With 1 CPU, Calamares still spawns 9 threads or so: eventloop,
dbus loop, QML loop, ... many of those are invisible to the
application. Contention occurs on startup when the UI is constructed,
and we end up with the module manager creating widgets alongside,
or ahead of, the main window UI. This can result in deadlock:
- in CalamaresApplication::initViewSteps
- in QML imports
This is partly because the signal-slots connections get "deep":
from loadModules() we emit *modulesLoaded* which ends up showing
the main window in initViewSteps(). Avoid this with a QTimer:
drop back to the event loop and release whatever locks are held,
so the QML thread can get on with it already. Then the timer
goes off and the view steps are created.
- Give LVM jobs a dummy argument Device* so that they
fit the functionality of makeJob for partitioning.
For those jobs that already take an LVMDevice*, this should
be the self-same device, but that isn't checked.
* Use the minSize when the target storage is smaller than the sum of sizes
* Percentage-defined partitions should be computed after setting hard-defined ones
This fixes issues when 0 byte partitions were created when the disk is too small.
Also fixes an issue with percent-defined partitions being forced to be defined at the end of the disk.
- create dirs as needed (this will normally be done by
unsquash, but for tests with paths it needs to be done
by hand)
- log what file is being checked
- filePath() doesn't like the absolute paths we have
(they're absolute in the chroot, and existing code
just sticks rootMountPoint in front)
Document keyboard change for Turkish F layout, and document
the keyboard configuration value better, with alternate
path used in e.g. openSUSE
FIXES#1397
Some tests -- notably the keyboard module -- need to have the
QRC for the module loaded as well (e.g. because of data in the
QRC). Add a RESOURCES parameter to calamares_add_test()
like calamares_add_plugin() already has, to build the
resources into the test.
Keyboard test now passes, since it was missing the data for
lookups before.
- both changing the autologin and changing the user (login) name
affect global storage, and both may need to change the autologin
username; split it into a free function.
- the fullname change was bypassing the login in changing the
login name, **but** then it needs a back-workaround to keep
the "custom" setting off (when custom is off, auto-fill username
and hostname is active).
- after loading the config, fill GS already.
- when finalizing GS, get the autologin settings again.
- setup the visibility and initial checked-state of the reuse-user-
password-for-root near where it gets connected; do similar
for the require-strong-password
- squash the lambda slot into the regular slot: no sense in
connecting twice to the same signal with the same receiver.
- only connect config->ui once
- only connect at all if the setting is visible (e.g. when weak
passwords are allowed for the require-strong checkbox, or
when root's password will be written for the reuse-password)
- switch to QStringList as parameter, since consumers (that is,
the debug dialog, which is what this is for) are interested
just in the **names** of the jobs.
- to allow mutex locking in const methods, mark them mutable.
