00:01:47 | | Join massiveH [0] (~massiveH@2600:4040:a992:a300:dd2e:e4a0:28b:b534) |
00:02:08 | | Join Trzyzet_ [0] (~Trzyzet@86.8.78.139) |
00:02:45 | | Quit Trzyzet (Ping timeout: 260 seconds) |
00:02:45 | | Nick Trzyzet_ is now known as Trzyzet (~Trzyzet@86.8.78.139) |
00:10:35 | *** | Saving seen data "./dancer.seen" |
00:34:54 | _bilgus | thats weird |
00:35:40 | _bilgus | hmm there is something with those strings or the count |
00:36:24 | _bilgus | i'm in the middle of a build atm but ill throw another up in a bit |
00:36:28 | MarcAndersen | maybe because the gigabeats has tone control and other special strings? |
00:37:05 | _bilgus | IDK i'm just speculating that it has an upper limit because I don't see anything else |
00:37:48 | _bilgus | i'll try adding back the chessbox strings and leaving those pictureflow ones out if it fails then we know its a size issue |
00:37:53 | _bilgus | I guess?? |
00:38:12 | MarcAndersen | That might be the case. was the chessbox before the pictureflow? If so it might be those few pictureflow strings that made it overflow? |
00:38:15 | _bilgus | if its still good i guess we can add the strings back one at a time till it fails |
00:38:33 | _bilgus | no I just picked chessbox because it has lots of strings |
00:39:00 | _bilgus | since those builds don't have plugins it was an easy target |
00:39:04 | MarcAndersen | So then it may be those few pictureflow strings that made the limit go over? |
00:39:29 | _bilgus | at this point there are a lot more |
00:39:35 | _bilgus | like 20 or so |
00:39:44 | MarcAndersen | but what about that time? |
00:41:04 | MarcAndersen | I think we're getting it squashed. The ipods also has more strings than the sansas right? |
00:41:11 | _bilgus | well IDK yet but I really doubt it because the cound was at like 1075 and that commit brought it to 1080 or so i'd expect it to fail at a power of 2 like 1025 or 1769 or something |
00:41:25 | _bilgus | count* |
00:42:10 | MarcAndersen | Can you continue on your own or do you need me anymore tonight? |
00:44:18 | _bilgus | we can wait till tomorrow if you want to give me a few and try one last build i can go from there |
00:44:39 | MarcAndersen | How long would it take, i can wait a bit if you want? |
00:44:45 | _bilgus | I assume this build is nearly finished its been going forever |
00:45:03 | MarcAndersen | I'll wait |
00:48:58 | _bilgus | well all that waiting for the build to fail I guess at least i'm not waiting anymore compiling yours now |
00:49:45 | MarcAndersen | Why did it fail? |
00:51:15 | _bilgus | its another patch I'm missing something apparently |
00:52:57 | _bilgus | http://www.mediafire.com/file/781hpk1i4lvg1ic/gigabeatS_rockbox_HEAD-PF.zip |
00:53:13 | _bilgus | if that one works its got to be something with the strings |
00:58:35 | MarcAndersen | Does not work. Says "y" after each option and spells garbage. |
00:59:08 | MarcAndersen | Maybe my bisect wasn't too bad anyhow? |
00:59:55 | _bilgus | I'm going to say its something to do with the size of the voice file |
01:00 |
01:00:15 | _bilgus | maybe speachy will have some idea |
01:00:28 | _bilgus | but it can wait till tomorrow thanks for the help |
01:00:34 | MarcAndersen | Good point. Can you go on from here? |
01:00:57 | _bilgus | yeah i'll look around and try to get a few ideas of what we can try |
01:01:40 | MarcAndersen | Thank you. I also learned a lot today about bisect and other stuff thanks to you. Good night. |
