The timeout utility guarantees an exit code of 124 when the process died due to
timeout. Since awesome only ever exits with 0 or 1, we can use this to reliably
detect timeouts and print a matching message.
Signed-off-by: Uli Schlachter <psychon@znc.in>
coreutils provides a timeout command. Use that instead of (badly) inventing our
own version of it. This "timeout" command seems to be new. Let's hope everyone
has it and think about alternative solutions only when needed.
Fixes: https://github.com/awesomeWM/awesome/issues/1075
Signed-off-by: Uli Schlachter <psychon@znc.in>
When awesome exits with a non-zero code, this is something interesting that we
should log. Do so.
The "set +e" / "set -e" dance is required so that we do not abort because the
wait builtin returns a non-zero code.
Signed-off-by: Uli Schlachter <psychon@znc.in>
This script runs under "set -e", so any command exiting with a non-zero status
makes it abort. However, we do not care about failures from grep to find
anything, so handle that case gracefully.
Signed-off-by: Uli Schlachter <psychon@znc.in>
Before this commit: When we are GC'ing an object, we clear its metatable, since
otherwise crashes could occur in various places. This means that if someone
tries to use such an object, they get an unhelpful error message like "attempt
to index userdata object" and they don't understand what the problem is. Also,
this means that foo.valid does not actually work after GC.
This commit changes this behaviour. Instead of setting an empty metatable, we
now create a metatable with an __index and __newindex method. These metamethods
produce better error messages that they sat the underlying object was already
garbage collected. Better yet, the __index metamethod makes foo.valid be false
instead of causing an error, so that the existing machinery for detecting
invalid objects continues to work.
This commit also adds a functional test that verifies this behaviour.
Signed-off-by: Uli Schlachter <psychon@znc.in>
To ensure that some features such as SNID rules work, we need
to ensure that the screen isn't set by other code paths. Only
a single algorithm can be executed for the screen. As soon
as many algorithms are executed on events such as "manage", it
will most likely regress again.
This commit make sure of that by disabling the default normal source
of c.screen. After that, any other c.screen changes can be
considered bugs.
The code was written so that it assumes that disconnecting the last signal also
removed the corresponding entry in the signal array. This lead e.g. to an
index-out-of-bounds access in some cases.
Signed-off-by: Uli Schlachter <psychon@znc.in>
The "test launching client on a specific screen" suit is very slow.
However, it is also necessary to avoid issues such as #1069 or #154
from regressing again.
This is a temporary fix until a faster test client "daemon" is
developped.
While testing using "the real deal" and with all the tests would be
better, it would add a lot of complexity to the testing framework.
This module generate multiple multi-screen scenarios and some obvious
issues that they can cause. Over time, as more steps are added, it
will provide "good enough" testing for multiple screens.
Individual test suits can require() this utility to replicate their
steps for each multi-screen scenarios.
This uses the new support introduced in f0f31bc305 in the docs and in
tests/run.sh, removing an useless use of cat/echo.
Signed-off-by: Uli Schlachter <psychon@znc.in>
The build is no longer aborted when one of the "example tests" produces a
message on stderr. However, on Travis this requirement is still made. This
should catch "bad errors" via Travis while not breaking the build for users.
Fixes: https://github.com/awesomeWM/awesome/issues/821
Signed-off-by: Uli Schlachter <psychon@znc.in>
Running the tests sadly takes much to long. Since I don't have a good idea what
to do about this (I'd like to run all tests in a single Lua process, but that
doesn't seem to be possible easily), instead let's just make it more explicit
what is being done. This commit prints a message for each test that is being
run.
Signed-off-by: Uli Schlachter <psychon@znc.in>
Now that tests are no longer scanned for recursively, the hack of passing values
back and forth via the environment is no longer needed and can be removed.
While at it, this also exchanges the "useless use of regex" for an explicit
string replacement.
Signed-off-by: Uli Schlachter <psychon@znc.in>
Instead of recursively walking the directory tree, this commit makes the code us
GLOB_RECURSE to find all files and then handles them on after another.
Signed-off-by: Uli Schlachter <psychon@znc.in>
The namespace of e.g. "tests/examples/awful/mouse/coords.lua" is "_awful_mouse".
This is purely based on the path of the file.
Previously, this was computed while recursively scanning the directory tree.
This commit instead moves this to an extra function that handles this task.
Signed-off-by: Uli Schlachter <psychon@znc.in>
Previously, while recursively scanning the directory tree, the code in here also
scanned for template.lua files and remembered the latest one it found. This
commit adds a function which finds the right template.lua for a given file name,
making this search explicit and easier to understand.
