| <<<Back 1 day (to 2012/10/10) | 2012/10/11 |
chrisl | Hmm, a one node cluster - no wonder it's taking a long time :-( | 08:52.21 |
Robin_Watts | chrisl: speedy. | 09:18.06 |
chrisl | Robin_Watts: yeh - strange that the one node remained when all the others went down..... | 09:19.45 |
Robin_Watts | chrisl: Is there a make macro trick that I can use to enable/disable threadsafe devices? | 09:35.57 |
| Akin to: | 09:36.07 |
| #ifdef GS_THREADSAFE | 09:36.44 |
| #define NTS(X) | 09:36.46 |
| #else | 09:36.48 |
| #define NTS(X) X | 09:36.50 |
| #endif | 09:36.52 |
| So I can wrap device names in NTS() if they are not thread safe ? | 09:37.04 |
| I fear not :( | 09:42.54 |
tor8 | Robin_Watts: ooh, there's a neat feature in chrome I just discovered. multiple user profiles in the browser, so you can have separate sessions open. now I can have both private and artifex gmail up without having to use multiple browsers :) | 09:48.18 |
Robin_Watts | nice. | 09:49.50 |
| I refuse to use gmail though, so... :) | 09:50.04 |
chrisl | Robin_Watts: no, not that I'm aware of. Let me have a think..... | 09:51.03 |
Robin_Watts | Hmm. Email carriers/ISPs generally assert Common Carrier status - they claim to be like telephone companies, where they can't be held responsible for the contents of the calls. | 09:51.30 |
| So if someone plots murder over a telephone line, the telephone company isn't counted as colluding in it. | 09:51.53 |
| Given that googles T&Cs mean they read everything that's sent through them, does that mean that doesn't hold any more? :) | 09:52.33 |
tor8 | Robin_Watts: I'm fond of webmail, never did like local mailboxes, and gmail was the only decent web mail out there for the longest time. and now I'm too lazy to switch... :/ | 09:58.57 |
chrisl | Robin_Watts: how about having the non-threadsafe devices get redefined to the null device in a thread safe build? | 10:01.16 |
tor8 | Robin_Watts: all governments snoop everything on the interwebs these days anyway :( just treat all email as public correspondence :)) | 10:01.19 |
Robin_Watts | chrisl: How ? | 10:02.00 |
chrisl | Something like: http://pastebin.com/sjnM7fip | 10:02.49 |
| I'm not sure how you're going to identify non-threadsafe devices...... | 10:03.46 |
Robin_Watts | but how can I do that in the makefile ? | 10:04.04 |
| At the moment in the makefile we have something like: | 10:04.18 |
| DEVS1=device1 device2 device3 device4 ... | 10:04.36 |
| I was hoping to be able to change that to be: | 10:04.43 |
| DEVS1=device1 NTS(device2) NTS(device3) device4 ... | 10:04.59 |
chrisl | Oh, I see. I suspect if it can be done, it will be difficult to do in a non-make specific way | 10:06.18 |
| The problem is, those device lists aren't targets, so we can't do the usual trick with appending _<variable> to switch between targets. | 10:09.06 |
| We can probably do something in configure, and something different nmake | 10:10.58 |
Robin_Watts | will ponder it while running. | 10:20.13 |
chrisl | Robin_Watts: the final "stage" in including a device in the build is (I think!) when genconf writes the gconfig.h header file. How about augmenting genconf to take a -devs-exclude parameter, which would be a list of devices to drop, and putting the logic in genconf.c to exclude those. Then we can maintain the list of non-threadsafe devices in a makefile. | 10:54.54 |
paulgardiner | Robin_Watts, tor8: there are a few commits on paulg/master when you have a moment. | 11:04.30 |
Robin_Watts | chrisl: Would that stop it building those devices? Or just not link them in? | 11:45.21 |
| I was thinking that we could do something with the .dev suffix. | 11:45.54 |
| Suppose we have: DEVS=a.dev b.dev c.dev etc | 11:46.04 |
| Then elsewhere we have: | 11:46.10 |
