When an imagebox was drawn with width or height zero, it tried to calculate the
needed scale factor for making the image fit. Sadly, this would be a division by
zero aka infinite in this case.
Fix this by just not drawing anything if there is no space available.
Signed-off-by: Uli Schlachter <psychon@znc.in>
This reverts commit c1cb7883b5.
The cairo surface used for uploading the wallpaper doesn't use a temporary
connection but is on our main connection since commit 5024843e9. Thus, the above
commit broke all of our cairo drawing instead of just making cairo not touch our
temporary connection.
Since commit 5b4666432f, we use set_text() instead of set_markup()
on the tooltip's textbox. This means it is no longer possible to use pango
markup in the tooltip which was not intended.
Fix this (properly) by introducing a :set_markup() function on tooltips (and use
it in the timer function to restore the old behavior).
Signed-off-by: Uli Schlachter <psychon@znc.in>
This inlines the set_defaults() function into its only caller and makes us less
stupid with the font property.
Signed-off-by: Uli Schlachter <psychon@znc.in>
This makes gears.color() cache patterns in a weak table and returns that cached
pattern when we get called with the same argument again.
To benchmark this change, the following code was used:
local time = require("socket").gettime
function benchmark(func)
local begin = time()
local iter = 0
while time() - begin < 1 do
func()
iter = iter + 1
end
return iter
end
for _, arg in pairs({
"#00aa00",
"solid:#00aa00",
"radial:50,50,10:55,55,30:0,#ff0000:0.5,#00ff00:1,#0000ff",
"linear:1,2:3,4:0,#000000:1,#ffffff",
"png:/home/psychon/Wallpaper/Bars.png",
{ type = "solid", color = "#00aa00" },
{ type = "radial", from = { 50, 50, 10 }, to = { 55, 55, 30 }, stops = { { 0, "#ff0000" }, { 0.5, "#00ff00" }, { 1, "#0000ff" } } },
{ type = "linear", from = { 1, 2 }, to = { 3, 4 }, stops = { { 0, "#000000" }, { 1, "#ffffff" } } },
{ type = "png", file = "/home/psychon/Wallpaper/Bars.png" },
}) do
collectgarbage("collect")
print(benchmark(function() gears.color.create_pattern(arg) end), arg)
end
Before this change (larger numbers are better, this measures how many times we
can create the given pattern per second):
29525 #00aa00
29344 solid:#00aa00
3446 radial:50,50,10:55,55,30:0,#ff0000:0.5,#00ff00:1,#0000ff
4845 linear:1,2:3,4:0,#000000:1,#ffffff
32855 png:/home/psychon/Wallpaper/Bars.png
29883 table: 0x1bb67e0
3868 table: 0x1bb6830
5339 table: 0x1bb6c60
32772 table: 0x1bb6fe0
After this change:
126188 #00aa00
125962 solid:#00aa00
125125 radial:50,50,10:55,55,30:0,#ff0000:0.5,#00ff00:1,#0000ff
125213 linear:1,2:3,4:0,#000000:1,#ffffff
113659 png:/home/psychon/Wallpaper/Bars.png
125586 table: 0x1232680
125249 table: 0x12326d0
125468 table: 0x1232b00
113711 table: 0x1232e80
As you see, this makes some cases about 35 times faster (although I have to
admit that something like this can be expected from such a synthetic benchmark).
Signed-off-by: Uli Schlachter <psychon@znc.in>
Instead of loading files from disk every time we need them, add a cache to
gears.surface as a weak table that maps strings to cairo surfaces.
If this cache should be avoided, there is a new gears.surface.load_uncached()
function which works just like gears.surface.load() worked before.
Signed-off-by: Uli Schlachter <psychon@znc.in>
By default, tag.history.restore switches between the previous history
states, which is not what should get done when deleting a tag.
Without this, deleting multiple tags in a row, will jump back to the
first/fallback tag, instead of the older history entries.
The bugs this fix are:
* Invalid request using nil as screen
* Stop messing indexes in the old screen
* Prevent c.screen <-> t.screen mismatch
* Prevent no tags being selected in the old screen
When a screen doesn't have any tags selected, then just tag the new client with
all of the screen's tags. That way, we don't lose clients.
Also, if we failed at coming up with tags for a client, don't completely untag
it. This means that it can keep its old tags if it had any.
Signed-off-by: Uli Schlachter <psychon@znc.in>
This reverts commit bbe86e4e25. That commit caused
unintuitive, special behavior. Instead, when someone wants us to set a screen,
we really should do so.
The current premise is that c.screen should be the same as
awful.tag.getscreen(t).
The addition in `ewmh.tag` appears to be the important part here,
changing the order in awful.rules.execute is (maybe) only for
consistency across the codebase.
