| <<<Back 1 day (to 2012/02/28) | 2012/02/29 |
Robin_Watts | henrys: Yes. Turn on hpgl mode when pcl starts up, and we should hopefully get what we want. | 00:05.03 |
| mvrhel_laptop: Is your push still messing around? | 00:07.30 |
| Pushing a clean one would seem a good test if this fails. | 00:11.10 |
Robin_Watts | tries a PCL clusterpush with hpgl mode on. | 00:22.57 |
| marcosw: Can you remove marcos.tar from cluster/user please ? | 00:23.28 |
| It has an unwanted effect on the dashboard :) | 00:23.40 |
marcosw | wil do | 00:23.45 |
| btw, I'm watching mvrhel_laptop's cluster job (having just restarted it) and it seems to be running okya. | 00:24.13 |
Robin_Watts | The dashboard has links to a script that is supposed to bring hosts up/down. | 00:24.27 |
| but it's not working, as the script hasn't got permission to write to the cluster dir. | 00:24.50 |
| I think possibly the script used to run as setuid root, but that's somehow been lost. | 00:25.13 |
| Did you change that ? | 00:25.19 |
marcosw | no. | 00:25.54 |
Robin_Watts | IS feet in your office? | 00:26.09 |
marcosw | yes, but I'm supposed to deliver to Miles tomorrow. | 00:26.25 |
Robin_Watts | Is it worth memtest86ing it overnight before you do ? | 00:26.39 |
marcosw | It's possible that the setuid bit was lost when I restored the cluster files from backup, after you accidentally did a git reset. | 00:26.56 |
Robin_Watts | This isn't in the cluster dir. It's in /usr/lib/cgi-bin | 00:27.20 |
marcosw | I've been running mprime on feet for the last several hours without any issue. I was going to de-overclock it and put it back on the cluster. | 00:27.41 |
| In that case I definitely didn't change the setuid flag. | 00:27.56 |
marcosw | thinks that a watched cluster never fails | 00:36.19 |
Robin_Watts | Bah. perl won't run setuid. I'll do some simple shellscripts tomorrow. | 00:39.34 |
marcosw | I've found the likely source of mvrhel_laptop's cluster problems. The log files are huge, presumably due to debugging output. This is going to cause problems when they are transferred to casper. | 00:43.45 |
| e.g the log file on my i7 is 7.5 gigs :-) | 00:45.42 |
| When I looked at the logs it appeared that the transfers failed because casper went down, but I'm now thinking it was just a timeout because of the large size. How to fix this?â¦. | 00:47.35 |
mvrhel_laptop | marcosw: wow | 00:48.27 |
| marcosw: if the log file has some big debug output, then apparently the one line that I wanted to remove from the code is clearly needed. I should be able to look at the log and see some of the files for which this is an issue | 00:49.34 |
| brb | 00:49.37 |
marcosw | This line appears to be repeated a lot: " **** File has unbalanced q/Q operators (too many Q's) ****" | 00:49.54 |
| In particular the unbalanced q/Q time is in the log file for the i7 118,401,398 times | 00:54.38 |
henrys | Robin_Watts:what I was trying to say before is I did turn on the new mode and the very simple path fails with a stray line running up the page. | 01:05.01 |
marcosw | the cluster now removes duplicate lines from the log files, so the many repeated "*** File has unbalanced q/Q operators (too many Q's) ****" lines will get folded into one line. | 01:26.56 |
mvrhel_laptop | ok. so I should push again? | 02:54.42 |
| I don't understand how my one line could cause the q/Q output | 02:54.56 |
marcosw | mvrhel_laptop: I'm not sure either, but I've no fixed the cluster so that really large log files are truncated to the first 10,000 lines and then duplicate lines are removed. So you may clusterpush at your convenience. | 03:59.49 |
mvrhel_laptop | thanks marcosw | 04:11.41 |
marcosw | mvrhel_laptop: the problem with the large number of warnings is likely not your fault. It appear to be a memory issue. | 04:36.18 |
mvrhel_laptop | ok | 04:36.28 |
marcosw | the disturbing thing is that running the exact same command on the same cluster node doesn't produce the large number of warnings, it immediately seg faults. | 04:48.29 |
henrys | marcosw:same result on every machine? | 04:52.28 |
marcosw | no. but running on the exact same machine produces different results. | 04:53.06 |
| i.e. on peeves the latest clusterpush from mvrhel_laptop produces thousands of warnings for the file tests_private__comparefiles__Bug691783.pdf.plank.300.1 when run by the cluster | 04:53.59 |
| but if I run the exact same command from a shell I get a seg fault. | 04:54.10 |
| this appears to be a recent development. | 04:54.34 |
henrys | I did a diff of the mvrhel_laptop tree on macpro vs. head and didn't see anything terribly suspicious. | 04:58.43 |
marcosw | I think it's a memory moving around thing, running the same code on different machines produces different results. | 04:59.25 |
