- Change timeout to 60s (5 mins is too long)
- Catch exceptions when fetching the image and skip (better to get all
images except 1 than crash)

Signed-off-by: Callum McIntyre <[email protected]>

Add the context to data

Context includes useful data about how the test ran - the soms, machines
used, VMs used, and other test parameters. This is useful for
differentiating between tests with the same/similar names, and without
it we need to click through to the RAGE graph to determine how tests
differed. ragelib now passes context of each row for consumers to use
however suits them bet.

Signed-off-by: Callum McIntyre <[email protected]>

…ge number of points

Private/patrickcoo/cp 38420

Signed-off-by: Edwin Török <[email protected]>

To prepare the library to work normally with pip install we:
- define ragelib as a regular python package
- move to setup.py and requirements.txt to pyproject.toml and setup.cfg

We remove Pipfile because it's only useful for locking library versions,
only that should only be done on non-library packages as these may cause
conflicts otherwise. If there are any bounds these can be define in the
dependencies

Signed-off-by: Pau Ruiz Safont <[email protected]>

`expected.alert_is_present()` is always true, because it is a Selenium predicate
(a `Callable` that expects a web-driver).

This currently raises a NoSuchAlertException.
We have to actually call the expectation with a driver, which will check if an alert is present,
or return false (it internally catches this exception), instead of calling switch_to ourselves.

Fixes: 69db96c ("Reorder element checking as to avoid 10s wait for dialogue pop-up")

Signed-off-by: Edwin Török <[email protected]>

Selenium has broken backwards compatibility with 3.9.0:
* executable_path is no longer valid, now a service parameter needs to be used
* firefox_options got renamed to option

Add an upper bound on major version to avoid the library breaking again.

Signed-off-by: Edwin Török <[email protected]>

Signed-off-by: Edwin Török <[email protected]>

Package library in a standard way

Fix race condition on alert presence

Update to Debian12 and Python3.11.2 requirements

There is some race condition between page loads and fetching the images: force new instance.
Even though the code waits for certain page elements to appear and be in the correct state, apparently this only works on the first page.
When loading the next page these elements will already be there and the page will be considered already loaded even though it hasn't fully finished loading yet.

This is a "big hammer" approach, followup commits will use tabs instead of an entire new process.

The effect of the race condition is that you see graphs from completely unrelated measurements shown in place of other graphs
(which is a distinct bug from the "duplicate title causing overwrite" bug)

Signed-off-by: Edwin Török <[email protected]>

Switch to a new tab for each test, and then close it and switch back to the empty page.
This ensures that every image fetch correctly waits for the entire page load, but without the overhead of spawning an entirely new browser process every time.

Signed-off-by: Edwin Török <[email protected]>

Avoid leaving browser processes behind on errors

Signed-off-by: Edwin Török <[email protected]>

Fix race condition when fetching images