This is related to https://invent.kde.org/neon/neon/calamares-settings/-/merge_requests/1
which adds .. the default things from the example configuration to the
configuration file KDE neon ships. The default layout doesn't add
any subvolumes at all, which seems to be non-functional.
If nothing is configured, complain and use /@ as the lone subvolume.
PARTITION_UNSAFE is a debug mode. It is not used in
production, because it allows you to pick an install
device that would be dangerous (e.g. the current / device).
Existing code kept two copies of a list of pointers,
and deleted pointers from one of the lists and returned
the other -- which now contains dangling pointers.
Refactor by applying suitable lambdas to a single
copy of the list; this avoids copying the list so
there is no danger of dangling pointers.
The bootloader model knows about both rows and
devices, so we can look up both at once. The
existing implementation as a non-member was rather
sketchy and wasn't used except as support for
restoreSelectedBootLoader().
If the pakcage manager fails in some way, convert to a readable
error message instead of leaking the exception to the caller
(which produces a traceback, which is harder to read and less
informative)
Widgets are easier to style if they have a name, and easier to spot
in the widget tree as well. Give the requirements-checker
parts meaningful names.
SEE #1685
- in legacy mode, *id* can have an effect and leads to
"packagechooser_<id>"; if unset, uses the the module
instance id instead, still as "packagechooser_<instanceid>".
- in packages mode, *id* is not used and only the whole
module Id (generally, "packagechooser@<instanceid>")
is used, but in packages mode there's no need for other
packages to mess with GS settings for this packagechooser.
When the module is loaded and the viewstep created, it doesn't have a
module Id **yet**. That is set after reading more of the configuration
file. It **is** set by the time setConfigurationMap() is called,
so pass it on to the Config object then. This means that packagechooser
modules can skip the *id* config key and use the module Id.
The model needs to be attached to the widget; because of changes
in the order that widget() and setConfigurationMap() are called,
the model is created earlier, but needs to be connected later.
- the %4 is left-over from the feature-summary string,
- replace it with ""; don't change the source string
because that will break translations right now.
- don't pass the item IDs to packages module, use the
packages lists for each item
- document the item list in more detail (including the packages member
and new install-method item)
Rip out most of the ViewStep that deals with configuration,
move it to a Config object (not one that supports QML yet,
though), and massage the model a little.
The single-values *package* member in a PackageItem was not used,
so remove it -- to show that it really isn't used. This is prep-
work for putting the package name *back*, as multi-valued,
and using the *packages* module.
- if the queue is emptied, there was no usable data; set
failure to NoData rather than BadData.
- FetchNextUnless::done() is done only if the parameter is true (that
is, it's done!); otherwise should continue.
- drop the debugging line because that has already been
logged by the call to `runCommand()` that backs
`target_env_call()`.
- use the same (top-level) function rather than having a
function and elsewhere a very-similar method.
The code path for setting the locale / language automatically
emits currentLanguageStatusChanged(), but the code that updates
GS connects to currentLanguageCodeChaged(). This was altered in
the 3.2.28 release cycle. Since then, automcatic locale selection
wasn't setting *locale* in GS, so that a click-through kind of
locale selection would not set it; then the packages module
has no *locale* setting for localization packages.
The combination of status and code signals (machine- and human-
readable) is ok. Introduce a setter to the language that does
the necessary signalling, so that setting the language automatically
also DTRT.
FIXES#1671
The code path for setting the locale / language automatically
emits currentLanguageStatusChanged(), but the code that updates
GS connects to currentLanguageCodeChaged(). This was altered in
the 3.2.28 release cycle. Since then, automcatic locale selection
wasn't setting *locale* in GS, so that a click-through kind of
locale selection would not set it; then the packages module
has no *locale* setting for localization packages.
The combination of status and code signals (machine- and human-
readable) is ok. Introduce a setter to the language that does
the necessary signalling, so that setting the language automatically
also DTRT.
FIXES#1671
- Move variables closer to where they are needed
- Do the winnowing / selection always, but in unsafe mode return
the un-winnowed list of devices
- Massage build documentation a little
When loading the lists, no initial string-value was being
set for the model, layout and variant; the configuration
could pass right through and pick up empty strings instead.
If the user does not change the model, the UI would show
"pc105" but the internal setting would still be empty.
FIXES#1668
The machinery in `setConfigurationMap()` was just duplicating
checks already in place in the `getString()` and `getBool()`
methods, and there's no special need for efficiency here,
so prefer the more readable and short code.