henrys | well he suspected a gc problem with the change. | 05:00.49 |
marcosw | i.e. this command: ./gs/bin/gs -sOutputFile=/dev/null -dMaxBitmap=10000 -sDEVICE=ppmraw -r300 -sDEFAULTPAPERSIZE=letter -dNOPAUSE -dBATCH -K1000000 -dClusterJob -dJOBSERVER ./tests_private/comparefiles/Bug692368.pdf | 05:01.16 |
| seg faults on my amd64 box but on peeves generates infinite (?) number of warnings: **** File has unbalanced q/Q operators (too many Q's) **** | 05:01.40 |
| this is with mvrhel_laptop's source. when I run it with master it's fine. | 05:02.48 |
henrys | a debug and -Z? might tell us what's going on. | 05:03.34 |
marcosw | yeah, building a debug version now. | 05:04.09 |
| #0 0x0000000000468229 in gs_md5_process (pms=0x7fffffffbaa0, data=0x0) at ./debugobj/md5.c:209 | 05:07.28 |
| #1 0x00000000004690d6 in gs_md5_append (pms=0x7fffffffbaa0, data=0x0, nbytes=564) at ./debugobj/md5.c:355 | 05:07.28 |
| #2 0x00000000006227cf in gsicc_get_buff_hash (data=0x0, hash=0x7fffffffbc78, num_bytes=564) at ./base/gsicc_cache.c:237 | 05:07.28 |
| #3 0x0000000000622770 in gsicc_get_icc_buff_hash (buffer=0x0, hash=0x7fffffffbc78, buff_size=564) at ./base/gsicc_cache.c:224 | 05:07.28 |
| #4 0x0000000000622a0d in gsicc_get_cspace_hash (icc_manager=0x1819020, dev=0x1ab9ad8, cmm_icc_profile_data=0x19b8110, hash=0x7fffffffbc78) at ./base/gsicc_cache.c:301 | 05:07.29 |
| #5 0x00000000006228f8 in gsicc_compute_linkhash (icc_manager=0x1819020, dev=0x1ab9ad8, input_profile=0x19b8110, output_profile=0x1b96f30, rendering_params=0x7fffffffbe60, | 05:07.29 |
| hash=0x7fffffffbc70) at ./base/gsicc_cache.c:268 | 05:07.29 |
| #6 0x00000000006232c0 in gsicc_get_link_profile (pis=0x17fd048, dev=0x1ab9ad8, gs_input_profile=0x19b8110, gs_output_profile=0x1b96f30, rendering_params=0x7fffffffbe60, memory=0x17da928, | 05:07.30 |
| devicegraytok=1) at ./base/gsicc_cache.c:604 | 05:07.30 |
| did mvrhel_laptop's latest change involve the icc code? | 05:07.56 |
henrys | yes | 05:08.01 |
marcosw | I'll send mvrhel_laptop an email with the gdb stack trace | 05:11.05 |
mvrhel_laptop | hmm | 05:11.28 |
henrys | I guess this would be complicated but it would be nice if these cluster user trees were git trees so we could just do git diff. The diff -r is rather painful. | 05:12.09 |
| but I guess it is good enough. | 05:12.20 |
mvrhel_laptop | henrys: so. now we also know why I did not allow changing of the profiles after they are set | 05:12.53 |
henrys | mvrhel_laptop you just commented out the *manager_default_profile != NULL business right? | 05:13.03 |
| mvrhel_laptop:up to you I can probably work around it is in the other languages, but these seems rather quirky to me. | 05:14.02 |
| s/these/this | 05:14.14 |
mvrhel_laptop | well. I agree. I was not at the time able to figure out what was going on. coming back now, I may be able to find out what the heck the issue is | 05:14.43 |
| my frustration at the time resulted in the the length comment in the code | 05:15.00 |
| lengthy | 05:15.05 |
| let me do this. Let me try to isolate and understand the issue tomorrow | 05:15.25 |
| before you have to do a limbo move to get around this | 05:15.38 |
henrys | okay sounds good, thanks | 05:15.53 |
mvrhel_laptop | better to fix the limitation here than to have program around it all over the place | 05:16.05 |
henrys | it's unfortunate you don't see it on windows - maybe with -Z? | 05:16.36 |
mvrhel_laptop | yes. I will do that tomorrow. | 05:16.50 |
henrys | okay I'm out for the night too, see ya tomorrow. | 05:17.16 |
mvrhel_laptop | ttyl | 05:17.21 |
henrys | thanks marcosw. | 05:17.25 |
mvrhel_laptop | ok. I have one of the exploding cases here in windoze.. | 05:53.30 |
| and sure enough it is resetting the profiles in the manager later on in the code to the same thing and freeing them | 05:55.48 |
| hmm. ok. well I am hopeful that I can get this figured out tomorrow | 05:56.14 |
| and sure enough this is occurring following the vmreclaim | 05:58.32 |
| oh now I remember. | 05:58.35 |
| the damn user param stings are not being maintained | 05:58.52 |
| ray_laptop and I worked on this some | 05:59.01 |
| there is a bug someplace else that was causing this | 05:59.35 |
| will chat with him tomorrow about it | 05:59.47 |
| good night for now | 05:59.50 |
CoolCoder | hi. anyone here | 07:22.33 |
| When i run ghostscript for converting pdf to jpg (please check this log http://pastebin.com/tJfTCXA3 ) after page "Loading StandardSymL font ..." this works in terminal. But when i do with php after page 8, Its not rendering and hanging. Any can knows the reason? | 07:25.50 |
kens | Hmmm.... | 08:05.47 |
| For the logs, Robin_Watts your path change made thousands of changes to PostScript output, is that expected ? | 08:06.13 |
Robin_Watts | no. :( | 10:27.02 |