Signed-off-by: Uli Schlachter <psychon@znc.in>
There was a problem that the examples were considered to have failed as soon as
they produced any kind of output, but there were legitimate cases of warnings
being printed that triggered these checks. Commit 4819be4f4f used a
regular expression to detect and ignore this warnings.
This commit reverts the above commit and instead silences the warnings by
monkey-patching the function that prints the warnings into a no-op.
Signed-off-by: Uli Schlachter <psychon@znc.in>
The requirement to call add_signal() was added to catch typos. However, this
requirement became increasingly annoying with property::<name> signals and e.g.
gears.object allowing arbitrary properties to be changed.
All of this ended up in a single commit because tests/examples fails if I first
let add_signal() emit a deprecation warning.
Signed-off-by: Uli Schlachter <psychon@znc.in>
There was a race with autofocus. To overcome this, add another step that tests
that moving the client back to a visible tag and focusing it updates
_NET_CURRENT_DESKTOP. Afterwards, we kill the client so that it can no longer
interfere.
Signed-off-by: Uli Schlachter <psychon@znc.in>
As the documentation generation insert increasingly large ammount
of code into the lua files, the coverage data is getting less and
less accurate. This try to fix this by only collecting such data
after the `configure_file` calls are done.
kill is a wrapper around the POSIX kill() function and unix_signal is a table
that maps signal numbers to their names and signal names to their numeric value.
Signed-off-by: Uli Schlachter <psychon@znc.in>
Apparently, there is such thing as not leaking enough...
Also try to clear the widgets from mywibox. This seem to help.
Time will tell.
Fixes#914, unfixes #808
Currently, tests/run.sh expects the directory layout that our wrapper Makefile
sets up before running CMake. This commit adds support for any other directory
configuration as well.
Signed-off-by: Uli Schlachter <psychon@znc.in>
GDK_SCALE=1 is needed to overwrite other settings that people might have which
would make geometry-related tests fail.
NO_AT_BRIDGE=1 gets rid of the following message that I am seeing:
** (lua:8321): WARNING **: Error retrieving accessibility bus address:
org.freedesktop.DBus.Error.ServiceUnknown: The name org.a11y.Bus was not
provided by any .service files
Signed-off-by: Uli Schlachter <psychon@znc.in>
Any clients with these tags end up somewhere random (the first tag on the first
remaining screen). This certainly can be improved in the future, but at least
this is a start.
Signed-off-by: Uli Schlachter <psychon@znc.in>
GtkApplication does magic like ensuring that the application ID is unique and
there is only a single instance of each application running. We don't want nor
need that for the tests.
Signed-off-by: Uli Schlachter <psychon@znc.in>
Startup notification support in urxvt is optional while GTK always supports
startup notification. Thus, use the new GTK-based test client for the SN tests.
Fixes: https://github.com/awesomeWM/awesome/issues/848
Signed-off-by: Uli Schlachter <psychon@znc.in>
Tags are accessible as s.tags on a screen object. Yup, that's harder to find
than a variable that is defined in the default config, but such is life.
Now that awful.rules supports specifying tags by name, I guess that the number
one reason for needing the tags table is gone.
Signed-off-by: Uli Schlachter <psychon@znc.in>
There was a regression when refactoring the API. It was no longer
possible to disable the automatic tag selection.
Due to recent changes, it was no longer possible to disable the
default tag selection handler. This commit extend the already
existing request::tag mechanism to let handlers select the tags.
This commit makes all C code that previously returned a screen index now return
a screen object, continuing the deprecation of screen indicies. Note that this
is an API break and will likely cause all kinds of problems for users.
The change also breaks some tests which are suitably fixed in this commit.
Signed-off-by: Uli Schlachter <psychon@znc.in>
When e.g. test-leaks.lua fails, it will cause a Lua error before starting the
test runner. This means that the test will just hang, because nothing causes
awesome to quit.
Handle this by starting a timer when the test runner is loaded and quitting
awesome in there if no test run was started yet. This only works if all tests
load the runner before doing anything that could fail, so the require("_runner")
is moved to the beginning in every test.
Signed-off-by: Uli Schlachter <psychon@znc.in>
Step 1 starts a client and uses awful.rules to move it to a not-selected tag.
Because this rule still has focus=true, this calls awful.ewmh.activate() via the
request::activate signal. This function makes the client urgent because it is on
a not-selected tag.
Step 3 does the same thing, but also uses switchtotag=true. Now
awful.ewmh.activate() doesn't make the client urgent because it successfully
focused this client. However, the test was wrongly assuming that the client
became urgent (copy&paste error? I don't know).
