14:36:33 <joanie> #startmeeting
14:36:33 <tota11y> Meeting started Thu Jun 30 14:36:33 2011 UTC.  The chair is joanie. Information about MeetBot at http://wiki.debian.org/MeetBot.
14:36:33 <tota11y> Useful Commands: #action #agreed #help #info #idea #link #topic.
14:36:53 <joanie> #topic 1.1 Call for volunteers: Quantifying improvement from Mike's change to only emit events if an AT is listening.
14:37:16 <joanie> #info Mike has made the changes needed to only emit events if an AT is listening.
14:37:55 <joanie> #info This change might in theory solve much of the broader community's concerns about enabling AT support by default due to performance problems.
14:38:22 <joanie> #info It would be ideal if someone could verify this in some proven, respectable, convincing, with numbersy sort of way.
14:38:27 * joanie opens floor up
14:38:31 <bnitz> It sounds like a great fix.  Are you looking for quantifying some of the performance improvements?
14:38:54 <bnitz> Thanks you answered my questions and type faster than me.
14:39:06 <joanie> I would like to have data that says "There is no performance hit whatsoever if AT is enabled by default but no ATs are listening."
14:39:15 <joanie> So how can we get this data?
14:40:17 * joanie reminds everyone that she is a teacher and she will start calling on people. :-P
14:40:26 <jhernandez> xDD
14:40:45 <bnitz> joanie:I wonder if some of the existing mago tests could verify this.
14:40:46 <jhernandez> gtk-perf could be usefull for to check the performance
14:41:04 <jhernandez> isn't it?
14:41:09 <joanie> bnitz: you first than jhernandez. Please elaborate
14:41:10 <bnitz> jhernandez:I thought that might be at too low of a level.
14:41:44 <bnitz> The mago test community maintains some scripts which do things for various applications.
14:42:03 <bnitz> I think it's well maintained and works well for ubuntu, less so for other distros.
14:42:09 * bnitz looks for link.
14:42:27 <joanie> jhernandez: While bnitz is looking for a link....
14:42:31 <fregl> it would also be interesting to measure memory in addition to cpu - that is doable with valgrind --tool=callgrind
14:42:52 <joanie> do you have knowledge about gtk-perf sufficient to write such a test?
14:43:05 <joanie> fregl: makes sense
14:43:07 <jhernandez> joanie: no i don't
14:43:49 <joanie> fregl: could valgrind be *the* tool (memory+cpu)?
14:44:04 <bnitz> http://bazaar.launchpad.net/~mago-contributors/mago/mago-testsuite/files
14:44:10 <joanie> thanks bnitz
14:44:42 <jhernandez> i know about gtk-perf and since the main problem is the performance, I was suggesting gtk-perf as a quick/easy/first approachment
14:44:45 <fregl> joanie: memory definitively. cpu maybe with a different tool but rather not.
14:45:18 <joanie> fregl: you know enough about valgrind to create the memory test?
14:45:53 <fregl> sorry, but I don't have time nor enthusiasm. but if someone wants to run this I can tell the command line basics how to do this.
14:46:00 <bnitz> Any BSD people here?  I have some dtrace based launch performance scripts, they look for when an application maps to the X server.
14:46:08 <joanie> fregl: understood. And thanks.
14:46:51 <joanie> As much as I love dtrace, the "typical" gnome environments now are Fedora 15 and Ubuntu.
14:47:34 <mgorse> I don't think it's that hard to figure out valgrind--it has a manpage, anyway--I think the hard part will be figuring out what to test
14:47:36 <bnitz> This patch looks cool  https://bugzilla.gnome.org/show_bug.cgi?id=618189   A facility for scripting the (gnome)-shell to generate performance events, and    then replaying the log to collect "metrics"
14:47:36 <tota11y> 04Bug 618189: normal, Normal, ---, gnome-shell-maint, RESOLVED FIXED, Framework for performance measurement
14:47:42 <clown> backing up a bit for clarification:  improving performance implies a comparison -- is the comparison between (1) AT enabled by default vs. (2) AT enabled only when at least one AT is listening?  Or (1) no AT enabled vs (2) AT enabled when one AT is listening?  Or both?
14:47:51 <bnitz> great minds...
14:48:08 <bnitz> tota11y I think this could be the most useful/easy to use.