kens | I'll leave it with you Robin_Watts | 10:27.43 |
| TIme for me to go | 10:27.47 |
svilayphiou | hi, a client wants to put a pdf on its website, but I made on Indesign back then, and I don't have it anymore. I need to remove the cropmarks and bleeds from the prepress PDF. What option should I use? | 10:38.50 |
chrisl | CoolCoder: I can help with your gs from php problem, but have a look on stackoverflow, there have been a number of discussions on the subject. I think the most common problem was using the wrong php call to run the gs exectuable - you need to use the one that carries the shell environment with it, and also make sure that the php user's environment on the system has everything correctly setup... | 11:37.54 |
| ...for GS. | 11:37.56 |
CoolCoder | chrisl: I was getting the result when try to convert pdf to jpeg. http://pastebin.com/kKAeqzMu this is the command i was using. the issue is for only some files | 11:39.51 |
| like the files have "Loading StandardSymL... " see this output on terminal http://pastebin.com/tJfTCXA3. after page 8 its going to an infinite loop on php | 11:40.49 |
| I have searched stackoverflow for this issue. But there is no result. Actually this issue is not regarding the PDF->JPEG conversion. I have successfully done it for some files. But some files this is an issue of infinite loop | 11:42.49 |
chrisl | CoolCoder: sorry, I had a typo in my reply - it was supposed to be "I can't help...... but look on stackoverflow...." | 11:43.54 |
CoolCoder | ok. :) | 11:44.35 |
chrisl | CoolCoder: The thing I was thinking of, I think, was using shellexec instead of exec. | 11:45.39 |
CoolCoder | ok. let me try | 11:46.24 |
| same thing chrisl: it dies for some files. The issue is the same. when gs loads "StandardSymL" for terminal it runs, but on PHP it hangs. | 11:53.07 |
chrisl | CoolCoder: well, if it works normally, then it's rather hard for us to help! Like I say, the similar sounding issues I've seen discussed have usually boiled down to differences in the environment between running in a normal terminal, and running in the php environment | 11:55.17 |
CoolCoder | same thing chrisl: it dies for some files. The issue is the same. when gs loads "StandardSymL" for terminal it runs, but on PHP it hangs. | 12:04.45 |
chrisl | CoolCoder: well, if it works normally, then it's rather hard for us to help! Like I say, the similar sounding issues I've seen discussed have usually boiled down to differences in the environment between running in a normal terminal, and running in the php environment | 12:09.08 |
CoolCoder | ok. let me check. thanks | 12:14.33 |
chrisl | CoolCoder: another thought (I don't know how php's callout mechanism works) - if Ghostscript is producing a lot of backchannel output (like font message) it might be the stdout buffer filling up, and the process blocking until it empties. | 12:16.28 |
CoolCoder | can we disable this? | 12:17.20 |
chrisl | add -q or -dQUIET | 12:17.48 |
CoolCoder | ok | 12:21.14 |
| chrisl: http://pastebin.com/RxLVvRiK | 12:23.59 |
chrisl | No '-'? | 12:24.43 |
CoolCoder | no | 12:25.29 |
chrisl | gs -dQUIET -I../Resource/Init -sDEVICE=pnmraw -r300 -o stuff%03d.pnm <pdf file> | 12:26.57 |
| Works for me | 12:27.10 |
CoolCoder | it runs for me. But the issue is there still. Its some internal issue. I think some permission issue with PHP while trying to load "StandardSymL". I will have to find some other way to do the task | 12:32.15 |
| will ps code works? | 12:32.23 |
chrisl | CoolCoder: the GS PDF interpreter is written in Postscript, so there is little "real" difference to GS whether you're running a PDF job or a PS one. | 13:02.58 |
CoolCoder | ok | 13:03.49 |
chrisl | CoolCoder: can you su to the php user, and try running Ghostscript from there? | 13:04.31 |
| CoolCoder: or are you running on Windows? | 13:05.35 |
CoolCoder | ok. let me check how to do it | 13:05.35 |
| linux | 13:05.38 |
| ubuntu | 13:05.58 |
chrisl | Okay, so su-ing to the php user, and testing GS might be an option | 13:06.25 |
CoolCoder | Let me try. this might be a permission issue it seems. Because apache might be trying to load StandardSymL | 13:07.30 |
paulgardiner | tor8: Hi. I was wondering where we are with the Android app. Henry said you might be discussing what needs doing at the meeting last week. I guess the fact that the hardware acceleration changes worked for some devices but not others, makes it less clear what's best to do. I've made one further small change, to do with invocation from the TP's File Manager, but other than that I've been... | 14:42.22 |
| ...holding off until we know more what's required. | 14:42.22 |
tor8 | paulgardiner: I think it's in pretty good shape as it is. obviously I'd like to see HW acceleration work on my samsung as well, but even without it's fast enough now. | 14:43.21 |