01:03:56 | _bilgus | nn |
01:31:35 | | Quit massiveH (Quit: Leaving) |
02:00 |
02:02:18 | | Join chris_s [0] (~chris_s@ip-095-223-074-147.um35.pools.vodafone-ip.de) |
02:10:36 | *** | Saving seen data "./dancer.seen" |
02:18:31 | chris_s | amachronic: I'm seeing subtle image corruption at the left and right end of scrolling lines with 5d0c382, at least with certain fonts. |
02:18:37 | chris_s | Here's a screenshot of the M3K-Simulator with the default theme using the 20-Inter-V font from the Interpod theme. |
02:18:44 | chris_s | https://ibb.co/31m48Nj |
02:21:25 | | Quit chris_s (Quit: Connection closed) |
02:27:39 | | Join chris_s [0] (~chris_s@ip-095-223-074-147.um35.pools.vodafone-ip.de) |
02:27:54 | chris_s | Actually, the line doesn't have to be scrolling. I think, the last character just has to be partially off screen |
02:28:26 | | Quit chris_s (Client Quit) |
02:44:10 | | Join chris_s [0] (~chris_s@ip-095-223-074-147.um35.pools.vodafone-ip.de) |
02:44:31 | chris_s | could anti-aliased fonts only |
02:44:47 | | Quit chris_s (Client Quit) |
04:00 |
04:10:18 | | Join amachronic [0] (~amachroni@user/amachronic) |
04:10:37 | *** | Saving seen data "./dancer.seen" |
04:12:58 | amachronic | chris_s: that's definitely an alpha blit bug, probably missed it since I don't use antialiased fonts |
04:13:15 | amachronic | I checked them in the simulator but obviously not well enough |
04:59:06 | | Quit amachronic (Quit: amachronic) |
05:00 |
05:53:58 | | Join amachronic [0] (~amachroni@user/amachronic) |
05:54:32 | rb-bluebot | Build Server message: New build round started. Revision 593103cd8b, 303 builds, 8 clients. |
05:54:32 | rb-bluebot | lcd: Fix 4bpp optimized 16-bit alpha blit by Aidan MacDonald |
06:00 |
06:10:38 | rb-bluebot | Build Server message: Build round completed after 968 seconds. |
06:10:40 | rb-bluebot | Build Server message: Revision 593103cd8b result: All green |
06:10:41 | *** | Saving seen data "./dancer.seen" |
07:00 |
07:07:38 | | Quit amachronic (Quit: amachronic) |
08:00 |
08:10:42 | *** | Saving seen data "./dancer.seen" |
09:00 |
09:52:25 | rb-bluebot | Build Server message: New build round started. Revision 282a54b23c, 303 builds, 9 clients. |
09:52:25 | rb-bluebot | [Feature] filetree fallback to loading a playlist from disk by William Wilgus |
09:57:45 | _bilgus | nice |
09:58:17 | MarcAndersen | Any news about the voice bug? |
09:59:23 | _bilgus | no haven't heard from speachy yet and I spent about an hour looking through the voice/lang code for anything that could limit the number of clips I just don't see anything |
09:59:59 | _bilgus | I thought maybe the device was running out of ram but it has 64MB so thats pretty unlikely |
10:00 |
10:00:25 | _bilgus | have you looked in the debug menu does the voice menu show any errors? |
10:01:11 | MarcAndersen | It's hard when it can't spell, I'd need sighted assistance |
10:01:24 | _bilgus | maybe we could make a plugin that goes through every voice clip and see if it fails at a particular point or just all together |
10:02:21 | MarcAndersen | By the way have you tried if it does it in the simulator? I haven't. |
10:02:39 | _bilgus | no that would be nice brb |
10:05:35 | MarcAndersen | It unfortunately uses another voice file so I'm not sure |
10:06:42 | MarcAndersen | I'm pretty sure it's the size because that pf commit expands it by around 20 kb |
10:08:25 | rb-bluebot | Build Server message: Build round completed after 961 seconds. |