("efficiency" here means "we're saving one method call in
case the configuration is not set")
In 4ffa79d4cf, the spelling
was changed to consistently be "autoLoginUser" in the *users*
module, but that changed the Global Storage key as well,
and the *displaymanager* module wasn't changed to follow.
You'll need a VM with 2 disks to demonstrate:
- Configure Calamares to pick "none" as initial action on
the partition page (this is a safe choice),
- Enter partition page,
- No action is selected, and the next> button is greyed out.
- Click erase; notice next> is now available.
- Change devices, notice no action is selected, but next>
is still available. Clicking on it, though, does nothing.
When changing to "no action", update the next-button's
availability.
Avoid the extra indirection through the otherwise-unused
prettyGptType(const QString&), construct table of names
only on first call to avoid static-initialization order
(though that's not important here).
In 4ffa79d4cf, the spelling
was changed to consistently be "autoLoginUser" in the *users*
module, but that changed the Global Storage key as well,
and the *displaymanager* module wasn't changed to follow.
In particular, we need a separate Job class to set the label; this
is invoked after we formatted a partition, and when no other changes
to the partition have been requested in the Edit dialog.
The partition- and filesystem-label setting code was already there,
but not in the call to createNewPartition(); now we set the
FS label twice (once in the call, once afterwards)
When creating or editing a new formatted partition, allow
to set a filesystem label (16 chars maximum). Modify
the KPMHelpers to accept it as a new parameter. Partitions
created by default may get a meaningful label too.
The skip-checking is now in the functions for adding plugins and
subdirectories, so that third-party building should get it
as well, for free. Since AddModuleSubdirectory and AddPlugin
use the newly split-out module, handling SKIP_MODULES and USE_*
consistently across module repositories is now easier.
While here, make accumulating-the-skipped-modules explicit.
- use load() to start loading
- the FetchNextUnless class is useful in more spots in
the loading process
- set status explicitly on success (otherwise, a failure in a
previous URL would leave a failure message lying around even
when the module shows something useful)
- m_queue was not initialized to nullptr, crashes
- split queue-is-done to a separate slot rather than a lambda
- prefer queueing calls to fetchNext(), for responsiveness
Require a ; after RETRANSLATE macros. They are statement-like;
this makes it easier for some of them to be recognized by
clang-format and resolves some existing weird formatting.
Now all the business logic is in Config, the door is open to
building a QML-ified netinstall module. I'm not sure that
would be worth it: packagechooser offers more space for a
nice UI and should be QML'ed first.
There was a mix of autologin and autoLogin, leading to confusion
in the code. QML is sensitive to this, so go to one consistent name.
(Although the names of the settings in the `.conf` file are
different again)
Although the example configurations shouldn't really be used
as a sample of how to configure **your** Calamares for your
distro, many distro's do just copy the examples. So leave
traces of the OEM-configuration settings in the example,
and give the standard configuration a 'nothing changed'
set of presets.
- you can still call set*(), eg. from the UI, when the field is
not editable. Although the code previously ignored the change,
this would lead to a mismatch between what the UI is showing
(the changed value) and what the Config has (old value).
Emit a changed-signal (notify) with the old value so that the
UI is changed *back* as soon as possible.
- warn about fields applied twice (program error)
- warn about fields not used (configuration error)
- add operator<< for "clean" looking preset application
Build up the list of known presets by what the Config-object
expects, not by what the Config file provides. This allows
early detection of mis-matched configurations.
Presets can only apply to Q_PROPERTY properties, and the
preset must match the property name.
It's possible to ignore the "user setting" for restart-now
and call doRestart(true) directly. This is intended for
use with specific UIs that make that choice clear for the user.
Hook up both [finished] and [finishedq] to the "traditional"
restart-if-the-box-is-ticked logic although the example
QML doesn't expose that box.
module builds & runs, config connections are not registering
no errors
finishedq.qml is offering a different option though, running commands directly in qml
plasma-framework executer is used for that
- move most of the business logic to Config
- make retranslate of the page more robust (e.g. changing language
after failure would restore the un-failed message)
There's still some bits left.
- the configuration is still duplicated in the widget, and
functionality still needs to move to the Config object
- the ViewStep is cut down to almost nothing
- the host system's /etc/group is being read, and that varies between
host OS versions; since I was doing today's release on KaOS, the
test was failing because of arbitrary differences between the
default groups on each Linux flavor.
