#musicbrainz

much better than my 4-nested-subqueries monstrosity :D

max(id) is cheating, technically

but we don't reuse ids, so it should work

I was already trying to use max(id) anyway :)

mezood is now known as mezod

Does anyone know if what I want is doable with Lucene?

intgr: yes, it's definitely doable

probably a little too expensive with the current design of the search server

Do you need to add new indexes for that though?

no

Ah ok

./c

you can search on multiple fields

Or go a step further and implement "free form search" that can search for any object.

even give them different weights

yeah, that's what I intend to do eventually

I might have the time to hack on MB now; dunno if I can overcome my "inertia" though. :)

Cool

intgr: get over it!

we need more hackers :D

the search server design right now is not exactly "scalable"

we use different index for every search type

Well, you just pop up more boxes if you want to scale, no?

yes, but to add a new search you have to add a new index

Ah

much of the data could be shared between indexes

I'm sure lucene takes that into account

but when they are on different indexes, it can't do anything

Ah ok

so the "free form search" could be technically faster

Another question I have; what does the Qt threadpool stuff do in Picard's OFA scan code? It sure as hell isn't parallelizing fingerprinting.

there is only one fingerprinting thread

that's a limitation of libofa

I was thinking of parallelizing it (as I bought a new shiny computer recently ;) but didn't dare to touch it.

Screw threads, use processes.

subprocess is portable enough

Probably more portable than libofa anyway.

pronik`: let me know when you get back

I have some code written for using processes

never finished it though

Ok

So it's fine if I rip out the threadpool code?

well, it's used for other things too

Hmm

Ah I see.

moving all of to a subprocess would be much better

and having configurable number of fingeprinting processes

So I would start with botching picard.util.queue.Queue?

Or do you want to have the other queues as threads?

you still need queues

I'm not sure if picard.util.queue.Queue does something special right now though

By "botching" I meant "changing"

Looks like a wrapper to me

well, I'd start by writing the subprocess

Scratch everything; where's your subprocess code?

I'll see what you did there

one moment, I'll try to find it

Just wondering, how do you feel about hg given that Picard uses bzr?

I've never used it much

Nice, Wall Street Journal want to do an article including MusicBrainz :)

:)

I think I'll leave that phone call to rob though :)

if you want my general opinion on DVCS then: bzr ~ git > hg

I don't see any advantage of using hg

(over bzr)

What advantage do you see in bzr over hg?

I like Mercurial, that was my first DVCS

but now I wouldn't use anything but Git, heh

well, first of all I have my tools for bzr

Fair enough

hg seems like a mix of different ideas to me

I've started using bzr when it was the only dvcs usable on windows

How do you mean "a mix"?

the definition of branch and repository in hg is very weird

Huh, I just realised I can write our merge functions with less duplication by making use of the fact that aliases/annotations are roles

before 'merge' => sub { $self->merge_aliases() } seems good

luks: Never really used named branches in hg; I'm afraid of branching in git as I was with svn ;)

I couldn't live without local branches anymore

But I haven't used git much tbh.

Especially not now we do code reviews at musicbrainz

I'm using bzr the git way

I use mq for most of my still-in-churn development.

so I have a number of branches without trees (since branches are directories in bzr)

use a single working tree, and switch between those branches

aCiD2: I can use tab completion for branch names, can you? :)

Yup :)

cool

git-bash.sh :D

it shows which branch i'm on in PS1 too, which is handy

Anyway enough about version control masturbation. :)

it's been discussed to death here :P

but I'm glad we finally have a system that lets anyone use what they want

Heh, it's the hot topic these days.

intgr: well, I can't find the code

hg-git <3

but there wasn't much anyway

luks: What's the right approach to this?

Do something like pp does -- send pickled input to subprocess and get pickled results?

yes

Do you want to keep the HTTP queues as threads? (which makes more sense to me)

the transport layer should be QProcess, ideally

because it allows async calls

HTTP queues are not threads, are they?

but HTTP should be rewritten in QNetworkAccessManager anyway

Ah they aren't?

I see

the way I personally see it working is that the fingerprinting subprocess will respont to two requests

hm

actually only one

you give it a file name, it decodes it, calculates the fingerprint and returns the result

Right.

once you have the fingerprint, you will do HTTP lookups

nothing changes there

Yeah

What's load_queue, save_queue and other_queue?

Reading/writing files?

yep

other_queue is for CD lookups and directory listing

Right, that shouldn't be parallelized I guess.

yes

luks: Any source code hosts besides Sourceforge and Launchpad that support bzr?

Savannah, and standard web hosting

*any

google code is a huge plus for hg

but then, hosting bzr on a svn server is trivial :)

Google Code is pretty pale compared to Bitbucket IMO

luks: I thought about the work we're doing merging edit_* tables

I don't know, I somehow never liked Bitbucket and Github

surely that's not quite enough, wouldn't we need to actually update the entity ids in the edit as well?

I don't like Github either.

Nor Launchpad

aCiD2: well, that's questionable

Ie, I enter an edit artist edit for artist id = 5, then merge artist 5 into artist 4. Now artist 4 has an edit_artist pointing to an edit on artist 5, which no longer exists

aCiD2: but I'd consider it historical information

so it doesn't really make any sense

The other day I just spent 10 minutes figuring out how can I list the bugs I've opened on launchpad. Then spent 10 minutes figuring out how to navigate back to the place where I can report a bug :(

that's what the current code does

and it's really confusing

lol

hrm

I don't know

but "Merge artist: A and A" seem strange to me