| I think moving forward we should look into the search UI behaviour and the link highlighting toggle | 14:44.00 |
paulgardiner | Ok good. Do we need to check with anyone else before I work on that? | 14:45.42 |
tor8 | I don't think so. | 14:47.15 |
Robin_Watts | I would have thought that that would be covered under what henrys said last week. | 14:47.17 |
tor8 | Robin_Watts: does paulgardiner have an account on casper, so that he can host his git repo there? | 14:48.04 |
paulgardiner | Henry gave me the go ahead for two days (although for a different task). I've since then used up one day. | 14:48.10 |
Robin_Watts | I don't think paulgardiner has an account on casper. | 14:48.35 |
| paulgardiner: Did Miles get you to sign an NDA ? | 14:48.46 |
paulgardiner | Yep | 14:49.04 |
Robin_Watts | Ok. So there shouldn't be a problem in giving you a casper account then. | 14:49.21 |
| Henrys will be here within an hour or so, so we can double check with him. | 14:49.46 |
paulgardiner | Great. Thanks. | 14:50.03 |
| On the subject of the search UI, what actions do we think should remove the current search highlights? Should they disappear the moment one types? | 14:52.21 |
kens | test with Acrobat ? | 14:53.00 |
sebras | paulgardiner: when the text in the search field changes. | 14:53.35 |
| paulgradiner: I think that's when the ios app clears search highlights. | 14:53.54 |
paulgardiner | Ok, that sounds sensible. | 14:54.11 |
kens | Acrobat maintains highlights until you press the 'search' button | 14:54.30 |
paulgardiner | ... which was the alternative I was wonderined about. | 14:54.56 |
kens | Not saying its right, just that's what it does :-) | 14:55.15 |
sebras | I'm unsure which one is better. :) | 14:55.18 |
paulgardiner | I might let laziness decide it then (least code) :-) | 14:56.15 |
| I guess the main thing to be avoided is for a failed search for a different search string to leave the old highlights in place. | 14:57.03 |
sebras | yes, I agree. this is what caught my eye. | 14:57.41 |
kens | Yes, that's important | 14:58.04 |
paulgardiner | Ok, thanks. | 15:02.03 |
Robin_Watts | Personally, I'd like the highlights to disappear when an actual search starts. | 15:09.48 |
| if I search for something, that might point me to something else to search for. When I bring up the search box again, I'd like the highlights to stay until I actually *start* the next search (hit search or hit return) | 15:10.53 |
| It would be bad if bringing up the search box cleared the old results. | 15:11.14 |
kens | That's as per Acrobat | 15:11.16 |
paulgardiner | Yeah, that does sound best to me | 15:19.19 |
marcosw | Those of you who suggested a memory issue with feet were correct; running memtest overnight showed a problem. I've since removed half the ram and am rerunning the test to isolate the issue. | 15:24.27 |
Robin_Watts | It failed even when not overclocked? | 15:28.39 |
henrys | yes paulgardiner should be fine for casper. | 16:17.02 |
Robin_Watts | chrisl: Can you sort that out ? | 16:17.29 |
| (or marcosw?) | 16:17.40 |
| I can never remember the correct runes to add a user. | 16:17.49 |
chrisl | Robin_Watts: I can do it - we'll need an ssh key | 16:22.36 |
paulgardiner | chrisl: which I should generate? | 16:23.19 |
Robin_Watts | paulgardiner: Can you make an ssh key? 2048 bit dsa one by preference. | 16:23.22 |
henrys | chrisl:can you put the latest ufst cdrom on casper (in a private place) | 16:23.24 |
Robin_Watts | I used putty to generate a key, then saved out the openssh version from there. | 16:23.47 |
| because that way I have a key that works in putty too. | 16:24.01 |
chrisl | henrys: The latest UFST I have access to (6.2) is already in svn-private | 16:24.01 |
henrys | chrisl okay great the whole works - fonts and all. | 16:24.27 |
chrisl | henrys: yes, I just dumped the lot in there! | 16:24.48 |
henrys | good | 16:24.54 |
| paulgardiner:note your private key will be exposed to artifex admins on casper. | 16:26.00 |
paulgardiner | Should be ok to use my existing one: it's only the public part I need pass to you, yes? | 16:26.15 |
Robin_Watts | paulgardiner: Better to make a completely new one. | 16:26.34 |
paulgardiner | Ok | 16:27.00 |
chrisl | paulgardiner: yes, only the public key required | 16:27.34 |
| paulgardiner: if you can e-mail it to chris.liddell (at) artifex.com | 16:27.56 |
henrys | I guess it would be okay for him to just have a public key on casper. | 16:28.21 |
paulgardiner | Will do. So the risk with my current one, is it could have become compromised without my knowing it? | 16:28.34 |
chrisl | henrys: I only have a public key on casper | 16:28.51 |
