/srv/irclogs.ubuntu.com/2015/09/22/#ubuntu-unity.txt

=== maclin1 is now known as maclin
=== maclin1 is now known as maclin
* guest42315 sudo snappy install coffee07:05
Saviqtsdgeos, hey, got a sec? need a sounding board08:23
tsdgeosSaviq: sure08:24
Saviqtsdgeos, last issue with autopkgtests08:24
Saviqtsdgeos, paths.h is generated on build, but when testing archive/ppa-built packages, the non-installed paths there are not the same as what you get in the testing environment08:25
Saviqe.g. when the packages were built08:25
Saviqit was file:///build/unity8-aDQvXH/08:26
Saviqbut when the tests are ran08:26
Saviqit's file:///tmp/adt-run.jl4VOK/build.64t/ for example08:26
Saviqbasically, our binaries get some random paths baked in08:26
tsdgeosright08:27
tsdgeoshmmm08:28
tsdgeosno, not right08:28
tsdgeoswhy does this happen?08:28
tsdgeosdoes adt copy the things around?08:28
Saviqthis actually makes ApplicationWindow test fail under adt-run *if* you don't build the packages at the same time (e.g. what you should do - take the packages as built in the PPA, not build them locally)08:29
Saviqtsdgeos, PPA builds the packages, the build path on the builders is /build/foo08:29
Saviqtsdgeos, then, adt takes those packages as they are, builds just the missing mocks and runs the tests, but the build path is different - /tmp/adt-foo08:30
tsdgeosah so there's two builds08:30
Saviqin effect, yes08:30
Saviqbut even if there were not08:31
Saviqthe path baked into paths.h on the builders08:31
Saviqare invalid as soon as the build is done08:31
Saviqat that point paths.h is only useful when running installed08:32
Saviq== running unity8 itself08:32
Saviqso, short term I could put the graphics from qml/graphics/applicationIcons and qml/Dash/graphics/{phone,tablet} in qrc08:34
Saviqthey don't belong where they are any more anyway08:34
Saviqbut that still means paths.h is useless non-installed08:35
Saviq(btw this worked for adt-run when you did build packages within adt-run because the build dir was reused)08:35
Saviqthe real fix would be to runtime-detect our build root08:36
tsdgeosyep08:38
tsdgeosmake it be XDG_SOMETHING08:38
Saviqwdym08:41
Saviq?08:41
tsdgeosso there's all these XDG_ variables08:41
tsdgeoswe can just assume our stuff is asumed in XDG_something/path08:42
tsdgeosand then set XDG_ accordingly for the tests?08:42
tsdgeosnot sure how viable that would be08:42
tsdgeoshow did you plan to detect the root?08:42
tsdgeospwd or something?08:43
Saviqwe'd need to up from pwd until reached build root08:43
Saviq+go08:43
Saviqbut that sounds nasty08:44
tsdgeoswould it even work?08:44
tsdgeosi mean in paths.h we have08:44
tsdgeos        return QString("@CMAKE_BINARY_DIR@/po/locale");08:44
tsdgeosand08:44
tsdgeos        return QString("@CMAKE_SOURCE_DIR@/qml/");08:44
Saviqright08:45
tsdgeoshow do we find those if we have two build roots?08:45
tsdgeosis the source even around?08:45
Saviqyeah, source is always around08:45
Saviqbut if we start requiring some env vars to be set, even more meh08:46
Saviqwe're not really interested about "previous" CMAKE_BINARY_DIR08:47
Saviqbecause it's invalid08:47
Saviqwe only care about current paths08:47
SaviqI can't shake the feeling that we're either doing something wrong or that someone must've solved this already...08:48
Saviqobviously the fact that QML makes relative paths nice and easy doesn't help08:48
Saviqbecause that only works until you start passing paths to C++08:49
* Saviq tempted to start moving all that stuff to qrc, only reliable way08:49
Saviqnot that it really solves the issue08:49
Saviqon that note, can you use qrc in qml-only files? how do you tell Qt where the resources are?08:50
Saviqmzanetti, ideas? (read up to ~10:23 ↑)08:56
* mzanetti reading08:57
mzanetti+1 for runtime detection... /me never liked the paths.h09:01
mzanettinow... I don't have the solution obviously09:01
* mzanetti thinks09:01
Saviqthe only thing I could think of is putting a special file in project root and go up from $CWD to find it09:03
Saviqbut that seems meh09:03
Saviqwe could have paths.h as a library and LD_PRELOAD it... that one seems overkill09:06
mzanettiSaviq, well, putting a special file is meh indeed, but why not trying to find the actual files?09:06
Saviqmzanetti, as meh as a special file I'd say09:07
