| | <<<Back 1 day (to 2021/05/04) | Fwd 1 day (to 2021/05/06)>>> | 20210505 |
| artifexi- | <ator> @Robin_Watts small fix to the pdfwrite device on tor/master | 12:23.54 |
| | <Robin_Watts> looking | 12:29.53 |
| | <Robin_Watts> Yeah, how dumb is that. | 12:31.42 |
| | <Robin_Watts> lgtm. | 12:31.45 |
| | <ator> Thanks! | 12:36.49 |
| | <ator> @cgdae @Robin_Watts FZ_STEXT_MEDIABOX_CLIP seems to me like it ought to be the default behavior, any reason why it isn't? | 12:58.09 |
| avih | ator: for reference, we identified that the stack breakage started after bumping binutils (mingw) from 2.35.1 to 2.36.1, and later tests show that 2.35.2 is also broken. we didn't try to check what's the difference between 2.35.1 and 2.35.2 | 13:16.28 |
| | but i do wonder what it could be | 13:16.57 |
| | and recall only 64 bits builds are affected | 13:17.09 |
| | ator: do you think this be an ld bug? | 13:18.06 |
| ator | avih: has anyone managed to get a gdb stack trace from the crash or valgrind error from this? | 13:21.13 |
| avih | ator: valgrind impossible AFAIK (windows), and personally i never managed to get gdb working on windows. if you can help me with that, or if you want a broken mujs.exe to try it yourself, i can provide | 13:22.14 |
| ator | right, it's only windows then. for some reason I thought you could replicate on linux too (probably all that talk about gcc) | 13:22.55 |
| | does it crash in wine? :) | 13:23.33 |
| avih | well, we don't know if that's only windows (64), but it's possible. best testcase would be to try and compile with gcc 10.2.1 with binutils 2.36 or some such on linux. i didn't try that. | 13:23.44 |
| | (i don't know how to create such setup) | 13:24.06 |
| | and i didn't try in wine :) | 13:24.13 |
| ator | I have no idea where to start debugging with windows. Haven't used windows for development in years. | 13:24.17 |
| avih | but you could try, i don't see why it wouldn't crash in wine, but i don't have a wine setup | 13:24.31 |
| ator | send me the problematic mujs.exe and maybe I can try it with wine or in a virtualbox with a windows | 13:24.57 |
| avih | k, i hope i have one available :) (i'm rebuilding the toolchain for each test, that's long time...) | 13:25.33 |
| | ator: https://0x0.st/-B-i.zip (let me know you can open the zip). to reproduce: 1. run mujs.exe 2. at the prompt enter: foo (and press enter) <-- result: REPL exits without error instead of not-exiting and displaying reference error | 13:28.01 |
| | mujs is exactly 1.1.2 in this binary | 13:29.15 |
| ator | I got the file and it starts up in wine. how do I make it crash? | 13:29.37 |
| avih | (i don't recall if this has debug symbols. i'll need to rebuild the toolchain again if you need that. let me know) | 13:29.49 |
| ator | and you just said | 13:29.54 |
| | yeah, I get a crash if I hit 'foo' and press enter | 13:30.02 |
| avih | ator: steps to reproduce same message as the link | 13:30.04 |
| | ok | 13:30.07 |
| ator | I'm guessing those regular crashes are somehow down to longjmp/setjmp since they occur when there's an error | 13:30.54 |
| avih | that's without symbols, right? | 13:30.56 |
| ator | non-error code seems to run fine | 13:30.59 |
| avih | yes, i presume so too, but lua(jit) seems fine | 13:31.12 |
| | but then again, i didn't actually test errors in luajit... hmmm | 13:31.33 |
| | in lua is much less likely to hit longjmp than in js, because typically this needs a buggy script, but in js it's the normal order of operation to use try/catch | 13:32.21 |
| | (i REALLY hope this aborts with lua errors too) | 13:33.12 |
| | (testing) | 13:33.18 |