The fix is of course not to require the client to become urgent.
Signed-off-by: Uli Schlachter <psychon@znc.in>
Currently, an error in the default config in the right place isn't noticed. Fix
this by doing two things:
- Also grep for "error" (this catches runtime errors with a stack trace)
- Make _runner print a "success" message at the end and also grep for that
Fixes: https://github.com/awesomeWM/awesome/issues/689
Signed-off-by: Uli Schlachter <psychon@znc.in>
Both the test runner and the wibox use gears.timer.delayed_call(). The test
runner uses this to call steps and the wibox uses it to trigger redraws. When
running under LuaCov, the Lua code becomes slow enough that the wibox didn't
redraw yet when the leak check is run. This causes the check to fail, because
the client is still referenced by the tasklist and thus cannot be garbage
collected.
Fix this by waiting one more iteration before running the leak check.
Signed-off-by: Uli Schlachter <psychon@znc.in>
This opens xterm, closes it and makes sure that the client object representing
xterm is GC'able at the end. The test will fail currently.
Signed-off-by: Uli Schlachter <psychon@znc.in>
The benchmarks in tests/test-benchmark.lua have two modes. When CI=1 is set in
the environment, only a "quick" and less exact test is done. Otherwise, a slower
and more exact measurements is taken. This was added so that we do not waste CPU
time on travis.
However, most of the time the user running "make check" doesn't want exact
measurements either. So instead of only being quick when CI=1 is set, this
commit changes the logic to always being quick unless BENCHMARK_EXACT=1 is set.
Additionally, a message is printed next to the benchmark results so that the
user is reminded to set this var if the measurements should actually mean
something.
Signed-off-by: Uli Schlachter <psychon@znc.in>
While setting up an environment to run the integration tests in, the run.sh
script uses sed to generate versions of several files that refer to the
not-installed version of files. One of these needs to replace the call to
beautiful.init().
Before commit 20c9723c5b, the corresponding line was:
beautiful.init("@AWESOME_THEMES_PATH@/default/theme.lua")
Now this wants to find and replace the following:
beautiful.init(awful.util.get_themes_dir() .. "default/theme.lua")
To handle both versions, this commit adds some wildcards to the sed-expression
so that any line containing a call to beautiful.init is found and replaced.
Signed-off-by: Uli Schlachter <psychon@znc.in>
Having many arguments can easily get confusing and hard to understand. This
commit uses a table instead so that we have names that identify what each
callback does.
Signed-off-by: Uli Schlachter <psychon@znc.in>
When using `spawn` without startup notification support, the
`startup_id` property on the client should be nil.
This adds rxvt-unicode to the Travis build, because it supports startup
notifications, but xterm does not. We should replace xterm with it in
the existing tests then later.
* This commit add a new module to avoid a (4 level) loop dependency
* It is now possible to call awful.spawn() with a table of properties
* awful.rules is used to execute the rules.
* Everything is public to allow alternative workflow modules such as
Tyrannical to use their own callback implementation.
This makes it possible to add something similar to a __index / __newindex
metamethod to all our C objects. Based on this, Lua can then easily implement
arbitrary properties on our capi objects.
I have no idea why this needs collectgarbage() to be called twice.
On the other hand, I can explain the change in tooltip.lua. Lua 5.2 introduced
"ephermeron tables". This means that in the following sitation, lua 5.2 can
collect the entry from the table, while 5.1 keeps the entry alive, because the
table has a strong reference to the value and that in turn has a strong
reference to the key:
t = setmetatable({}, { __mode = "k"})
do
local k = {}
t[k] = function() print(k) end
end
collectgarbage("collect")
print(next(t, nil))
To handle this incompatibility, this commit just removes the whole indirection
through the module-level variable "data".
Signed-off-by: Uli Schlachter <psychon@znc.in>
Apparently some of the last commits speeds up create_wibox() a lot. This
highlights that this is a bad test: After creating thousands of wiboxes, awesome
needed 15 seconds to draw all of them and in the end some dbus timeout aborted
the test run.
However, it's irrelevant how quickly we can create wibox. The interesting number
is how quickly we can display a new wibox. Thus, this commits changes the code
so that it also measures the time that is needed to update the wibox. This way,
we don't accumulate a huge number of pending repaints and everything's fine.