FIXES#1604
(Admittedly, this fixes the problem only when there's Plasma Solid automount
present, and not any of the other kinds; but none of those have been reported
yet, and adding them into AutoMount.cpp is opaque to the rest of the
system)
- It shouldn't be necessary to explicitly .get() pointers for
logging, and it's convenient to know when a pointer is smart.
* no annotation means raw (e.g. @0x0)
* S means shared
* U means unique
- switch logging in job to VERBOSE because we don't want to be printing
pointers to the regular session log
- switch logging in test to VERBOSE to actually see the messages from the Job
- hook the test into the build
- Although unique_ptr is only used when ICU is enabled, include it
always because it is likely that we'll use more unique_ptr
in the implementation at some point.
Ever since signed shim binaries for multiple architectures became
available, the shim binaries installed in Linux distributions have
been renamed to include the EFI architecture in the binary names.
This started in Fedora, but is now used in openSUSE and Ubuntu too.
Reference for shim binary names comes from shim spec in Fedora:
d8c3c8e392/f/shim.spec (_23-32)
The QLatin1String() might be replaced by char[], that trades one
initialization for two but with a simpler data section; this
probably is not worth profiling.
With Qt 5.15.2 (and clang), `k->first` works, but this breaks
with Qt 5.11 (and gcc), this is not available and the dereference
must be written differently, `(*k).first`.
- This kind of runs around the selection model on the view,
but we're drawing radio buttons ourselves **anyway**
and the list of themes knows which is selected / current
independent of the view.
- make ThemeInfo and ThemeInfoList internal, expose only
ThemesModel to the rest of the PlasmaLnF module
- don't build the widget anymore (needs to be replaced by
a delegate)
- also individual changes need to be signalled
- use QSignalBlocker to avoid spamming changes when calling
aggregate change methods
- refactor findById() so that also a row number can be
obtained, which is needed for the change signals.
- put a filter model in place, so only the themes with "show" set
are displayed
- rip out the messing about with widgets, soon to introduce a model-
based UI
This is the bugfix part (rather than the "clean up this widgets mess")
of issue-1573, ensuring that the LookAndFeelPackage setting is
saved to the target system config file.
- Need to update the variant that is in use, **and**
explicitly update it in the widget, in order to re-load
the keyboard image for the newly-selected layout+variant.
The lookandfeeltool does not (always?) write the LookAndFeelPackage
key that the KCM does -- and which this module reads on startup
to find the default LnF. This seems to be a regression in recent
lookandfeeltool versions (or in the KCM code that backs it).
Workaround supplied by jghodd.
Fixes#1573
The module preserves the extended attributes at rsync and the overlay
filesystem stores extended attributes by inodes.
The overlay filesystem keeps traces of the lower directory by encoding
and storing its UUID to the attribute trusted.overlay.origin. If the
index feature is on, that attribute is compared to the UUID of the lower
directory at every subsequent mounts and causes mount to fail with
ESTATE if it does not match.
This filters the namespace trusted.overlay.* by using the rsync option
--filter='-x trusted.overlay.*' to make sure the overlays extended
attributes are not preserved.
Fixes:
# mount -t overlay -o lowerdir=...,upperdir,...,workdir= overlay /mnt/etc
mount: /var/mnt/etc: mount(2) system call failed: Stale file handle.
# dmesg
(...)
overlayfs: "xino" feature enabled using 32 upper inode bits.
overlayfs: failed to verify origin (/etc, ino=524292, err=-116)
overlayfs: failed to verify upper root origin
Previously, the auxerror information was never stored, and
the messages were all un-numbered or un-explained.
Now, consume that information and store it when check()
is called, ready to be used when (possibly much later,
or after a translation change) explanation() is called.
Mount guesses the filesystem if it is unset or if it is set to auto,
thanks to blkid. That is the case for the bind mountpoints like /dev or
/run/udev in mount.conf. See `mount(8)` for more details.
The drop-down of zones was initially unfiltered, so you could start
in Europe/Amsterdam and the zones drop-down would also show Australian
zones; picking Perth would have weird effects, since Europe/Perth
doesn't exist and so you'd end up in New York instead.
- set the filtering region immediately, rather than only when the
region changes.
Various file writes were not being checked, and the code
was a bit tangled; specifically keyboardq did **not**
configure properly on KaOS and now seems ok.
The root mount-point can end with a / while the mount-point read from
the file /etc/mtab does not end with a /.