| a.dev: foo | 11:46.31 |
| bar .... | 11:46.33 |
| We could instead have: | 11:47.06 |
| a.$(DEV): foo | 11:47.08 |
| bar... | 11:47.10 |
| a.$(NTS): null_device | 11:47.12 |
| In normal builds we set DEV to dev and NTS to bogus | 11:47.53 |
| In threadsafe builds we set DEV to bogus and NTS to dev | 11:47.56 |
chrisl | I think if we have <device>.bogus (or whatever) we'd get errors. Also, we can't do conditionals in the makefiles (other than those specific to nmake) | 11:55.44 |
| Robin_Watts: ^^ | 11:55.51 |
Robin_Watts | chrisl: The DEVS list would always list blah.dev | 11:56.16 |
| We'd just end up with unused targets blah.bogus, which should be fine. | 11:56.42 |
| The DEV and NTS setting would be done in the top level makefiles (the ones that are specific to nmake or gnu make) | 11:57.14 |
chrisl | I think we'd still end up with *something* in gconfig.h for each non-threadsafe device, which would cause problems | 11:58.25 |
| I'm being called for lunch..... bbiaw | 11:58.55 |
Robin_Watts | yeah, the null device. | 11:59.30 |
| OK, have a good one. thanks. | 11:59.35 |
kens | Robin_Watts : VMWare player 4 can open my old VMs | 13:40.28 |
| So I've added them all, opened and edited all the settigns in the hope that then upgrading to version 5 will allow me to open them there | 13:40.55 |
Robin_Watts | kens: Cunning. | 13:45.28 |
kens | Am now playing teh 'restart WIndows' shuiffle, I should run a sweepstake on how many times I need to reboot it while installign software... | 13:46.02 |
Robin_Watts | kens: I updated my nvidia graphics drivers under windows last night. | 13:47.27 |
| It managed it without rebooting. | 13:47.34 |
kens | I did that yesterday too | 13:47.39 |
Robin_Watts | I was amazed. | 13:47.41 |
kens | 170+Mb of download | 13:47.48 |
| Oho, installing VMWare player 5, and it found the VMs I already loaded under 4, a good sign | 13:49.35 |
| Nope 'this virtual machine's policies are too old'.... | 13:49.59 |
| Hmm the user forums suggest creating a new VM and attaching the .vmdk to it. Bizarre. | 13:51.37 |
Robin_Watts | The dk being the disc? | 13:52.12 |
| That could work. | 13:52.18 |
kens | More suggestiosn to rename the .vmpl file and edit teh .vmx file. I'll try that | 13:52.42 |
| Robin_Watts : How do I persuade Windows 7 to show me the file extensions in explorer ? | 13:55.46 |
Robin_Watts | Urm... Folder options, same as ever? | 13:56.11 |
kens | Can't see it anywhere :-( | 13:56.24 |
Robin_Watts | Organise => Folder and search options ? | 13:57.00 |
kens | Hmm, let me see | 13:57.28 |
Robin_Watts | Organise => Folder and search options => View -> ... | 13:58.10 |
kens | Yes, that's it thanks, its different in my Vista setup (I may well have that set for an older 'cklassic' view as well) | 13:58.18 |
Robin_Watts | The options are the same as my memory of XP, but it's a different route to get to it. | 13:58.50 |
kens | Yes, same options, amusingly hidden in a different place (used ot be 'properties) | 13:59.17 |
paulgardiner | Robin_Watts, tor8: not that I'm in any hurry, but did you see my message about some commits needing some reviewing sometime? | 13:59.42 |
Robin_Watts | paulgardiner: I did. | 13:59.48 |
| I'll look in a mo. | 13:59.52 |
paulgardiner | No hurry at all. Just checking it was noted. | 14:00.09 |
Robin_Watts | paulgardiner: I'd forgotten, so prodding was wise :) | 14:00.45 |
paulgardiner | :-) | 14:00.51 |
kens | OK removing the .vmpl file and editing the.vmx makes it work. | 14:01.12 |
Robin_Watts | kens: excellent! | 14:01.23 |
kens | Making progress.... | 14:01.41 |
Robin_Watts | tor8: My skypey mupdf customer has some local changes to mupdf. | 14:01.42 |
| They have a new 'flush' function in the device interface that is called after each page. | 14:02.01 |