14:48:08 <tota11y> bnitz: Error: "I" is not a valid command.
14:48:31 <clown> note that ubuntu doesn't use gnome-shell..
14:48:45 <joanie> First and foremost, clown, I think we want to compare: AT support disabled and AT support enabled but no ATs listening
14:48:52 <joanie> clown: oneiric has it
14:49:16 <joanie> but to back up for everyone for clarification
14:49:34 <joanie> the broader (not us) GNOME community (including Release Team, etc.) have the following claim:
14:49:53 <fregl> I stand corrected, collegues tell me valgrind is supposed to be good for cpu testing also
14:49:54 <joanie> "We cannot enable AT support by default because it drags the desktop to a halt even when no ATs are being used."
14:50:15 <joanie> I would like to make a case that boils down to "Your assertion is now bogus." ;-)
14:50:33 <joanie> But we need to know if it is indeed bogus before we make our own bogus claims. :-)
14:50:40 <joanie> mgorse did the work
14:50:46 <joanie> Now we need the testing and data
14:50:52 <joanie> fregl: cool, thanks for the update
14:51:14 <joanie> So valgrind sounds like it might be the tool and mgorse doesn't think it would be that hard to learn
14:51:33 <bnitz> As much as I like valgrind, I don't think this can be our one tool, it may give us useful bits of information though.  CPU+memory doesn't necessarily correlate to user-perceived performance.
14:51:55 <joanie> bnitz: fair enough
14:52:07 <joanie> how do you propose we measure user-perceived performance?
14:52:09 <fregl> bnitz: it's about numbers not perception I thought
14:52:11 <joanie> other than by user testing?
14:52:32 <TheMuso> p/c
14:52:47 <joanie> If we were to have the numbers, we could go to the RT and say "look, all fixed!"
14:52:53 <joanie> and they could then see what they perceive
14:52:55 <bnitz> #info what about ShellPerfLog (the tool I mentioned ealier) https://bugzilla.gnome.org/show_bug.cgi?id=618189
14:52:55 <tota11y> 04Bug 618189: normal, Normal, ---, gnome-shell-maint, RESOLVED FIXED, Framework for performance measurement
14:52:56 <korn> For Joanie's assertion, wouldn't numbers actually be the better thing?
14:53:46 <joanie> Well, that tool has Owen's name all over it. So perhaps valgrind plus that.
14:53:49 <joanie> ;-)
14:54:00 <joanie> Anyhoo I don't want to spend the whole meeting on this topic.
14:54:07 <joanie> mgorse: This is your fix we're testing and you know what we're looking for. What's your time these days like?
14:54:19 <joanie> i.e. could you look into valgrind
14:54:21 <clown> numbers are a good thing.  still, note that they will probably show some greater use of resources -- for what value do you say that this greater use == "drags the desktop to a halt"?
14:54:23 <joanie> to get us these numbers
14:54:36 <bnitz> joanie:One step down from manual testing with a stopwatch.  Automate launch and other user activity.  Timestamp
14:54:48 <bnitz> Look at times between timestamps.
14:54:54 <joanie> bnitz: we could do that
14:55:06 <joanie> At this point I need volunteers. :-)
14:55:19 <joanie> Who can do what -- not just in spirit and knowledge, but in time?
14:56:03 <mgorse> I think I should write to some list (desktop-devel maybe), explain the change, and ask if people have things in particular that seem slow / that should be tested
14:56:31 <clown> mgorse:  +1
14:56:49 <bnitz> I can investigate the options but I don't have a good/popular test execution environment
14:56:52 <clown> of course, they might answer :  "the desktop is slow".
14:56:54 <joanie> We know the culprits though: Trees :-)
14:56:59 <joanie> clown: yup
14:57:21 <|Lupin|> hello, there
14:57:23 <bnitz> Unless our A11yTest distro running in a virtual box displayed via Sunray will be our gold standard ;-)
14:57:34 <joanie> hahaha
14:57:36 <joanie> anyhoo let's move on
14:57:43 <clown> bnitz:  fabulous benchmark, that!
14:57:56 <bnitz> yep
14:58:32 <joanie> #info Valgrind might be a good solution. The tool for gnome-shell from bug 618189 might as well. Lastly, just timestamping typical activities.