This leads to skip the unmounting of the root mount-point and causes the
removal of the root mountpoint directory to fail with EBUSY because it
is still mounted.
This uses the python functions os.path.normpath() to normalize the root
mount-point (i.e. to drop the trailing /) and os.path.commonprefix() to
determine the longest common prefix between the two mount-points. If the
returned prefix is identical to the normalized root mount-point then the
mount-point must be added to the list of the mount-points to unmount.
More generally, the python modules should rely on the os.path functions
to compare for paths instead of using strings. It covers this way lots
of corner cases (path with "//", "/../", "/./", ...).
- put the writing of each kind of file in its own block -- this should
become separate functions -- so that variables become more local
and debugging can be improved.
- while here, fix the error message for /etc/default/keyboard:
it would complain and name the vconsole file path if it ever failed.
- factor out the flags-we-want from the flags-we-already-have
- the use of ->activeFlags() meant that the state on *disk* was
being compared with the flags-we-want; if a partition was re-edited,
then you couldn't change the flags back to the state-on-disk
(eg. enable a flag, then change your mind and disable it).
- set the flags before refreshing the partition, because the
refresh checks for EFI bootability and that needs the new flags,
not the old ones.
- remove the m_defaultFSType from PartitionLayout, because it is
set on construction -- which is too early, before the configuration
has been read.
- make the default FS explicit in the init() calls which pass in
a configuration; this needs support in the intermediate
PartitionCoreModule.
- the "simple" constructor for PartitionEntry left the FS type
set as the constructor left it -- which is Unknown by default.
This leads to install failures in systems that don't set a
special layout but just want a single / -- because the FS is
set to Unknown.
- massage the constructor and consumer of the code, push
Ext4 FS in the tests and use the configured default in production.
The return of the call to libcalamares.utils.mount is never tested and
it may fail silently; this causes some mounpoints to be missing.
This adds a warning if mountpoint cannot be mounted.
chcon: failed to get security context of '/tmp/verity': Operation not supported
06:44:23 [6]: static CalamaresUtils::ProcessResult CalamaresUtils::System::runCommand(CalamaresUtils::System::RunLocation, const QStringList&, const QString&, const QString&, std::chrono::seconds)
Running "env" ("mount", "-t", "unformatted", "/dev/sdb2", "/tmp/calamares-root-kv8dqgb5/tmp/verity")
.. Finished. Exit code: 32
.. Target cmd: ("mount", "-t", "unformatted", "/dev/sdb7", "/tmp/calamares-root-kv8dqgb5/tmp/verity") output:
mount: /tmp/calamares-root-kv8dqgb5/tmp/verity: unknown filesystem type 'unformatted'.
Some compile flags changed recently, triggering assert()
in the jobqueue when there is more than one. There's no
real reason for JobQueue to be a singleton, but it wants
to be. So clean up pointers a little more enthusiastically.
- Use classes to prompt lupdate to extract with a better
context (e.g. the class name, rather than plain "QObject")
so that the translation-lookup can use the named context.
- Add hard-coded "default" variant
- Add totally bogus Tajik translations, for testing purposes
Now that Calamares is compiled as C++17, we get this:
src/modules/locale/timezonewidget/TimeZoneImage.cpp:28:55: warning: out-of-line definition of constexpr static data member is redundant in C++17 and is deprecated [-Wdeprecated]
/* static constexpr */ const QSize TimeZoneImageList::imageSize;
The code doesn't match the comment: there are no by-ref captures
in the code, and the shadowing of parameters and local variables
is confusing. Remove one variable that is passed in as an argument
(and just pass the value as argument) and copy-capture the other
rather than doing weird argument passing.
- remove unused this captures from lambda
- rename variables that are short, cryptic, and shadowed
- remove documentation for parameters that don't exist
This commit adds the new configuration `swapPartitionName` to the file
partition.conf.
This option sets the partition name to the swap partition that is
created. If this option is unset, the partition is left unnamed.
- only widgets get language change events, so we need to
hook that up to the ViewSteps and inform the Config
object so that it can (re)load translations for the
keyboard data.
- Rename the "size" locals using "sectors" in their name. Size may be
confusing or not enough specific as it can be interpreted a size in
Byte.
partSizeMap -> partSectorsMap,
totalSize -> totalSectors,
availablesize -> availableSectors,
size -> sectors,
minSize -> minSectors
maxSize -> maxSectors
- Create a the new local currentSector to iterate over the sectors;
instead of using the parameter firstSector.
