Fix gravity handling when reparenting a window back to the root window

[psychon]

Fixes #2308
CC @sigprof (I was faster than "coming weekend", whoooh; the most complicated part of this was that I broke test-gravity at first, but did not notice the error (this now has to wait for a specific PropertyNotify to know that unmapping is done, but it also listened for other PropertyNotify messages; the result was that that the state machine got really, really confused)).

[psychon]

Meh. :-(
This works when running just test-gravity.lua, but Travis points out that this fails when run together with the other tests. What the heck!?!

Read on stderr: ERROR: For window with gravity SouthWest in state MAPPED, expected y = 43+-1, but got 43
Read on stderr: ERROR: For window with gravity South in state MAPPED, expected y = 43+-1, but got 43
Read on stderr: ERROR: For window with gravity SouthEast in state MAPPED, expected y = 43+-1, but got 43

[psychon]

Uhm... okay. For some reason, the titlebars have a different size when just running the test vs when running all tests (22px for "make check", 16px for just the one test). I don't know why, nor do I know if this is a red herring or the actual problem (since the problem is a difference of one pixel, I would guess that some rounding in the division is involved, but both these sizes are even...)

[blueyed]

Related to some other config (file) being used maybe?

[sigprof]

Not sure what exactly breaks the test, but the code in titlebar_resize() is definitely incorrect for gravities which require centering. The problem is that titlebar_resize() passes only the size delta into xwindow_translate_for_gravity(), and if that delta is divided by 2 and truncated, the end result may be wrong.

To test this, I added some bindings to clientkeys:

    awful.key({ modkey,           }, "KP_Add",
        function (c)
            local _, size = c.titlebar_top(c)
            c.titlebar_top(c, size + 1)
        end,
        {description = "increase titlebar height", group = "client"}),
    awful.key({ modkey,           }, "KP_Subtract",
        function (c)
            local _, size = c.titlebar_top(c)
            if size > 0 then
                c.titlebar_top(c, size - 1)
            end
        end,
        {description = "decrease titlebar height", group = "client"}),

Then if I run https://github.com/ccg/metacity/blob/master/src/wm-tester/test-gravity.c and use Mod4+KP_Add and Mod4+KP_Subtract on windows with WestGravity, CenterGravity and EastGravity, I see that those windows do not remain vertically centered when the titlebar height is changed. Note that clicking with button 1 inside those windows will make test-gravity reposition them properly.

Changing the titlebar size in steps of 2 instead of 1 makes the problem disappear, because in this case truncation to 0 does not happen. Changing the border width also does not cause problems, because the total width occupied by borders on opposite sizes is always even.

The proper way to calculate the coordinates is to call xwindow_translate_for_gravity() twice — first with the old titlebar sizes, then with the new ones — and then use the difference of those results.

However, all of this should not matter if the titlebar size is changed from 0 to the final value in one step, which is what should usually happen — so the problem with that test is most likely caused by something else.

[sigprof]

I tried to run test-gravity.lua with both even and odd titlebar size (needed to edit awesomerc.lua for this, because I did not find a way to replace the request::titlebars handler in test-gravity.lua), but was unable to reproduce the Travis failure. (However, test-awful-rules.lua fails if the titlebar size is changed, because it contains local tb_height = gears.math.round(beautiful.get_font_height() * 1.5).)

[psychon]

Rebased on master. Still fails on Travis, but no longer fails locally. No idea what to do about this...

[actionless]

mb add scrot call and upload it to fileshare, so you could see what happens in xvfb inside travis:

https://github.com/themix-project/oomox-gtk-theme/blob/master/test/test.sh#L138-L141

[psychon]

A screenshot wouldn't tell me more than what the error message already says: Something is off by one pixel. I guess I would rather need to add lots of debug-printfs to the C code to figure out where this difference comes from (a race between... something?). But debugging on Travis is painful since the round-trip-time is so high. :-(

[actionless]

are you able to reproduce the problem inside docker image with the same ubuntu version as on travis? (i guess, 12.04)

[psychon]

I don't know and I am not going to install docker in the hopes that this might be specific to the ubuntu version. All the code that decides the positions of windows is in awesome, so the same between what is run on Travis and what is run here. I might stare a bit at test-gravity.c and see if I can come up with a possible race that would explain this difference... :-/

[psychon]

Rebased on master for #2355

[codecov]

Codecov Report

Merging #2323 into master will decrease coverage by 7.61%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master    #2323      +/-   ##
==========================================
- Coverage   84.03%   76.42%   -7.62%     
==========================================
  Files         477      332     -145     
  Lines       32671    18376   -14295     
==========================================
- Hits        27454    14043   -13411     
+ Misses       5217     4333     -884

Flag	Coverage Δ
#c_code	?
#functionaltests	?
#lua53	76.42% <ø> (-10.76%)	⬇️
#samples	71.7% <ø> (ø)	⬆️
#unittests	58.32% <ø> (ø)	⬆️

Impacted Files	Coverage Δ
lib/beautiful/gtk.lua	6.33% <0%> (-87.2%)	⬇️
lib/gears/wallpaper.lua	11.36% <0%> (-81.23%)	⬇️
lib/awful/layout/suit/fair.lua	19.14% <0%> (-78.73%)	⬇️
lib/awful/client/shape.lua	22.95% <0%> (-77.05%)	⬇️
lib/awful/layout/suit/spiral.lua	23.68% <0%> (-76.32%)	⬇️
lib/wibox/init.lua	25.27% <0%> (-68.69%)	⬇️
lib/awful/spawn.lua	27.67% <0%> (-58.41%)	⬇️
lib/awful/ewmh.lua	30.81% <0%> (-58.4%)	⬇️
lib/awful/layout/suit/corner.lua	26.73% <0%> (-58.23%)	⬇️
lib/wibox/drawable.lua	42.16% <0%> (-55.12%)	⬇️
... and 181 more