14:58:32 <tota11y> 04Bug https://bugzilla.gnome.org/show_bug.cgi?id=618189 normal, Normal, ---, gnome-shell-maint, RESOLVED FIXED, Framework for performance measurement
14:58:37 <clown> must run that distro from a 1x CD drive.
14:58:49 <joanie> #info No one at the meeting could volunteer their time for these tasks.
14:58:54 <mgorse> I can probably do some kind of testing as long as it doesn't require me to actually read the screen without Orca running :) In that case I'd need to figure out what needs doing and could come up with a plan at least
14:59:18 <joanie> #action Joanie will investigate what other resources we have/can draw upon to achieve this work.
14:59:28 <joanie> mgorse: perhaps we can tag team it
14:59:32 <joanie> I dunno valgrind
14:59:45 <bnitz> mgorse: didn't you have a basic pyatspi test, give me all children of this accessible 1000 times or something like that?
14:59:45 <joanie> but if you get me up to speed and provide a test which theoretically should work, I can run it.
14:59:50 <clown> any idea what the learning curve is for valgrind?
14:59:58 <bnitz> Maybe too far from the real world use.
15:00:29 <mgorse> bnitz: I don't think I've do at the moment; I've written things like that before but wouldn't help with what we're trying to do now
15:00:32 <joanie> anyhoo, 25 minutes into this meeting, I'd like to move on.
15:00:35 <joanie> sorry!!
15:00:49 <joanie> #topic 1.2 call for volunteers: Writing tests for the Gail into Gtk merge.
15:01:07 <bnitz> # Action brian to investigate gnome-shell --perf=<script>
15:01:28 <joanie> #info The Gail to Gtk integration is moving along pretty rapidly and there is talk about integrating the 'a11y' gtk branch into master soon.
15:01:46 <joanie> #info We need more tests written, and we need the output of the existing tests verified.
15:02:07 <joanie> #info Joanie has looked through the existing tests and is starting to write new ones. But other help would be greatly appreciated.
15:02:13 * joanie yields floor
15:02:23 <mgorse> These tests are glade files, right?
15:02:24 <clown> joanie:  are these unit tests?
15:02:53 <joanie> These are the accessibility-dump tests Benjamin (company) talked about last week.
15:03:32 <clown> which examined a glade file, and dumped out the a11y tree?
15:03:33 <joanie> mgorse: but can be written in text editors if need be. They are pretty simple files if you look.
15:03:41 <joanie> clown: yup
15:04:14 <clown> is it possible to create a dump of what should be output to compare with the actual dump?
15:04:40 <joanie> What I have done is generated the output using Gtk+ 3 master (rather than the a11y branch)
15:04:59 <joanie> so that when they test it with the a11y branch, differences will be caught
15:05:11 <joanie> and then proofed the text output for a sanity check
15:05:18 <clown> what does "test it with the a11y branch" mean?
15:05:20 <joanie> looking for anything obviously borked
15:05:45 <TheMuso> There is a branch of gtk in git where the gail merge work is being done.
15:05:48 <mgorse> Where are these tests? Are they in the gtk+ repository?
15:05:54 <joanie> well, the a11y branch of gtk+ is where Benjamin and Matthias are moving all of gail into Gtk itself
15:06:11 <joanie> mgorse: they are in the gtk+ repository under the a11y branch
15:06:18 <joanie> in gtk+/tests/a11y
15:06:27 <mgorse> thanks
15:06:43 <joanie> anyhoo, we need to move on. But please everyone, write tests, read tests, etc.
15:06:50 <joanie> this merge is going to happen
15:07:18 <joanie> #topic 1.3 Call for volunteers: Best practices guide for implementing accessible objects.
15:07:40 <joanie> #info Joanie and Frederik are going to resume working on this task from the ATK hackfest
15:08:08 <joanie> #info The Gail into Gtk work is including some "best practices" style changes that are not documented.
15:08:34 <joanie> #info We really need to get a "best practices" guide on the object hierarchy, roles, properties, and such out.
15:08:50 <joanie> #info Joanie and Frederik would not mind volunteers on this front. :-)
15:08:52 <fregl> clown: for valgrind there is no real learning curve. it is  a one line command that dumps lots of data into files. then you need a tool to visualize the result.
15:09:12 <clown> fregl:  thanks.
15:09:13 <bnitz> joanie:Are these new tests?