henrys | anyway we're a pretty trustworthy lot ;-) | 16:29.21 |
Robin_Watts | IF you ever want to be able to ssh from casper to another cluster node, then you'll need to put the private part on casper too. | 16:29.21 |
henrys | Robin_Watts | 16:29.29 |
paulgardiner | :-) | 16:29.30 |
henrys | right | 16:29.31 |
paulgardiner | Oh yes true | 16:29.47 |
henrys | I doubt paulgardiner will find the need for that | 16:30.10 |
Robin_Watts | Given that making a new key is pretty trivial, and pageant etc cope with multiple keys, it's best to go with a whole new key. | 16:30.24 |
| And you should have a pass phrase on your key too. | 16:30.49 |
| (or so Rillian told me when he was setting me up) | 16:30.58 |
chrisl | paulgardiner: oh, and let me know if you have a preference for a username | 16:32.10 |
marcosw | Robin_Watts: yes, the memory failed even when the computer was not over clocked. | 16:33.47 |
henrys | marcosw:is it a faulty simm? | 16:53.27 |
| I've been meaning to overclock my machine but never got to it. | 16:53.57 |
marcosw | I'm hoping so. I've had computers fail when two simms were installed but not when either one was individually installed. I'm testing with one simm installed, so far no errors. | 16:55.35 |
Robin_Watts | At this stage, you have to wonder whether it would not simply be more time efficient to tell them to replace the machine with a new one that you know works. | 17:00.52 |
kens | Off for a piuzaa, night all | 17:01.01 |
Robin_Watts | Hunting for the cause of such errors can take ages. | 17:01.09 |
| Could be a SIMM at fault, could be the motherboard, could even be the PSU. | 17:01.30 |
| (This is the machine you bought complete, right?) | 17:01.52 |
mvrhel | morning henrys | 17:07.24 |
henrys | howday | 17:07.41 |
| or howdy | 17:07.45 |
mvrhel | so when I got to digging into the issue it all came back to me | 17:07.54 |
henrys | uhoh | 17:08.31 |
mvrhel | that piece of code was a kludge to overcome an issue that is occurring on the user param strings during the vmreclaim | 17:08.35 |
| for some reason during a reclaim the strings get blown away | 17:08.52 |
henrys | so they are allocated in gc memory? | 17:09.34 |
mvrhel | I believe the user parameter strings are. I am going to dig into this rabbit hole further today. At the time, ray had tried to figure out what the issue was and could not so I don't know if I will have much success but I will see what I can find | 17:10.32 |
henrys | okay can you send a patch so I can reproduce it maybe I'll see something but as you said if ray_laptop worked on it already it's probably hard. | 17:11.59 |
mvrhel | henrys: ok. let me poke around at this a bit more. I think I am seeing another issue here | 17:18.49 |
henrys | Robin_Watts:you got my mail about the new path code? | 17:21.18 |
mvrhel | I have to step out for a bit. need to go to the airport to get some tickets. I had some vouchers and apparently to use them you can't do it on line but have to show them in person. | 17:21.30 |
Robin_Watts | henrys: About it having stray lines ? | 17:22.03 |
| Ken pointed out this morning that it produced lots of changes in ps and pdf, so I backed it out. | 17:22.32 |
| I think I found the stupid typo, so retesting now. | 17:22.41 |
henrys | oh okay how did I miss that. | 17:23.08 |
Robin_Watts | If this works, I'll do a pcl push with the hpgl flag on by default and look at the diffs. | 17:23.25 |
| henrys: You probably foolishly assumed that I'd test before comitting. ahem. | 17:23.55 |
henrys | I can just send you the patch to enable it and you can push then commit if you want. | 17:24.01 |
| or I'll add it, one line. | 17:24.35 |
Robin_Watts | If you want. It's a 1 char change for me to force it on for a test. | 17:24.49 |
chrisl | paulgardiner: before I change the name of the file, can you try casper again, please? | 17:29.08 |
mvrhel | henrys: so I do see an issue with my stuff and I think this is the actual cause. the sting params coming back after vmreclaim actually seem ok | 17:29.36 |
| string.. | 17:29.43 |
henrys | for the commit history it would be better to have you go ahead and add the setting to pctop.c around 403 right after gs_setaccuratecurves | 17:29.52 |
Robin_Watts | ok. | 17:30.13 |
henrys | so we're arguing over pdf vs ghostscript to somebody who isn't going to buy either. | 17:33.03 |
| s/pdf/mupdf | 17:33.15 |
| mvrhel:oh good. | 17:34.24 |
Robin_Watts | "MuPDF development work" ? | 17:34.32 |
| I didn't get the feeling he'd declared that he wasn't prepared to pay for it. | 17:34.50 |
henrys | yes I'm jumping to conclusions... | 17:35.41 |
| we probably should have a unified staff response to what is best for given situation but I doubt we could reach it. | 17:36.30 |
| mvrhel:glad you found that I hate to have that kind of stuff lurking around. | 17:37.42 |