10:08:27 | rb-bluebot | Build Server message: Revision 282a54b23c result: All green |
10:09:00 | _bilgus | MarcAndersen, I'm beginning to lean towards that as well |
10:09:22 | _bilgus | but I don't see anywhere where it matters |
10:10:43 | *** | No seen item changed, no save performed. |
10:11:11 | MarcAndersen | No, specially when gigabeats has 64 mb ram and ipodnano1g has 32 |
10:12:33 | _bilgus | it works in the sim |
10:12:56 | MarcAndersen | So you say the bug is gone? |
10:13:05 | _bilgus | yeah |
10:13:18 | MarcAndersen | Dam it |
10:13:20 | _bilgus | not that the sim is a 1:1 representation |
10:13:53 | MarcAndersen | Then we need real targets. I'd be willing to test some more and help out where I can. |
10:15:21 | _bilgus | I'll add spelling to the voice debug menu |
10:16:09 | MarcAndersen | It already does, but when it produces garbage I don't think it would work |
10:16:32 | MarcAndersen | You did it for me some time ago |
10:22:10 | _bilgus | does the voice menu talk/spell thought I think its only the top level |
10:22:33 | MarcAndersen | Oh I think you're right now I think about it, sorry.. |
10:23:01 | _bilgus | it has some stats about the number of clips and cache misses etc |
10:25:49 | _bilgus | MarcAndersen, on my second buid od the sim with the talk stuff added to the voice menu it is now reproducing the bug |
10:26:10 | _bilgus | just starts talking random entries for each and every menu |
10:28:06 | MarcAndersen | Is it saying a letter after each option, and if yes which one? |
10:28:20 | _bilgus | S |
10:28:48 | _bilgus | but its the spelling thats really screwed up |
10:29:05 | MarcAndersen | That was also what my daily builds from the website did, some of mine said y and some x |
10:30:13 | MarcAndersen | It's good that the sim does it so more people can try. |
10:30:27 | _bilgus | we reproducing it in the sim at least gives me a chance to figure it out |
10:30:33 | _bilgus | well* |
10:52:21 | | Quit woffs (Quit: Gateway shutdown) |
10:59:06 | MarcAndersen | I just build a windows sim and that produced it too |
11:00 |
11:07:30 | | Join woffs [0] (3cd46299b2@woffs.de) |
11:17:39 | speachy | _bilgus: I vaguely recall there being a upper/lower split in lang IDs. |
11:17:53 | _bilgus | yeah 0x8000 |
11:18:18 | _bilgus | it appears to be an overflow atm voice zero plays LANG_CANCEL |
11:18:30 | _bilgus | but it might just be a red herring |
11:19:23 | _bilgus | i'm looking for commits that might have changed the overflow on the id |
11:30:14 | _bilgus | it appears id2index is working right -> get_clip id: 32769 index: 880 talk_id: Say voice clip 0x8001 |
11:32:11 | _bilgus | just saw this message notice: Invalid mode encountered. The stream is corrupted. |
11:32:54 | _bilgus | maybe its a buffer overflow in buf lib I'll try enabling asan |
11:33:21 | _bilgus | probably should have done that first |
11:52:11 | _bilgus | nothing from asan |
11:52:44 | _bilgus | wonder if it could be genlang or the voice clip generator |
12:00 |
12:04:40 | _bilgus | in the hex editor the header count is right hndle table is a little more difficult to decipher |
12:10:46 | *** | Saving seen data "./dancer.seen" |
12:11:21 | _bilgus | the offsets and sizes seem to jive i guess ill check what the offset and size returned by buflib is |
12:12:08 | MarcAndersen | This bug is harder to track down than I thought. |
12:45:07 | _bilgus | well the index has the right offset and size coming back from buflib |