15:09:22 <aleiva> in fact is something like a format
15:09:23 <joanie> bnitz: which tests?
15:09:46 <aleiva> but why we need a C debugger?
15:09:53 * clown it's a volunteer meeting...
15:09:55 <aleiva> maybe i am late
15:09:57 <bnitz> joanie: gtk+/tests/a11y
15:09:57 * joanie laughs
15:10:04 <joanie> bnitz: See http://live.gnome.org/Accessibility/Minutes/20110623
15:10:20 <joanie> clown: this is why I want to move through this stuff.
15:10:22 <joanie> BUT
15:10:38 <joanie> the thing is, we have a crapload of stuff that needs doing
15:10:45 <aleiva> sure
15:10:46 <joanie> and we need folks who can pitch in to pitch in
15:11:13 <joanie> So I'm asking, nay begging
15:11:25 <clown> you might try nagging...
15:11:34 <joanie> that's next
15:11:36 <joanie> ;-)
15:11:40 <aleiva> well nagging is not an option
15:11:46 <aleiva> IMHO
15:11:49 <joanie> aleiva: Says who?
15:11:53 <joanie> :-P
15:12:09 <aleiva> so we need to assign task and get it done
15:12:17 <joanie> aleiva: Yep
15:12:17 <clown> how about "gentle encouragement"?
15:12:28 <aleiva> clown, that sounds good
15:12:30 <joanie> aleiva: Can we put you down for any of these tasks?
15:12:39 <aleiva> clown do you have some options
15:12:50 <aleiva> joanie dunno
15:12:58 <aleiva> Im late, sorry
15:12:59 <clown> aleiva:  not sure what you mean by "options"
15:13:12 <aleiva> clown possibilities
15:13:20 <aleiva> people willing to work
15:14:01 <clown> aleiva:  I'm willing to work -- that's not the problem.  the problem is the myriad other things I have to do.  So ...  I'll think about it, and try to fit it in where I can.
15:14:26 <aleiva> clown: I understand you very well
15:14:53 <aleiva> what I was saying is about other options
15:14:55 <joanie> Well, the last item I won't even bother with as a topic as we don't have volunteers for anythign more pressing
15:15:02 <aleiva> yup
15:15:04 <joanie> the wiki and docs, etc. bite
15:15:10 <joanie> anyhoo.... moving on
15:15:18 <joanie> #topic GAIL treeview children / children-changed signals with no child argument
15:15:25 <joanie> mgorse: you added that I believe
15:15:27 <joanie> floor is yours
15:15:31 <mgorse> yeah
15:15:50 <alibezz> joanie I didn`t candidate for more pressing things because I lack experience
15:16:00 <mgorse> Currently, it can be expensive to ref a child of a tree view in gail
15:16:09 <alibezz> but if I can help with the wiki/guide, I`m willing for that
15:16:21 <aleiva> nice
15:16:24 <joanie> alibezz: Ok, let's talk after the meeting
15:16:26 <joanie> thanks!
15:16:40 <mgorse> and, when a child is added to a tree view, it sends a children-changed event with no detail for the child
15:17:03 <joanie> mgorse: ah, this is what we were talking about last night, right?
15:17:15 <mgorse> joanie: yeah
15:17:43 <mgorse> Currently, if atk-bridge sees a NULL child for a children-changed event, it calls atk_object_ref_accessible_child to fetch the child
15:18:09 <mgorse> A while ago, Li identified this as causing a performance issue for tree views, and I thought that I'd removed that behavior in AT-SPI2, but apparently I hadn't
15:18:49 <mgorse> but it can be expensive particularly if a lot of rows are added, and so children-changed events keep getting fired for new cells, and gail needs to keep crawling through the tree to fetch the given child
15:19:18 <joanie> mgorse: yup. Can you remove that call? Or is something preventing you from doing so?
15:19:56 <mgorse> joanie: I'd like to do that, but, after talking to you, I'm realizing that some things would need to be tested after doing it
15:20:07 <bnitz> mgorse:Was that your example with spreadsheets as well or is that a different bug.
15:20:21 <mgorse> ie, Gecko and Impress, I guess
15:20:33 <joanie> Oh right....
15:20:47 <joanie> But would the removal be just for trees/tables?
