts | Some things already are much faster. | 17:07.08 |
| But some things are much slower. | 17:07.19 |
| What you win on the swings, you lose on the roundabouts. | 17:07.33 |
mvrhel | hehe. what things are slower? | 17:07.42 |
| funky rops | 17:08.14 |
Robin_Watts | I have Rick Perrys disease. | 17:08.20 |
| rops, chunky images, and ... | 17:08.36 |
| and general overhead for small operations. | 17:09.00 |
mvrhel | what do you mean by chunky images? | 17:09.03 |
Robin_Watts | images where the data is supplied in chunky form (and it isn't being halftoned) | 17:09.30 |
mvrhel | ok | 17:09.38 |
| so for pdf/ps to halftone we really should be looking good | 17:10.01 |
Robin_Watts | yes. | 17:10.14 |
mvrhel | if the pcl has some rop action going on, that could slow us down | 17:10.16 |
henrys | we hope that's rare. | 17:10.33 |
mvrhel | right | 17:10.37 |
Robin_Watts | rops seem ubiquitous in pcl though :( | 17:10.46 |
henrys | and I think that a reasonable assumption. | 17:10.49 |
| yes but many are optimized away. | 17:11.03 |
mvrhel | in most standard docs, its probably not too common I would think | 17:11.04 |
kens | I don't see it often causing a probelm in pdfwrite | 17:11.16 |
Robin_Watts | fair enough. | 17:11.36 |
henrys | kens anything for the meeting? | 17:12.10 |
mvrhel | henrys: wat is going on with bug 692674. I wonder if someone should take a look at that | 17:12.20 |
kens | Nothign from me henrys | 17:12.31 |
Robin_Watts | mvrhel: It's vanished! | 17:12.49 |
alexcher | kens: I've posted bug 592682 about strange code in pdfwrite. Please take a look. | 17:13.00 |
Robin_Watts | mvrhel: Ignore me, I mistyped. | 17:13.08 |
henrys | mvrhel:assigned to marcos | 17:13.15 |
| I'll send him mail. | 17:13.32 |
kens | alexcher I saw it, I'll take a closer look when I have time | 17:13.32 |
alexcher | kens: I've posted bug 692682 | 17:13.33 |
henrys | he's probably not seeing it. | 17:13.43 |
| kens:you want to just assign it to yourself? | 17:14.15 |
| or should I send email to marcos? | 17:14.24 |
kens | Isn't it assigned to me already ? | 17:14.31 |
| It seems to be. Wait, are we talking about the same bug ? | 17:15.05 |
henrys | mvrhel said 692674? | 17:15.44 |
kens | I'm talking about 692682 with alex | 17:15.56 |
mvrhel | well that is a performance issue between 8.71 and 9.04 | 17:16.03 |
| from customer 130 | 17:16.08 |
henrys | I'll send mail to marcos right now. | 17:16.20 |
| chrisl:any freetype news? | 17:16.31 |
kens | I'm inclined to say that 'yes its slower, it works better, if you don't want features and bug fixes, use the old one' | 17:17.05 |
chrisl | henrys: the Type 1 dict API is in (and is in today's freetype release!), so I'm good to go when my desk is clear again...... | 17:17.06 |
henrys | well that is good. | 17:17.44 |
chrisl | I just need to do the Ghostscript end of things, now! | 17:17.57 |
mvrhel | a minor detail | 17:18.08 |
chrisl | ;-) | 17:18.14 |
henrys | alexcher:are you working on openjpeg? | 17:18.31 |
chrisl | I also want to find time to implement the same API for CFF and CID fonts in freetype - but that's not high priority | 17:18.48 |
henrys | chrisl:I have to catch up with shelly email too, I keep forgetting. | 17:19.03 |
alexcher | henrys: yes, I'm about to commit a working version. | 17:19.50 |
henrys | ray_laptop:meeting stuff? | 17:19.56 |
chrisl | henrys: I was going to ask about that - do you want me to cast an eye over the jbig2dec changes he sent, and commit them - or are you planning to do it? | 17:19.58 |
mvrhel | so for bug 692508, we need to have freetype for XPS is that correct? | 17:20.24 |
ray_laptop | nothing for the meeting. I already sent a 'sorry' email to chrisl about my misreading of psromfs.mak | 17:20.52 |
henrys | chrisl:yes but I thought shelly could commit after a review? | 17:20.57 |
ray_laptop | chrisl: and I will apologize here, too | 17:21.04 |
