#musicbrainz-devel

The right button is the one with the cloud.

ok, i'll never use THAT service again, most of the links ask me to download an .exe

[all] all are tracks less than 7 minutes long, musical, and took >2minutes to process (hit my timeout)

rochusw: dropbox always seems to be pretty reasonable

hmm

so, ruaok's and rochusw's tracks work here

can you please run with https://www.dropbox.com/s/ck89u66biy9z5ed/strea... this extractor

and see if you get the same error

lol 59mb binary

annoyingly difficult to download from dropbox to a headless server :/

alastairp: that binary isn't compatible with my abzsubmit, but it appears to not hang on my example files

(it's not returning the codec info, which is why it's not working with abzsubmit)

i have my linux running on an older system, can't run the 64 bit binary there (the newer system is running windows for games)

the jenkins build doesn't hang with this file, too

buffer size error is fixed in master branch, do i miss smt or it wasn't (yet) merged in audioloader-codec branch ?

man, working with waf is kind of annoying :/

Wife Acceptance Factor was always hard to work with ;)

kepstin-laptop: yes, I don't care about submitting, we just want to try the extractor

alastairp: yeah, that particular build of the extractor doesn't hang on my test files.

how can i attempt to reproduce that build?

what git rev should I be building?

Master

rochusw: ah, Jenkins is OK? Interesting

rochusw: I think there is a 32 bit build. Let me see

yes, the 32 bit build was not hanging with the file

hmm, yeah, I'm still hitting the same thing with a clean build of master on my system

... how do I tell waf to build with debug info?

I'm currently running that extractor through all of my failed files (not all of them hanging)

I kinda wish this was using a less uncommon build system :/

ah, has to be passed as env variables to configure

kepstin-laptop: --mode=debug also works

kepstin-laptop: ffmeg or libav?

hmm. that peakdetection loop is a bit odd

Ulrich Klauer: MBS-7928: Follow change of search server response format

alastairp: I can try that in a minute. gotta finish dinner first

... 'i' is a signed int ... but that should only be an issue after 811 minutes of audio at 44.1kHz

but yeah, looks like an infinite loop in the peak detection to me :/

I wonder if the code is hitting one of the undefined behaviour on signed integer overflow issues

that could result in different behaviour with different compilers

... that's exactly it

I think.

hmm.

I think that the compiler is assuming that i+1 < size-1 is always true

Yeah, compilers. Sigh

What's your setup?

or no

always false

Os, compiler?

linux, gcc 4.8

if the i+1 < size-1 is always false, then the other part of the array condition is never checked and the loop is never entered

but that makes no sense, hmm.

because i is 0 and size is 1025 in this loop iteration i'm in right now

but it looks like the compiler has completely optimized out the array accesses in this loop

maybe compile the code with opts off and see if that improves?

alastairp: https://gist.github.com/mayhem/98ed9a570835db75...

not, uhm, clear what is going on.

if the code runs with optimizations off but not with them on, that generally means that the code is not correct.

but it didn't hang on the same track.

but might point out the issue yeah

the new build is giving me these worrisome errors:

"[ WARNING ] AudioLoader: Error reading frame: Input/output error"

I don't think that's a problem

dima didn't say anything when it happened when we ran it

but that version works...

this is weird

k.

I'm re-running my test collection with this new binary

hmm. now that i'm looking at the assembly, it is definitely loading values from the array

thanks for being hardcore kepstin-laptop__

ruaok: this is the original binary

also, wow, that's not an array, that's an stl vector

ah, regardless, I'm running my test set.

needs a function call every time it loads values, lol

ruaok: that is, there's no bitrate or codec

and so won't submit

no matter, all the data has already been submitted.

kepstin-laptop__: mmmm, c++

ruaok: ok, no problem, just making sure you're aware :)

understood, thanks.

anything else I can do to help?

save reading assembly?

moose > assembly, no?

ah, no wonder i'm confused, gdb isn't picking up the fact that the 'i' variable is only stored in a register and not in ram in this function

looks like it is getting ingremented at some point.

hmm.

alastairp: the original binary does not seen to lock up

well, let's wait for kepstin to disassemble both binaries and work out what's happening :)

heh. :)

alastairp: replying to an old question, I'll probably add windows binaries at some point, but os x is just not possible with the setup I have

luks: OK, great

I support 10.4 for acoustid and compiling c++11 code in the mac mini would need me to update xcode, which would mess up the whole setup

just letting you know that we already have a bit of the infrasture there

right, that's fair enough

if you want to go for easy deployment, it would be great to replace fftw/resample with ffmpeg and not depend on taglib in abzsubmit

given that essentia already depends on taglib, I guess you could add all the tag reading code there

yeah, I wondered about that.

right, the only reason it's out of the extractor is that I want to check for an mbid before we run the extractor

but of course we could add a check to the extractor to exit early if there's no tag

and it would be also great to stop using that giant hack in RogueVector :)

so it looks like the compiler's optimized out the check on line 132, resultPos > _maxPos

and it's basically looping forever with i=j=0

luks: yeah, everyone wants to fix rougevector

after the million other things

* nikki: Order releases table on /cdtoc/ pages like on other pages which list releases

* Ulrich Klauer: Implement some core entity model methods via %ENTITIES

* Ulrich Klauer: Use entity property mbid/multiple for statistics

* Ulrich Klauer: Don't create entities that are then not used

but it's not running the peaks.push_back line either as far as I can tell.

* Ulrich Klauer: Remove some unnecessary imports

* nikki: Escape the no AcoustIDs message on the fingerprint tab (MBS-7927)

* Ulrich Klauer: MBS-7928: Follow change of search server response format

or else I'm still horribly confued

I should try compiling it without optimization, i guess.

naw, without opts is for wimps

-->out. keep up the good work all

well, looks like it still hangs with optimization disabled. so that's something.

hmm, never mind

it's just really really slow

and it does not fail

any particular reason you guys chose pytaglib over mutagen for abz client?

you can't use mutagen in a gpl3 application

I don't know if that was the reason, but it's a reason

LordSputnik: no

And honestly, I would have preferred a mit-ish license, but the department prefers GPL and I didn't see the taglib license before I used it

As I've alluded to before, this use of taglib is a hack anyway, and we should replace it with something that doesn't require an additional dependency

I see... anyway, I'm pretty sure I should be submitting stuff now, the client is certainly making the right noises

Does it keep a list of already processed paths somewhere?

<wa

yes

In an SQL db in .abzsubmit

which debian package contains this parallel-util?

rochusw: looks like it's in 'moreutils'

what a horrible package name.

there's a gnu version too, but I dunno which debian versions have it

package name of that is just 'parallel'

if you have a new enough debian to run the acousticbrainz stuff, you probably have gnu parallel