15:21:01 <mgorse> bnitz: If you mean the LibreOffice crash, then that's a different bug
15:21:18 <bnitz> ok
15:22:34 <joanie> mgorse: If it's just for trees/tables it might be safe
15:22:45 <mgorse> I'd like to just remove the call and only send the child if the atk implementor provides it. I'm guessing that in practice making that change should only affect trees/tables, but would need to test that, since it's possible that some other toolkit is doing something I wasn't expecting and atk-bridge has been working around it
15:22:46 <joanie> and is there some way to do it "lazily"?
15:23:45 <joanie> Well, trees and tables are also amongst the chief "a11y sucks" complaints we hear. So perhaps it would make sense to open a bug, attach a patch, and recruit testers?
15:23:58 <fregl> this boils down to actually having defined what should happen when a tree changes again
15:24:38 <mgorse> That's an interesting question, whether it could somehow be done lazily. I'd need to think about that. Anyway, filing a bug seems like a good idea to me
15:24:52 <joanie> mgorse: okay.
15:25:01 <joanie> mgorse: could you #info this for the minutes?
15:25:45 <mgorse> #info Currently, atk-bridge calls atk_object_ref_accessible_child for a children-changed event sent without a child
15:26:14 <mgorse> #info This may be a significant performance hit, and may be better to remove it. Will probably mainly affect gail trees, but need to test that it doesn't affect other things as well.
15:26:33 <fregl> I would love to have a signal to say: this tree has become invalid. sending all changes seems quite big. In iaccessible2 there is TableModelChanged which signals changes. then the client can ask for the details by calling some other function
15:26:37 <mgorse> #action mgorse will file a bug to remove the check, attach a patch, and call for testers
15:26:53 <joanie> Thanks mgorse!
15:27:07 <mgorse> and, yeah, I think it gets into what best practices should be for trees
15:27:18 <joanie> fregl: I believe we already have a bug for that too
15:27:23 <fregl> ok
15:27:33 <joanie> I'll look momentarily
15:27:40 <fregl> I need to leave, damn, this is really interesting
15:27:45 <joanie> Given the time and who is here today....
15:27:49 <fregl> I'll read the backlog laters
15:27:51 <joanie> I'm skipping a topic
15:27:57 <joanie> #topic Action item updates
15:28:03 * joanie calls on people :-)
15:28:13 <joanie> jhernandez: test distro final spin?
15:28:18 <joanie> (#info it)
15:28:58 <jhernandez> ok
15:29:42 <jhernandez> #info at this moment I'm syncing and rebuilding with latest versions. I hope to get some time for to build the spin this weekend or next week
15:29:57 <bnitz> jhernandez final?  That's good news for the tinderbox I'm basing off of it, but how is anything GNOME 3.?? ever final?
15:30:22 <jhernandez> bnitz: final only for a11y testing pursposes
15:30:28 <jhernandez> *purposes
15:30:56 <clown> jhernandez: did you get my email about the mag prefs dialog possibly having problems?  I'd like to try out the spin to check that before we publish it.
15:30:59 <bnitz> jhernandez:To be replaced/superceeded by?
15:31:12 <jhernandez> I'm going to work in another testing spins for the 3.2 series, but not only for a11y
15:31:39 <bnitz> A general purpose GNOME 3.2 test distro?  Sounds reasonable as long as it does A11y.
15:31:40 <jhernandez> clown: sure! I'll pass you the link before publish it ;)
15:31:48 <clown> jhernandez:  thanks.
15:31:51 <jhernandez> bnitz: yep
15:31:53 <bnitz> OpenSuSE based?
15:32:17 <jhernandez> bnitz: at this moment is the easiest one, but opensuse's not my favourite distro :P
15:32:18 <joanie> bnitz: you have a couple of action items yourself ;-)
15:32:23 <joanie> Send a mail summarizing his proposal regarding a testing-centered design.
15:32:31 <joanie> Announce his Mago tests on the gnome-accessibility-devel list, and also create a README and INSTALL.
15:32:34 <joanie> any updates
15:32:35 <joanie> ?
15:32:40 <bnitz> Yes I do and they're still open.  Sorry.
15:32:49 <joanie> That's ok. just checking
15:32:54 <joanie> clown:
15:32:59 <joanie> Investigate if it is feasible to call libatspi using javascript (ie: on gnome-shell)
15:33:21 <clown> joanie:  as far as I can tell, there are no JS bindings to atspi
15:33:38 <joanie> silly question: Can this be introspected?