mvrhel | henrys: I will work on this more when I get back. need to get to the united desk counter to turn in these vouchers. bought tickets for the wife and kids to go to my inlaws over spring break and to use these I have to do it in person | 17:41.27 |
henrys | no hurry. | 17:41.42 |
Robin_Watts | mvrhel: Ah, yes. They do believe in making it easy to use their vouchers, don't they... :) | 17:42.06 |
mvrhel | unreal | 17:42.13 |
henrys | united gave me 75.00 off a flight for delaying me last trip... woopee! | 17:43.42 |
| I should ask them if I can use it on Frontier so I can to the destination on time. | 17:44.12 |
| s/can/can get | 17:44.41 |
Robin_Watts | Ok, so my cluster test passed without any diffs, that's good. Now to try nobbling pcl. | 17:49.40 |
ray_laptop | Robin_Watts: oops -- I didn't realize Eric was testing 1200 dpi and that he needs -dMaxBitmap=20000000 in order to run page buffer mode with 1200 dpi :-( | 17:50.25 |
ray_laptop | eats crow | 17:50.32 |
| at least it explains why the timing was the same | 17:50.56 |
Robin_Watts | Has he actually changed the -dBandHeight lines though ? | 17:52.05 |
| I see lots in his mails about fiddling with -dMaxBitmap or not, but nothing about -dMaxBitmap. | 17:52.40 |
| nothing about -dBandHeight. | 17:52.47 |
ray_laptop | Robin_Watts: if the -dMaxBitmap is large enough to hold the page, then the BandHeight doesn't get used | 17:53.03 |
Robin_Watts | Right. | 17:53.11 |
| So if he's failing to get to use page mode, we'll still be getting the stupid number of bands per page. | 17:53.41 |
ray_laptop | Robin_Watts: I want to see how fast page mode is since that will be the best we can do with just clist tuning | 17:54.00 |
Robin_Watts | ray_laptop: Right. | 17:54.27 |
ray_laptop | Robin_Watts: actually even better than clist mode can ever be | 17:54.40 |
Robin_Watts | Yes, getting timings for page mode will clearly set a lower bound for clist based operation. | 17:55.07 |
ray_laptop | Robin_Watts: and we may be able to convince them to just let Ghostscript decide whether or not to use page mode based on trying the allocation. If the PDF is too large (also in memory), then the allocation will fail and we'll "fall back" to clist mode | 17:56.34 |
| but for the majority of files, we'll be run in page mode | 17:57.02 |
Robin_Watts | I could understand if they were reluctant to do that. | 17:58.05 |
| That's a big change to the pattern of memory use in the system, and they haven't tested for that. | 17:58.23 |
ray_laptop | Robin_Watts: why ? | 17:58.25 |
Robin_Watts | Just upping the bandheight is a smaller change. | 17:58.47 |
| but sure, let's wait and see what they say. | 17:59.00 |
ray_laptop | Robin_Watts: it's actually safer in that they won't get complexity errors due to the clist getting too big | 17:59.04 |
Robin_Watts | raising the bandheight will shrink the clist too. | 17:59.33 |
ray_laptop | Robin_Watts: slightly. And not as much as my planned method for tile storage in pseudo bands | 18:00.05 |
Robin_Watts | Not as much as your planned method, no. | 18:00.19 |
ray_laptop | Robin_Watts: have you tried to see how much 1024 vs 256 shrinks the clist ? (-Z: prints out the clist size if it's a debug build) | 18:01.40 |
henrys | paulgardiner:it is fine to run over 2 days then give me a holler okay? | 18:01.50 |
| paulgardiner:when do we start javascript - that's what I want to talk about ! I guess I'm easily excited. | 18:02.28 |
Robin_Watts | ray_laptop: No. I haven't tried anything like that. I'm disinclined to waste any more time on it until they actually do the test I've asked them to do many times in the past. | 18:03.27 |
paulgardiner | henrys: can start javascript pretty much as soon as you like. Looks really interesting, and just my sort of thing. | 18:04.26 |
henrys | paulgardiner:have we decided on an engine? licensing needs to work. | 18:04.57 |
Robin_Watts | paulgardiner likes leather, chains and whips too. | 18:05.06 |
| :) | 18:05.11 |
paulgardiner | Robin_Watts: true, but not as much as PDF javascript. | 18:05.42 |
henrys | We have to have this ready in time for me to do my tax forms. | 18:06.08 |
paulgardiner | :-) | 18:06.17 |
Robin_Watts | henrys: I thought V8 was the preferred engine (it's the one with the nicest license for our purposes) | 18:06.26 |
| but ideally, we'd want the work done in such a way that mupdf is realtively agnostic w.r.t the choice of engine. | 18:07.01 |
henrys | Robin_Watts:we did kick it around before but it's a hot community and I haven't kept up. | 18:07.07 |
ray_laptop | on their other (color printer) performance issue, I am NOT understanding what's going on. gs8.71 processes the file to 600 dpi ppmraw /dev/null in 35 seconds. (gs 9.05 does it in 27 seconds). On their simulator code it is taking 200 seconds (but that is a debug build) and the release build on their target takes > 24 minutes | 18:07.09 |