mvrhel | I think that will be the last XPS bug. I need to double check but I think I can close 692512 and 692513 which Robin helped on | 17:21.17 |
chrisl | ray_laptop: no problem. I've tested adding the "*" on Linux and Windows, and both are happy with it, so I'm happy to commit the fix | 17:21.44 |
| henrys: I can't remember if Shelly has commit rights | 17:22.02 |
ray_laptop | chrisl: OK, thanks | 17:22.05 |
henrys | chrisl, ray_laptop:up to you guys but maybe a discussion before going back to the customer and saying somebody is wrong is best. | 17:22.07 |
ray_laptop | henrys: I agree and will do that in the future | 17:22.57 |
henrys | then the person that made the mistake (if there was one) can go back and say by the way my colleague pointed out... I think that sounds better. | 17:23.18 |
chrisl | Sure, I tend to do that - but mainly because I'm still learning stuff about Ghostscript! | 17:24.06 |
henrys | well let's adjourn unless there's other stuff. | 17:24.08 |
kens | henrys do you want me to do somethign with #692674 ? | 17:24.50 |
henrys | kens:no I sent mail to marcos | 17:25.05 |
kens | OK thanks | 17:25.20 |
henrys | mvrhel:yes xps needs freetype but if chrisl does pcl xps will be easy. | 17:25.40 |
mvrhel | ok. great | 17:25.52 |
chrisl | henrys: sorry, I didn't follow your reply: do you want me to review Shelly's changes for jbig2dec | 17:25.57 |
henrys | no I'll review them, I think you have plenty to worry about. | 17:26.27 |
| ;-) | 17:26.33 |
chrisl | Okay, thanks! | 17:26.38 |
kens | Goodnight then everyone | 17:28.07 |
ray_laptop | g'nite kens | 17:28.20 |
henrys | bye kens | 17:28.30 |
| Robin_Watts:so are you going to try setting up a node? | 17:32.11 |
Robin_Watts | I'll talk to marcosw_ about it tomorrow, yes. | 17:32.23 |
henrys | okay good | 17:33.03 |
ray_laptop | I have to go now, may be back online in a while... | 17:35.51 |
mvrhel | brb | 19:00.14 |
Robin_Watts | henrys: w.r.t bug 692677 | 19:29.43 |
| So PXL sets the page size to A4. | 19:30.17 |
| Then it 'passes through' to PCL which has the default page size (which happens to be letter). | 19:30.35 |
| Could setting the page size in PXL not also set the PCL page size ? | 19:30.56 |
| (or could it at least set it if PCL hasn't already explicitly set it) | 19:32.36 |
mvrhel | lunch time while the code is tested with WTS removed. | 20:05.48 |
henrys | Robin_Watts:right but the PCL interpreter really doesn't actually exist at that time. The change would mean always starting the PCL interpreter that is a very impactful change for a feature that is never used. | 20:38.11 |
Robin_Watts | henrys: OK, so the PCL interpreter is first started when we first meet a passthrough? | 20:39.04 |
| Can we not change that so that as well as starting it, we set the page size to match that of PXL? | 20:39.58 |
mvrhel | hmmm segvs.. | 20:48.05 |
henrys | that upside down P was written by PXL on an a4 page we have to keep that and basically give that page to pcl to be drawn on. If PCL does not detect the page is different from it's current state it won't mess with the page. | 20:48.33 |
| so yes setting up PCL PCL with the same page size would work | 20:49.24 |
| my feeling is though the whole design I came up with is wrong. I think HP just starts both interpreters and has a parsing mode to switch between the two. Passthrough is a horrible afterthought - a royal pain. | 20:50.38 |
kens | passthrough is truly an awful kludge | 20:51.15 |
mvrhel | looks like I hosed up the clist | 20:52.57 |
| with my wts extraction | 20:53.03 |
henrys | Robin_Watts:thanks for talking about it. I do think the right thing to do at this point is to start PCL with the right page size. All of this only applies to "snippet mode" there is a "regular" passthrough mode where pcl is supposed to reset the page. | 21:01.31 |
mvrhel | bboab | 21:14.11 |
| bbiab | 21:14.13 |
| Forward 1 day (to 2011/11/16)>>> | |