| And they use it in the text device to force the current span to be written into the sheet/page datastructure. | 14:03.27 |
kens | OK all VMs now running from Windows 95 to <ahem> the VM which shall not be named | 14:14.39 |
Robin_Watts | excellent. | 14:15.05 |
kens | On with the setup.... | 14:15.15 |
Robin_Watts | has DOS 6 floppies in the draw here. And windows 3.1 if you want to make a VM for that :) | 14:15.33 |
| drawer, even. | 14:15.38 |
kens | I have a DOS 6.22 ISO :-) | 14:15.59 |
Robin_Watts | paulgardiner: pushed. | 14:16.31 |
paulgardiner | Great. Thanks Robin | 14:16.44 |
Robin_Watts | np. | 14:16.48 |
chrisl | Robin_Watts: I'm still struggling to see a way to handle the dropping of devices from the makefile in a make-independent fashion :-( | 14:17.54 |
Robin_Watts | chrisl: Don't panic too much. | 14:18.07 |
| Ultimately, we should just fix 'em all :) | 14:18.41 |
chrisl | It's such a pain that we can't use conditionals - what idiot left such a fundamental out of make?!?! | 14:18.49 |
Robin_Watts | chrisl: It's in make. | 14:19.01 |
| It's just different between gnu make and nmake :) | 14:19.12 |
chrisl | Ah, but it's not in Sun make, and a few others | 14:19.34 |
| Robin_Watts: were you able to get a list of offending globals? | 14:20.31 |
Robin_Watts | Yeah, with nm as you suggested. | 14:20.56 |
| I went through them all yesterday, and attached it to a bug. | 14:21.05 |
| http://bugs.ghostscript.com/show_bug.cgi?id=693376 | 14:21.19 |
| The globals are one problem. | 14:21.43 |
| The use of gs_lib_ctx_get_non_gc_mem_ptr (or whatever stupid name it has) is another. | 14:22.14 |
tor8 | Robin_Watts: that seems odd, the device interface is not a multi-page interface... | 14:23.20 |
chrisl | Robin_Watts: holy crap, that's used all over the joint :-( | 14:23.51 |
Robin_Watts | chrisl: Indeed. | 14:23.59 |
| It was my previous way of avoiding rewriting lots of stuff. | 14:24.51 |
chrisl | Even with taking that into account, it's used a lot more than I thought | 14:25.43 |
Robin_Watts | chrisl: I think it's restricted to lcms v1, and the contrib devices now. | 14:41.09 |
| gp_check_interrupts should never be called with mem NULL now. | 14:41.42 |
chrisl | Well, lcms1 will disappear when I get around to it. I guess I could have a hack at the contrib devices at some point | 14:42.10 |
Robin_Watts | chrisl: I was considering having a look at some of the globals in the contrib devices now. | 14:42.48 |
chrisl | Robin_Watts: okay. If you want to share the pain, I don't mind. | 14:43.29 |
Robin_Watts | And I'd probably do the gs_blahblah_t stuff in the files I touch at the same time. | 14:43.35 |
| BUT first I should make a list of which devices I need to exclude so I can do test builds on linux nicely. | 14:43.58 |
| And the cluster goes BOOM! | 14:48.02 |
chrisl | Okay, so that's my changes that kill the cluster - I guess I need to run a local test script to find why :-( | 14:49.40 |
Robin_Watts | paulgardiner: We have various v8 tgz's available to download. | 15:20.43 |
marcosw | what have you guys down to the cluster now? | 15:20.44 |
Robin_Watts | See http://ghostscript.com/~robin/ and look at v8-... | 15:21.03 |
| We ought to put the android one there too. | 15:21.10 |
chrisl | marcosw: my fault - dodgy code. It should recover itself | 15:21.15 |
Robin_Watts | Could you make a tgz (or a zip) in the same style, and let me have a copy please? | 15:21.43 |
paulgardiner | Robin_Watts: Right, good thought. Should I zip it up and send it to you? | 15:21.48 |
marcosw | okay. is this related to the large number of warnings that clogged up the cluster earlier? | 15:21.49 |
chrisl | marcosw: very likely, yes | 15:22.02 |
marcosw | i'm going to add code so that jobs which generate too much output are killed. | 15:22.40 |