henrys | tor8 will likely want to weigh in. | 18:07.18 |
paulgardiner | Yes, I was thinking that too. Run the javascript engine via a thin interface of our own, so that potientially we could slot in a different engine. | 18:07.41 |
Robin_Watts | (i.e. we put an interface on mupdf that's not tied to any particular engine, and then have a 'shim' the other side of that that's engine specific) | 18:07.45 |
| Spot on. | 18:07.57 |
henrys | paulgardiner:yes that would make sense. | 18:08.17 |
Robin_Watts | ray_laptop: I've not looked. I'm being occupied by the mupdf customer. | 18:08.31 |
paulgardiner | "Clever minds thinks alike" or "fools rarely differ". Not sure which. | 18:08.33 |
ray_laptop | Robin_Watts: np. | 18:08.39 |
| just whining | 18:09.00 |
paulgardiner | ... re Robin and I saying almost identical statements | 18:09.27 |
Robin_Watts | The MuPDF customer is complaining that the new mupdf code is slower than the old stuff. All my tests on my machine here suggest that the new code is > 20% faster on the page in question. | 18:10.19 |
| but his timings do seem to be strange. | 18:10.37 |
paulgardiner | henrys: one small issue, I hope small: it would be very handy for me if I could work part time on it, at least initially. | 18:10.43 |
henrys | paulgardiner:so 20 a week say? | 18:11.15 |
paulgardiner | Yes, sure, or a week on a week off. My thought was guarantee you half time, but probably do more like 2/3 to 3/4, if the variation is acceptable to you. | 18:12.37 |
henrys | Actually I think we get into legal issues with full time without hiring you anyway so that is preferred - certainly we would in the states I guess Miles has studied the UK issues. | 18:13.31 |
Robin_Watts | I believe we in the UK aren't technically employees (at least, I'm not because everything goes through my company). | 18:15.08 |
paulgardiner | henrys: I know of potential problems in the UK, but I think they would be my problem rather than yours... but still great if part time suits you too. | 18:15.31 |
Robin_Watts | I'd assume that Paul would be in the same boat (as our companies were set up at the same time, by the same people). | 18:15.41 |
henrys | yes part time should be fine. | 18:15.45 |
Robin_Watts | paulgardiner: You mean IR35? | 18:16.21 |
paulgardiner | Robin_Watts: yes. | 18:16.36 |
Robin_Watts | I went through my contract with qdos (IR35 consulatants) and discussed some minor changes with Miles. | 18:16.56 |
| And we came to a form of words that everyone was happy with. | 18:17.17 |
paulgardiner | henrys: well, there are no other problems for me. I'd be really keen to do the work. | 18:17.23 |
henrys | when tor8 comes around let's have a chat about the project's requirements then you can start. | 18:17.24 |
paulgardiner | Brilliant. | 18:18.30 |
henrys | I'd lean toward a spec up front but if tor8 wants to keep things less formal I'm okay with that. | 18:18.47 |
paulgardiner | I haven't thought about it much yet, but I wondered about starting the design by pretending we were designing the whole thing MuPDF the engine and how then work together, so that the inital design for how they interact wouldn't be poluted by realities. Then work out what compromises are needed for the real MuPDF and the real engine. | 18:20.47 |
| s/then/they/ | 18:21.04 |
Robin_Watts | I wonder whether the first part of the work shouldn't be writing the spec (and pauls suggestion may not be a bad way to get there). | 18:23.32 |
henrys | that suggests a spec. Knowing tor8 he's probably cut through right to the realities and will want to start there but ... | 18:23.48 |
paulgardiner | henrys: sounds sensible. | 18:25.04 |
| *: sounds sensible. | 18:25.23 |
henrys | let me get an email to tor8 to show up at IRC 9:00 am PST tomorrow and we'll talk more about it. Do those times work for you two? | 18:26.00 |
Robin_Watts | Sure. | 18:26.18 |
| Let me just blip tor on google chat in case he's around now... | 18:26.29 |
henrys | or that too. | 18:26.46 |
| I would rather wait for tomorrow and collect my thoughts I'm in the midst of something else right now. | 18:27.46 |
Robin_Watts | Oh, sure. | 18:27.55 |
| (Sometimes tor8 has irc on one machine, and then works on another, I think) | 18:28.32 |
paulgardiner | I'm sure 9:00 is fine with me. | 18:29.08 |
Robin_Watts | paulgardiner: (That's 5pm UK time) | 18:29.20 |
paulgardiner | Thanks. Was struggling. :-) | 18:30.24 |
Robin_Watts | paulgardiner: When I try and build the android app here, I get told that android/bitmap.h doesn't exist. | 18:48.18 |
| Is that something you've seen ? | 18:48.36 |
paulgardiner | No. I don't think I have. Strange | 18:56.22 |