Some results (but there is nothing to compare this with):
create&draw wibox: 0.0373947 sec/iter ( 28 iters, 1.59 sec for benchmark)
update textclock: 0.00198174 sec/iter (515 iters, 1.937 sec for benchmark)
relayout textclock: 0.000614439 sec/iter (1710 iters, 1.051 sec for benchmark)
redraw textclock: 0.00116882 sec/iter (865 iters, 2.962 sec for benchmark)
tag switch: 0.000705579 sec/iter (1498 iters, 3.703 sec for benchmark)
Signed-off-by: Uli Schlachter <psychon@znc.in>
This commit does two things: It gets rid of the reference to the layoutbox that
the default config created and it changes the widget dependency cache to not
keep widgets alive unnecessarily.
Signed-off-by: Uli Schlachter <psychon@znc.in>
Again, instead of directly connecting to various signals for updating a
tasklist, this commit changes the code so that there is just a single, global
connections and based on this a weak table with all tasklist instances is used
do the updates.
Signed-off-by: Uli Schlachter <psychon@znc.in>
Similar to what the previous commit does for layoutboxes, this changes the code
for the taglist so that there is only a single, global connection to the various
signals and these update all taglists via weak tables.
Signed-off-by: Uli Schlachter <psychon@znc.in>
This (ab)uses the integration tests to run a benchmark. This currently only
measures wibox drawing performance. To avoid wasting CPU-cycles, this does only
a quick run under travis while on "normal" runs the function under test is
executed in a loop to improve the precision of the measurement.
This benchmarks hopefully allow to optimize things in a clear fashion instead of
things like "it feels faster to me".
Results when run against the previous commit:
== Running test-benchmark.lua ==
create wibox: 0.0788958 sec/iter ( 13 iters, 1.103 sec for benchmark)
update textclock: 0.019493 sec/iter ( 56 iters, 2.507 sec for benchmark)
relayout textclock: 0.0160725 sec/iter ( 63 iters, 1.029 sec for benchmark)
redraw textclock: 0.0015601 sec/iter (647 iters, 1.875 sec for benchmark)
W: awesome: a_glib_poll:291: Last main loop iteration took 6.593912 seconds! Increasing limit for this warning to that value.
Results right before the new widget layouts were merged (commit 52154d0f15):
== Running test-benchmark.lua ==
create wibox: 0.0782874 sec/iter ( 13 iters, 1.095 sec for benchmark)
update textclock: 0.00736755 sec/iter (136 iters, 1.346 sec for benchmark)
W: awesome: luaA_dofunction:77: error while running function
[...]
error: /home/psychon/projects/awesome/build/lib/gears/object.lua:30: Trying to emit non-existent signal 'widget::layout_changed'
Closes https://github.com/awesomeWM/awesome/pull/451.
Signed-off-by: Uli Schlachter <psychon@znc.in>
This tried to use awful, but didn't load awful itself. Kids, this is why you
should make your variables local!
Signed-off-by: Uli Schlachter <psychon@znc.in>
When using Xephyr, this already starts the server with -noreset, but for Xvfb we
also need this flag. Without this flag, the DPI value that is set via xrdb gets
lost at server reset.
This wasn't a problem before commit 6d4837a53a. That commit moved the launch
of the dbus session after the setting of the DPI. So previously, waiting for
server startup was half broken (the dbus session already tried to connect to the
server to check for when it shuts down), but due to this no server reset
occurred and thus the DPI was correctly applied. After this commit, the server
immediately resets after xrdb is done setting the DPI and the value set is lost.
Signed-off-by: Uli Schlachter <psychon@znc.in>
This factors out a function wait_until_success that runs some command until it
succeeds (with a timeout) and uses this function in the two places where this
was already done before.
Note that this removes the "kill -0" trick for early exit again and instead will
use the timeout in case awesome dies during startup.
Signed-off-by: Uli Schlachter <psychon@znc.in>
When this script is not run under Travis, it will prepare a temporary config
file and a theme file that point to the files that were not yet installed.
Signed-off-by: Uli Schlachter <psychon@znc.in>
The code does some dances with xrdb to ensure that the server finished starting
up. However, before this it already tries to access the server via dbus-launch.
Since nothing uses dbus in this part of the code, we can just move this down.
Signed-off-by: Uli Schlachter <psychon@znc.in>
I tried calling this script with "tests/foo.lua" as an argument, but it actually
expects "foo.lua" since it changes into the appropriate directory itself.
Because this file does not exist, "cat $f | awesome-client" failed and thus this
whole script just hung (the test runner was never actually started).
Fix this by testing for the test-file to exist before trying to run it.
Signed-off-by: Uli Schlachter <psychon@znc.in>
Daniel Hahler
Uli Schlachter
Emmanuel Lepage Vallee
Crazy Lemon
Uli Schlachter
cmertz
Daniel Hahler
The Personal Things
Uli Schlachter
actionless