Robin_Watts | marcosw: If you do so, could you make it so that the output that caused the jobs to be killed is available? | 15:23.11 |
marcosw | that's the plan, though we'll see how well it goes :-) | 15:23.29 |
chrisl | Ah, so lots of warning output clog up the cluster - that would be it, then. | 15:23.43 |
Robin_Watts | ISTR (though I may be wrong, as usual) that when jobs are killed (say through timeouts) the logs are not uploaded. | 15:23.50 |
chrisl | Although, I'm *sure* my code was right! | 15:23.59 |
Robin_Watts | so you can't easily see which jobs were causing the timeouts. | 15:24.19 |
marcosw | chrisl: the warnings were filling up the disk. | 15:30.20 |
chrisl | marcosw: oh, that's bad :-( | 15:30.40 |
| Sorry | 15:30.42 |
marcosw | only happened for a small number of jobs. i.e. tests_private__pdf__PDF_1.7_FTS__fts_17_1700.pdf.psdcmyk.72.0 | 15:31.19 |
chrisl | I think anything with a Truetype in it, or that we fall back to the CID substitute font would have spouted errors | 15:32.23 |
marcosw | this wasn't that, it was the mismatched Q/q warnings, an infinite (?) number of them. | 15:34.51 |
| disturbingly enough I also had them on my iMac after I upgraded to mountain lion and the new Xcode, though in a different file than the cluster was reporting and not reproducible on linux. | 15:35.42 |
| on the iMac they were in Bug690497.pdf | 15:36.14 |
chrisl | We can get the mismatched q/Q warnings when we give up on a content stream, which could have happened with a font reading error | 15:36.49 |
marcosw | I'm now wondering if this is an indeterministic issue. | 15:36.50 |
| chrisl: but an infinite number of them? | 15:37.09 |
chrisl | I would not expect that, no | 15:37.23 |
marcosw | I'll keep a copy of your source and try to figure out what really happened. may be worth opening a bug. | 15:38.56 |
henrys | not much action at the show we had poor booth placement due to some sort of mixup in making the reservation. I guess show visitors figure if there that far out not worth checking it out. | 15:53.13 |
kens | Often the case, yes :-( | 15:53.34 |
Robin_Watts | Damn. And even if they offer you the money back for the booth, it's no consolation really. | 15:54.20 |
kens | Costs way more than that in lost time | 15:54.46 |
Robin_Watts | and plane/hotel bills. | 15:55.14 |
| chrisl: I've gone through configure.ac and separated the non threadsafe devices out into separate lists. | 15:55.53 |
henrys | I am not a believer in having a booth I'd much rather see us visit more shows and walk the floors stalking game. | 15:55.59 |
kens | But then you don't find anyone else doing the same. | 15:56.42 |
Robin_Watts | chrisl: So I now have HP_DEVS and HP_DEVS_NTS | 15:56.44 |
Robin_Watts | buys henrys a sandwich board. | 15:56.59 |
| THE END IS NIGH (for Adobe) | 15:57.21 |
kens | and a megaphone ? | 15:57.26 |
chrisl | Robin_Watts: Can't we just have one list of non-thread safe devices that we "post process" to remove them from the final tally - we already have too many device lists, IMHO | 15:57.55 |
henrys | oh the retailers need a booth and the folks running those booths are our customers. | 15:58.19 |
Robin_Watts | chrisl: Well, if that's the way you'd rather do it. | 15:59.26 |
| chrisl: I don't really speak configure. | 16:00.00 |
chrisl | Robin_Watts: well, the intention is for all these devices to be fixed at some point, so we'd only have to remerge the lists as we go. | 16:00.14 |
Robin_Watts | chrisl:Right, but moving a device from one list to another is something even I can do :) | 16:00.36 |
| Would it be possible to have a --threadsafe thing for configure? | 16:01.32 |