12:45:27 | _bilgus | so I guess tha leads back to the .voicefile |
12:45:44 | _bilgus | garbage in garbage out |
12:46:19 | _bilgus | I guess I do a voice make and check the clips before they are assembled |
13:00 |
13:26:28 | | Quit othello7 (Quit: othello7) |
13:40:20 | _bilgus | ok so our custom speex is putting out the right files |
13:41:25 | _bilgus | the size in the header is different though now I need to see if it strips the file I assume wavtrim was run before encoding |
13:45:53 | _bilgus | hmm wonder if this is normal or an issue mp3 file ./LANG_INSERT_SHUFFLED.mp3 not found! |
13:45:59 | _bilgus | quite a few of those |
13:50:11 | _bilgus | ah its voicefont.c and it appears to be normal |
14:00 |
14:10:48 | *** | Saving seen data "./dancer.seen" |
14:40:09 | _bilgus | Well I figured out the why |
14:43:06 | _bilgus | somehow its picking up IDs > the max regular IDs(LANG + VOICE) instead of labeling them voice only IDs and inserting them into the file stream prior to the voice IDs |
14:46:21 | _bilgus | it goes something like index[879] [1000] .. [1008] [880] [881]..[999] [1009] [1010].. |
14:46:41 | _bilgus | now I just have to figure out the why |
14:50:22 | MarcAndersen | Cool |
14:52:10 | _bilgus | hmm it might have something to do with the not found! messages afterall |
15:00 |
15:08:46 | | Join l0x0l [0] (~l0x0l@097-093-024-013.res.spectrum.com) |
15:13:47 | | Quit MarcAndersen (Read error: Connection reset by peer) |
15:17:19 | | Join MarcAndersen [0] (~no_znepna@93-163-32-211-static.dk.customer.tdc.net) |
15:21:12 | _bilgus | haha buffer overrun |
15:22:22 | _bilgus | https://github.com/Rockbox/rockbox/blob/master/tools/voicefont.c#L54= |
15:22:45 | _bilgus | we hit 1000 entries and then it started eating the rest |
15:28:16 | MarcAndersen | Is that easy to fix? |
15:34:26 | _bilgus | very much so |
15:35:11 | MarcAndersen | But it was hard to track down! |
15:39:20 | _bilgus | yeah I was sure to add some checking and an error code |
15:39:41 | _bilgus | save the next me some tme |
15:39:51 | _bilgus | time |
15:41:09 | MarcAndersen | So was it at my bisect bad commit that it went over 1000? |
15:46:13 | _bilgus | g#5033 |
15:46:16 | rb-bluebot | Gerrit review #5033 at https://gerrit.rockbox.org/r/c/rockbox/+/5033 : [BugFix] voicefont.c buffer overflow by William Wilgus |
15:46:22 | _bilgus | MarcAndersen, Yep |
15:46:37 | _bilgus | well 99 technically |
15:46:40 | _bilgus | 999 |
15:46:46 | MarcAndersen | When will it get into master? |
15:48:00 | rb-bluebot | Build Server message: New build round started. Revision 3694314058, 303 builds, 9 clients. |
15:48:00 | rb-bluebot | [BugFix] voicefont.c buffer overflow by William Wilgus |
15:48:05 | _bilgus | now |
15:48:08 | MarcAndersen | Oh |
15:48:14 | _bilgus | when thats done it should be fixed |
15:48:48 | MarcAndersen | I think I'll wait till the dailys are built and then update all my players |
15:49:15 | MarcAndersen | Thanks so much for the help! |
15:50:13 | MarcAndersen | Let's hope it's green |
15:51:37 | _bilgus | it will be |
15:52:14 | _bilgus | thank you for the bug report we don't use the voice much so things like that might get missed for a while |
15:52:48 | MarcAndersen | That's ok, as I said I learned a lot too! |
15:53:35 | _bilgus | @speachy now we know why that paticular commit caused the bug |
15:56:51 | MarcAndersen | Oh I don't need to wait because the fixed code is in rockbox itself, not the voice file right? |
16:00 |