15:33:55 <clown> there are some for Atk, still investigating the extent of those bindings.
15:34:31 <clown> re "introspected":  not sure.
15:35:02 <mgorse> clown: Do the atk bindings you mentioned involve introspection?
15:35:18 <clown> I assume so -- how can I tell?
15:35:45 <clown> all I do is "const Atk = imports.gi.Atk"
15:36:02 <clown> and voila, there is a JS object for the Atk namespace.
15:36:19 <mgorse> Okay; looks like it then. Does imports.gi.Atspi not work?
15:36:32 <clown> but I can't find a similar incantation  for Atspi
15:36:46 <clown> mgorse:  let me check right now.
15:37:01 <joanie> so while clown is checking, and given the time.
15:37:03 <TheMuso> Why would you want to use at-spi from within gnome-shell?
15:37:13 <joanie> #topic Miscellaneous Time
15:37:16 <TheMuso> actually, donn't mind me, just worked it out
15:37:24 <clown> !  that worked.  at least now complaints about not finding it.
15:37:24 <tota11y> clown: Error: "that" is not a valid command.
15:37:31 <joanie> lol
15:37:35 * clown thatnks tota11y
15:37:46 <joanie> I need to smartify the bot
15:37:54 <alibezz> I have a misc news ?)
15:38:02 <joanie> alibezz: go for it
15:38:04 <alibezz> two days ago, I gave an interview for Og Maciel
15:38:07 <clown> mgorse:  much thanks for that suggestion.  I'll explore further...
15:38:15 <alibezz> it was for a podcast in portuguese
15:38:33 <alibezz> and it was mainly about my involvement with gnome and accerciser, in particula
15:38:42 <alibezz> I spent a good time explaining the tool
15:38:47 <alibezz> and why it is useful
15:38:55 <alibezz> and how I came up to work with it
15:39:04 <alibezz> it`ll be online in a month or so here
15:39:09 * alibezz looks for link...
15:39:16 <alibezz> http://www.castalio.info/
15:39:25 <bnitz> Thanks alibezz, that sounds great!
15:39:31 <joanie> very cool alibezz!
15:39:47 <alibezz> another news is that, yesterday
15:40:07 <alibezz> a friend of mine who organizes a free software event called ensol
15:40:23 <alibezz> asked me if I were interested in talking about Accerciser
15:40:24 <jhernandez> alibezz: you rocks!!!
15:40:28 <alibezz> and accessibility
15:40:30 <alibezz> in august
15:40:37 <clown> #info there is a way to import atspi into JavaScript (gjs), namely "imports.gi.Atspi".
15:40:42 <jhernandez> i saw your tweet about the podcast
15:40:44 <alibezz> it`s not set already, but it`s quite likely
15:40:48 <alibezz> xD
15:41:12 <alibezz> thanks! It`s awesome to talk about accerciser
15:41:21 <alibezz> accessibility isn`t a very difunded topic in Brazil
15:41:37 <joanie> anyhoo, anyone else?
15:41:45 <joanie> or shall I bang the gavel to end the meeting?
15:41:46 <|Lupin|> pls
15:41:49 <|Lupin|> just one quick thing
15:41:55 <|Lupin|> (hello, everybody)
15:41:56 <alibezz> joanie, should I #info that?
15:42:03 <joanie> alibezz: please :-)
15:42:19 <|Lupin|> may I ?
15:43:03 <alibezz> #info alibezz gave an interview for Og Macial about her involvement with GNOME and specially with Accerciser. It`s going to be online in a month or so here http://www.castalio.info/
15:43:12 <alibezz> s/Macial/Maciel
15:43:29 * alibezz waves to |Lupin|
15:43:37 <|Lupin|> alibezz: thx
15:43:43 <|Lupin|> well right after the meeting
15:43:53 <|Lupin|> I'd like to show you a very old bug
15:43:58 <|Lupin|> on #a11y
15:44:10 <|Lupin|> so it wold be great if you gys could take a few minutes to have a look to it.
15:44:30 <joanie> and with that, we're over time
15:44:33 <clown> #action Joseph to explore the JavaScript Atspi import further to see how much is available.
15:44:33 <joanie> #endmeeting