henrys | We did go out to dinner with one of our large customers and he is interested in doing digital signage with PDF. There actually is a signage standard that calls for svg, I didn't mention that fact. | 16:01.39 |
chrisl | Robin_Watts: yes, I'll sort it out tomorrow (I have a squash match real soon!) | 16:02.06 |
Robin_Watts | That set -DGS_THREADSAFE and then ... | 16:02.15 |
| chrisl: Thanks. | 16:02.18 |
| I'll mail you my hacked file, so you have the lists of devices at least. | 16:02.30 |
chrisl | Great, thanks | 16:02.46 |
| Okay, off now.... | 16:02.59 |
Robin_Watts | break a string. | 16:03.07 |
chrisl | ;-) | 16:03.13 |
kens | Time for me to be off as well, night all | 16:24.05 |
Robin_Watts | ooh. wossa difference between gs/base/contrib.mak and contrib/contrib.mak ? | 17:06.09 |
malc_ | tor8: http://upload.wikimedia.org/wikipedia/en/timeline/99e6d81f0dcfcdef5404fdbcfec82923.png http://xkcd.com/1015/ | 17:07.20 |
sebras | malc_: now you're not being kind. :) | 17:26.32 |
malc_ | heh | 17:27.23 |
sebras | malc_: is there any reason you're sending those urls? | 17:27.41 |
| malc_: mupdf-related I mean. :) | 17:27.53 |
malc_ | sebras: not really, no | 17:30.39 |
sebras | malc_: if you like xkcd in general, try abstrusegoose.com too. | 17:32.46 |
malc_ | sebras: i had xkcd packed into a cbz, and recently after adding a "make cbz not suck" featue into my code i was able to read it in its entirety | 17:35.02 |
| thanks for the link btw | 17:35.19 |
henrys | marcosw:ping | 17:37.06 |
marcosw | henrys: morning | 17:37.12 |
henrys | the 8.54 -> 8.64 performance issue looks like the ball is in their court right? | 17:38.25 |
| they're buggin' me and miles and I told them to answer your question and we'll continue. | 17:39.26 |
marcosw | henrys: i agree, we can't duplicate it and in any case 8.64 is now longer supported. I did run a test yesterday to look at the memory issue they also reported but haven't analyzed the results. | 17:40.44 |
henrys | with a grenade launcher I think I could hit their facility from my house. | 17:41.09 |
| maybe I'd need a mortar | 17:41.48 |
| marcosw:anyway I've told her to continue working with you. | 17:44.35 |
marcosw | okay | 17:44.49 |
Robin_Watts | The windows builds do not include the contrib devices that the unix builds do. | 17:46.21 |
| Anyone object to me making them do so? | 17:46.34 |
henrys | be surprised if they compiled | 17:47.35 |
Robin_Watts | They do now :) | 17:49.14 |
| mvrhel_laptop: I tweaked the gscms code for retrieving colorants to avoid the use of a static earlier. | 19:34.35 |
mvrhel_laptop | Robin_Watts: oh yes | 19:53.14 |
| we talked about that at one point | 19:53.18 |
| didnt we? | 19:53.30 |
| I was thinking you already commited | 19:53.41 |
Robin_Watts | mvrhel_laptop: We did, but I got sidetracked. | 20:48.00 |
slestak_work | I am doing a text dump of a landscape pcl document using | 21:33.10 |
| usign the txtwrite device, the output appears transposed (vertical strings) and mirrored (reversed)/ | 21:33.48 |
| the transposed I can almost understand since, but the mirrored seems incorrect to me. | 21:34.34 |
sebras | Robin_Watts: git clone --recursive | 21:35.30 |
slestak_work | this is an example of what I am talking about http://www.screencast.com/t/vcWwvies1 | 21:36.22 |
| ive tried transposing this myself, but wanted to touch base to see that what I am getting as output was what was expected. the strings pulled out of here are "NET" "COST" "UNIT" "10762981-1" | 21:39.13 |
Robin_Watts | slestak_work: In case you read logs; please open a bug at ghostscript.com, and attach the source file you are using and the command you are using. | 23:12.52 |
| Basically, unless we can reproduce the issue on our machines here, we can't begin to bughunt it. | 23:13.15 |
| Forward 1 day (to 2012/10/12)>>> | |