| | hmm.. seems no. i do get a stack trace in lua when i add errors to a script | 13:35.49 |
| | ator: do you need a binary with debug symbols? | 13:37.32 |
| ator | that might help, yes | 13:38.17 |
| | *sigh* stupid windows 10 (in a virtualbox thankfully) is being helpful again | 13:39.03 |
| | "can't run visual studio while setup is running" | 13:39.19 |
| | oh yeah, it's auto-updating shit again... | 13:39.25 |
| avih | k, this might take half an hour to rebuild the toolchain. also, notice that as far as i can tell it's the same mingw stuff and same gcc - only binutils differs. that's why i think it could be an ld issue | 13:40.28 |
| | (ii now have a working setup, and i'll upgrade only binutils from 2.35.1 to 2.35.2 and hopefully the issue will return) | 13:41.41 |
| ator | oh ffs now it needs to connect to the cloud to verify my free subscription. how anybody can put up with windows 10 and visual studio is beyond me... | 13:42.12 |
| | and there it exploded with an "unknown error" and quit right out of my debugger session. isn't the cloud amazing? | 13:42.39 |
| avih | the cloud is amazing, yes :) | 13:44.25 |
| | (actually, now that i think about it, i know that lua uses longjmp, but i don't know for a fact that luajit uses longjmp) | 13:45.08 |
| | ator: once i have the compiler working, you can give me any piece of c code to test. it might be reproducible with setjmp/longjmp regardless of mujs. | 13:45.46 |
| ator | aaaand now it throws a "Script Error" when trying to log in to visual studio | 13:46.11 |
| avih | are you like debugging in a remote windows somehow? | 13:46.34 |
| ator | I'm trying to run the debugger from visual studio express (or whatever it's called) in a windows 10 VM | 13:46.56 |
| avih | ator: i don't know if windows debugger is good for mingw binaries. i think theoretically gdb should work with them, but i never got it working myself | 13:47.12 |
| ator | but microsoft is being obnoxious as usual, and making it painful enough that I'm one more error message away from rm -rf'ing the VM image | 13:47.36 |
| avih | well, i KNOW people used gdb with mingw binaries, and it _should_ work, but i couldn't get it working | 13:47.42 |
| | yeah, windows 10 can be a PITA if you don't keep it up to date regularly. | 13:48.26 |
| artifexi- | <cgdae> I'd be happy for it to be the default, but i thought (perhaps mistakenly) that there was a certain reluctance to the feature when it was added. | 13:52.15 |
| | <Robin_Watts> I have no objection to it being the default, but it's a change in behaviour, so having a way to restore the old way seems important (and having it as a flag gives us that) | 13:53.53 |
| avih | shit. i've upgraded only binutils, and the exe is still fine. this is not a good scenario for PEBKAC... | 14:16.31 |
| | wtf is going on... | 14:18.23 |
| | could this be some race while building gcc?!? | 14:18.53 |
| artifexi- | <ator> @cgdae @Robin_Watts one commit on tor/master changing the default then | 14:31.24 |
| | <Robin_Watts> @ator lgtm | 14:32.58 |
| avih | ok, i don't have energy now for debug cycles of 30 minutes. | 14:41.43 |
| | ator: sorry, no debug symbols for now. i';; get back to it when i'm more sane, apparently. | 14:42.03 |
| ator | avih: no rush, I ran out of patience with windows for today anyway :) | 14:47.51 |
| avih | :) | 14:48.07 |
| | so much pain in dev... | 14:48.25 |
| artifexi- | <cgdae> Nice to see the XOR operator being used 🙂 | 14:55.09 |
| | <Robin_Watts> Well, there isn't an &~= operator 🙂 | 15:00.29 |
| | <<<Back 1 day (to 2021/05/04) | Forward 1 day (to 2021/05/06)>>> | |
Log of #mupdf at irc.freenode.net.