16:05:51 | rb-bluebot | Build Server message: Build round completed after 1072 seconds. |
16:05:54 | rb-bluebot | Build Server message: Revision 3694314058 result: All green |
16:10:52 | *** | Saving seen data "./dancer.seen" |
16:14:39 | _bilgus | well as long as you generate the voicefile then yes |
16:15:07 | _bilgus | sorry was gerriting? |
16:15:08 | MarcAndersen | Oh I see it's not in rockbox, it's the generator. Then I'll wait. |
16:15:15 | _bilgus | its up |
16:15:48 | MarcAndersen | I know but I'll just wait for the daily voices to be built |
16:30:24 | _bilgus | ah |
16:32:29 | MarcAndersen | That makes it easier to deploy on all my players with rbutil on windows |
16:58:49 | speachy | _bilgus: that's.... really asinine. |
16:58:52 | speachy | good find |
16:59:03 | _bilgus | took literally hours lol |
16:59:26 | speachy | MarcAndersen: yeah, tonight all voices will regenerate and all will be well |
16:59:43 | MarcAndersen | That's excelent |
17:00 |
17:00:18 | speachy | I guess the gigabeast had some target-specific stuff that pushed it over the limit before some of the other devices. my daily drivers are all monochrome which excludes a lot of strings. |
17:00:24 | MarcAndersen | Oh do the voicefont executable get rebuilt before? |
17:01:26 | speachy | it uses the latest git to generate everything nightly, and building the executable is part of that. |
17:01:34 | speachy | technically builds it per-target |
17:01:41 | speachy | (should be identical though) |
17:02:19 | MarcAndersen | But the executable is in /tools, does that get rebuilt as well? |
17:02:24 | speachy | yeah |
17:02:29 | MarcAndersen | Fine |
17:02:35 | speachy | that executable is not part of what runs on the player |
17:02:51 | speachy | you could do a git update, make voice, and then you'd get a fixed voice file. |
17:03:03 | speachy | well, git pull anyway |
17:03:06 | MarcAndersen | I know, that was why I asked if it gets rebuilt before the voices |
17:03:36 | speachy | everything built is done from a completely clean work area; everyhting is regenerated as it's needed. |
17:03:57 | MarcAndersen | Ok |
17:15:26 | _bilgus | that solves weird issues too |
17:16:39 | MarcAndersen | Which issues? |
17:18:56 | _bilgus | build issues |
17:19:31 | _bilgus | thats why when you were doing git bisect I told you to make clean && make reconf |
17:20:22 | MarcAndersen | Oh I thought it was an issue in rockbox, I misunderstood |
17:20:42 | _bilgus | as things change you can have a file already build so the makefile won't rebuild it but there might be something missing or still in it that is no longer there |
17:21:13 | _bilgus | probably better makefiles would help but they are a terrible curse to work on |
18:00 |
18:10:55 | *** | No seen item changed, no save performed. |
18:23:37 | | Quit paulk (Ping timeout: 256 seconds) |
18:25:26 | | Join paulk [0] (~paulk@vpn-0-22.aquilenet.fr) |
18:30:20 | | Join massiveH [0] (~massiveH@2600:4040:a992:a300:dd2e:e4a0:28b:b534) |
18:53:29 | | Quit l0x0l (Changing host) |
18:53:29 | | Join l0x0l [0] (~l0x0l@user/l0x0l) |
20:00 |
20:10:58 | *** | Saving seen data "./dancer.seen" |
20:14:08 | | Quit JanC (Remote host closed the connection) |
20:14:28 | | Join JanC [0] (~janc@user/janc) |
22:00 |
22:11:01 | *** | Saving seen data "./dancer.seen" |
23:00 |
23:36:22 | | Quit massiveH (Quit: Leaving) |
23:37:11 | | Quit m01 (Quit: Konversation terminated.) |
23:39:32 | | Join m01 [0] (~quassel@vps-b172b88b.vps.ovh.net) |