mzanettislightly less meh09:07
mzanettinot super pretty still, yes09:07
Saviqwith the added problem that you might want to move or rename the non-special file09:07
mzanettiyeah well... the test would catch that pretty fast ;)09:07
Saviqsure, but I'd rather be explicit about it09:08
SaviqI just feel like this must be a solved problem09:09
Saviqcan't believe we're the first ones to ancounter it09:09
Saviq*encounter09:09
mzanettiSaviq, doesn't dpkg export some build_root var or something?09:11
Saviqmzanetti, even if, only useful if building a package09:12
mzanettiisn't that what we're talking about?09:12
Saviqmzanetti, the problem we're facing is running tests09:12
mzanettiyes, afaiu it's still dpkg who is building the last mocks and runs the tests, no?09:13
Saviqno09:13
mzanettihmm, ok09:13
Saviqadt-run09:13
Saviqbut even so09:13
mzanettiI thought that's dpkg-magic09:13
mzanettiso, doesn't adt-run export some information about the env?09:13
Saviqdon't think we should rely on that09:14
Saviqbecause you don't want to have to export this manually when running outside of adt-run09:14
tsdgeosSaviq: well, default to what we have unless the env var is exported and then use that?09:16
Saviqtsdgeos, yeah, I was thinking that, but still feels like a band-aid09:16
tsdgeosseems good to me tbh09:16
tsdgeosall these "let's detect where i am" are like this09:17
Saviqtrueth09:17
tsdgeosuse some env vars, if not use some paths i got while compiling, if not just try /usr and if not just bail out and go crazy09:17
Saviqaargh why does QtC not open CMakeLists.txt files any more but tries to build the project ;(09:25
Saviqoh ok, need to open from project pane not filesystem09:25
Saviqmzanetti, tsdgeos, opinion: should we rely on paths coming from paths.h to be /-terminated, or assume they're not?09:43
tsdgeosSaviq: if we're going to accept envvar, assume not09:43
tsdgeosthere's nothing more annoying that stuff failing because you have to add / to an envvar09:43
Saviq+109:44
mzanettiwfm09:44
mzanettiI usually add trailing slashes, but mostly not rely on them being there09:44
mzanettiwhich causes my code to often have // in it... but at least it works09:44
Saviqyeah, so assume not it is09:45
Saviqoh wow, QtC remembered changes after crashing, kudos ;D09:48
=== alan_g is now known as alan_g|lunch
=== alan_g|lunch is now known as alan_g
mzanettimterry, can't repro this: https://bugs.launchpad.net/ubuntu/+source/unity8/+bug/149848613:45
ubot5Ubuntu bug 1498486 in unity8 (Ubuntu) "Clicking on a 'folded' launcher icon should always result in a launchable icon" [Undecided,New]13:45
mzanettimterry, for me the clicked icon ends up being completely unfolded13:45
mzanettion krillin that is13:45
mterrymzanetti, huh.  I only tested on mako, assumed they had same size in landscape13:46
mterrymzanetti, do you have 3 icons not shown in the direction you are testing?13:46
mzanettihowever, I have seen issues that ListView.flick() doesn't always flick with the same strength13:46
mzanettimterry, yes, I have like 20 icons in the launcher13:46
mterryhumph13:46
mzanettiand trying somewhere in the middle13:46
mterrymzanetti, maybe it's mako-specific (like it matters on a pixel level, instead of a grid unit level)13:47
mzanettiprobably13:47
* mzanetti searches for a mako device13:47
mterrywill edit bug to not mention krillin13:47
mzanettimterry, hah... indeed. it flicks like twice as much on mako13:48
mzanetticycles through the whole list :D13:49
mterryoh interesting it's so different13:49
mterrynot the whole list. just like 3 icons...?13:49
mterryFor me, the nearest-folded icon will bounce between top and bottom when I click on it13:50
mterryBut always be folded13:50
=== pat_ is now known as Guest18418
=== dandrader_ is now known as dandrader|afk
=== dandrader|afk is now known as dandrader
=== Guest18418 is now known as pmgowan
=== alan_g is now known as alan_g|EOD
dandraderSaviq, do you know where things like Qt5Gui_PRIVATE_INCLUDE_DIRS come from?17:02
dandraderSaviq, ie, what file defines it17:03
Saviqdandrader, dpkg -L qtbase5-dev | grep cmake17:03
dandraderSaviq, thanks17:04
Saviqgrep PRIVATE -r /usr/lib/x86_64-linux-gnu/cmake/Qt5Gui17:04
=== dandrader is now known as dandrader|afk
=== dandrader|afk is now known as dandrader

Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!