| On the branch? | 18:57.26 |
Robin_Watts | Oh, no. I'm on the trunk. | 18:58.00 |
| D'Oh. | 18:58.05 |
tor8 | paulgardiner, Robin_Watts: re javascript, I'd prefer a non-c++ library if one exists. v8 worries me because it's very platform specific with all its native code generation. | 19:02.42 |
| but as you both say, a shim layer to separate the javascript interpreter from the mupdf library bindings sounds like a good plan | 19:06.08 |
Robin_Watts | henrys: The bmpcmp for the hpgl path stuff doesn't look good :( | 19:31.01 |
| It looks like lots of stuff is being stroked when I wouldn't expect it to be. | 19:32.21 |
henrys | the simple file I sent you works okay, I assume. | 19:33.55 |
| if you send me the patch I'll generate a few simple cases maybe it's all the same issue. | 19:34.34 |
Robin_Watts | Will sort that out after dinner. | 19:35.10 |
henrys | okay I'm going for a run. | 19:35.51 |
mvrhel | henrys: testing a fix that made at least one file work on my machine | 20:58.34 |
| hopefully this will fix them all | 20:58.45 |
| it was an interesting case where there had to be a softmask processing going on during a vmreclaim | 20:59.24 |
| during softmask processing the profiles in the manager are swapped out | 20:59.42 |
| so there was a mismatch in what the user params said there was and what there was so it was trying to free the soft mask profiles and replace them | 21:00.32 |
henrys | ah interesting. | 21:02.10 |
| marcosw:would you like to go back to the customer and propose robin_watts solution which sounds like a reasonable way of doing things. Did you have some idea about, whatever makes him go away is good by me. | 21:08.34 |
Robin_Watts | which customer is this? | 21:09.24 |
marcosw | that's what I was wondering... | 21:09.39 |
henrys | the portfolio stuff - why aren't you eating? | 21:09.45 |
Robin_Watts | has eaten. | 21:09.53 |
henrys | Robin_Watts:oh right that was a while ago | 21:10.42 |
marcosw | you mean adding a -dPortfolioEntry command line option? | 21:11.07 |
henrys | I was concerned about how we enumerate pages and I thought Robin_Watts' bug comment made sense. | 21:12.44 |
| are we talking about the same thing? | 21:12.54 |
marcosw | I think Robin_Watts' suggestion does make sense, however I was just writing a comment questioning how it might be used to extract pages from multiple portfolio pdf documents, such as the customer's example in the description. | 21:16.00 |
Robin_Watts | marcosw: Wanting to take pages 1-3 from the cover document, 1-2 from portfolio entry 1, and pages 3-4 from portfolio entry 2 ? | 21:17.01 |
marcosw | page 1 from the first document and pages 1 through 4 from the second document. | 21:17.35 |
henrys | mvrhel has submitted I'm headed to the dashboard to watch the excitement. | 21:17.39 |
| ;-) | 21:17.50 |
mvrhel | :) | 21:18.23 |
| so far so good | 21:18.28 |
Robin_Watts | gs -sDEVICE=pdfwrite -o out.pdf -dPortfolioEntry=1 -dFirstPage=1 -dLastPage=1 in.pdf -dPortfolioEntry=1 -dFirstPage=1 -dLastPage=4 in.pdf | 21:18.37 |
| out.pdf is only flushed when the output device changes, right? | 21:18.56 |
marcosw | I just added a very similar suggestion to the bug report. | 21:19.42 |
henrys | marcosw but we need feedback from phil yes? | 21:21.02 |
marcosw | yes, I'm writing the email as we speak (or would be if I wasn't monitoring the cluster). | 21:21.26 |
henrys | okay great, I have a quick errand to run and I'll bbiaf | 21:21.44 |
mvrhel | brb | 21:25.10 |
Robin_Watts | marcosw: Ah, your suggestion is NOT the same as mine. | 21:27.47 |
| In mine, I mention in.pdf twice. | 21:28.01 |
marcosw | yours is better, allowing different pages to be extracted from different files. | 21:28.18 |
Robin_Watts | Mine should work already, if only we have -dPortfolioEntry | 21:28.40 |
| s/have/had/ | 21:28.46 |
| OK. Press the big red 'cluster explode' button now! | 21:29.16 |
marcosw | btw, I assume your example can be simplified to: -dPortfolioEntry=1 in.pdf -dPortfolioEntry=2 -dLastPage=4 in.pdf | 21:30.51 |
| Robin_Watts: odd, the cluster didn't explode. | 21:32.25 |
mvrhel | hmm still a couple seg faults | 21:34.17 |
| however there are fewer | 21:34.59 |
marcosw | cool, my new code worked, though several of the jobs generated millions of warnings they were correctly truncated in the logs. | 21:35.28 |
mvrhel | ok. still something funky going on in my fix | 21:37.27 |
| following a vmrclaim | 21:37.42 |
henrys | Robin_Watts:based on your bmpcmp I would expect the very simple path example I emailed you to not work. | 22:02.04 |
mvrhel | henrys: so it looks like I still have a few issues. I will beat on those tonight. have to run the kids up to their ski lessons in a bit | 22:25.42 |
| Forward 1 day (to 2012/03/01)>>> | |