This expands the tilde in the path to beautiful.init and changes the
expansion in theme values to only match '^~/': tilde expansion is only
meant to be expanded at the beginning.
The latter is not really tested.
The current premise is that c.screen should be the same as
awful.tag.getscreen(t).
The addition in `ewmh.tag` appears to be the important part here,
changing the order in awful.rules.execute is (maybe) only for
consistency across the codebase.
Previously, odd things could happen if a widget was getting fitted into a
negative width or, even worse, width being NaN (not a number)!
This can e.g. happen due to a margin layout which doesn't get enough space to
even draw the margin that it is supposed to add.
Fix this by enforcing a minimum value of 0 for the width and height that a
widget gets fitted into.
Signed-off-by: Uli Schlachter <psychon@znc.in>
Given an imagebox i with i.resize_forbidden = false and a valid image set, the
call t:fit(0, 0) would return two times "not a number".
This is because the code first does some calculations to get the input image
into the available space and then tried to do some calculations needed for
scaling images up.
The first calculation already gave us h == 0 == w, the second calculation would
then calculate 0/0. This results in NaNs.
This was only noticed because NaN is not a valid table index in lua.
Fix this by returning 0,0 if we have an image of width or height 0 after the
first calculation. Since 0x0 images are valid in cairo, this also fixes the same
bug with such images.
Signed-off-by: Uli Schlachter <psychon@znc.in>
Most of these are unused since the drawing code was moved to lua. For example,
the old wibox code needed the metatable entries __next, __ipairs and __pairs so
that w.widgets worked correctly and could pretend to be a regular lua table.
Signed-off-by: Uli Schlachter <psychon@znc.in>
There is a strong hint that you are doing something wrong: You call
client.emit_signal(some_signal, c). Chances are high that this signal is
supposed to be emitted on the client object 'c' instead of the underlying client
class.
This applies to awful.rules' usage of this signal.
Signed-off-by: Uli Schlachter <psychon@znc.in>
Before commit 1b2826 in lgi, the get_rgba() function on cairo SolidPatterns was
specified like this:
get_rgba = { ret = cairo.Status,
{ ti.double, dir = 'out' },
{ ti.double, dir = 'out' },
{ ti.double, dir = 'out' } },
The above commit fixed this (without saying so) and the code became:
get_rgba = { ret = cairo.Status,
{ ti.double, dir = 'out' },
{ ti.double, dir = 'out' },
{ ti.double, dir = 'out' },
{ ti.double, dir = 'out' } },
The prototype for the corresponding cairo function is:
cairo_public cairo_status_t
cairo_pattern_get_rgba (cairo_pattern_t *pattern,
double *red, double *green,
double *blue, double *alpha);
As you see, this functions gets four double* as arguments and it will save its
result via those pointers. Old versions of lgi call this function with too few
arguments and this will cause a segmentation fault when cairo dereferences an
invalid pointer.
Signed-off-by: Uli Schlachter <psychon@znc.in>
The X11 server tells us about things that need to be redrawn via expose events.
When we get such an expose event before lua drew the drawable, we just fill the
exposed area with old data (which is black for newly-created drawables).
Fix this by tracking if we have any usable data in a drawable's double buffering
pixmap. This flag is unset whenever we throw away the old content (e.g. due to a
resize) and is set when lua gave us some new content to display.
Signed-off-by: Uli Schlachter <psychon@znc.in>
We now handle all "important" EWMH messages in lua and lua can decide to do
different things than the current obvious one.
Consistency!
Signed-off-by: Uli Schlachter <psychon@znc.in>
We can't just directly assign c->sticky directly, because we should emit the
right signals and unban the client if it is currently ban. All of that gets
handled by client_set_sticky().
Signed-off-by: Uli Schlachter <psychon@znc.in>
Commit e291c0398 already made awesome ignore invalid values for _NET_WM_DESKTOP.
That commit just made it tag clients with the first tag if they used an invalid
value for this property.
This commit reverts 90% of that change. What is still left is the range
validation. Previously, awesome would untag a client completely when it got such
a _NET_WM_DESKTOP property / message. Now, such messages are just silently
ignored.
Signed-off-by: Uli Schlachter <psychon@znc.in>
If a client gets moved to a screen without any selected tags, that client
disappears and it is non-trivial to get it back. Since this is unexpected and
annoying, make movetoscreen do nothing if the target screen has no tags
selected.
Signed-off-by: Uli Schlachter <psychon@znc.in>
This allows to apply properties and callbacks to a client like
awful.rules.apply would do, without the matching part.
This is useful with the new startup notifications, and lets you apply
the same properties and callbacks from e.g. a manage signal handler.
Emmanuel Lepage Vallee
Tin Benjamin Matuka
Uli Schlachter
Daniel Hahler