- Remove the variable end that does not help much; too many variable
already. Expand its expression instead.
- Introduces new constructors for PartitionEntry: copy constructory and
constructor with all attributes.
- Use the new constructor in method addEntry().
- The variant helper toString() takes a default value since commit
c9f942ad6 ([libcalamares] Add default value to variant helpers).
- Set the default value to 0 and simplify the retreival of size values
by calling the helper toString() and removing the temporary variables.
- The logic of the method initLayout belongs to the object
PartitionLayout. Move logic to that object.
- Use a single method initLayout in object PartitionCoreModule.
- Member m_partLayout in object PartitionCoreModule is no longer
allocated.
The translation happens whenever this code is run, which may
not match the language the user subsequently selects.
It also causes general problems with the translation,
since we end up with a possibly-partly-translated name.
- With debugging and untangling done, the lambdas are simple
and short, so return to using them. One point of improvement
might be to give the models suitable slots themselves,
to avoid even this indirection.
- Use the just-refactored XKBListModel to store the xkb key-value
pairs for variants, drop most of the complicated implementation,
store just a single list of values.
- Remove code that is duplicated in Config.
- Hook up UI for physical keyboard model, and back.
- For now, introduce some named slots with debugging output.
This makes debugging a lot easier since we have function names
to work with rather than anonymous lambdas
- Config already *has* everythind, but drop the useless copies
and duplicated code from the Page
- Plug the models model into the Page
- While here, document the model / layout / variant distinctions
The code doesn't fill the UI properly, and the drop-down
for the models combobox is not right, but at least the data
is shared.
- gcc (up to at least version 10) is worse at recognizing that all
cases have been handled, so it complains about all the switches
that cover enum values.
- both clang and g++ support __builtin_unreachable(); (as Kevin
Kofler pointed out) so we don't need the macro to do different things;
- the compilers have gotten better at detecting unreachable code,
so instead of inserting macros or fiddly bits, just drop them
and the unreachable code they comment.
- if the partition size is invalid, then warn about it but do
not print the (uninitialized) size of the partition.
- shuffle code to continue earlier, allowing the "good path"
code to be out-dented.
The special setup for nicely-named groups which have a single
hidden subgroup containing the actual packages, has the problem
that there is a non-empty subgroups item, but this results
in zero actual children: then the number of selected and partly
selected children is also zero in updateSelected() and therefore
the item ends up unselected.
Special-case this to avoid unnecessarily unselecting the item.
Reported by Vitor L.
The status for an empty login name is '' (empty), for ok -- this is
so that there is no complaint about it. But it's not ok to
continue with an empty name.
The status for an empty login name is '' (empty), for ok -- this is
so that there is no complaint about it. But it's not ok to
continue with an empty name.
- The EXPECT_FAIL value "Abort" stops the test (I wanted 'if this
unexpectedly passes, raise an error' -- should have read the
documentation more closely).
- Set the shell in the config object, not just in GS.
- add a status member so the different steps can show progress
as the user is created and configured. The progress values
are hard-coded guesses as to how much work is done for each step.
- while here, reduce the scope of the global storage variable
This is somewhat experimental and weird; the idea is that bool
arguments are a lot easier to understand if there are proper
names attached, rather than "true" and "false".
- os-proper may return an extra file after the device:
/dev/sda1:Ubuntu 19.10 (19.10):Ubuntu:linux
/dev/sdb1@/EFI/Microsoft/Boot/bootmgfw.efi:Windows Boot Manager:Windows:efi
I think we had this (kind of) module a long time ago and it was
removed for over-complicating things; re-introduce one now that
KPMcore is used in 3 different places and all would benefit
from consistent API handling / defines.
- handle swapfiles when writing /etc/fstab in the target system
- special-case mountpoint
- since swapfiles are not a partition, take the setting out
of partitionChoices
- create the physical swapfile as well (there's no other place
where it would make sense)
This still won't help if there's one really huge file that takes
several seconds to write, but if there's a bunch of files together
that is less than a file_chunk_count but take more than a half-
second to write, update anyway
If there's thousands of files in a squashfs (e.g. 400000 like on
some ArcoLinux ISOs) then progress would be reported every
4000 files, which can take quite some time to write. Reduce
file_chunk_count to at most 500, so that progress is reported
more often even if that wouldn't lead to a visible change
in the percentage progress: instead we **do** get a change
in files-transferred numbers.
- The total weight is only needed by the UnpackOperation,
not by each entry.
- Use a chunk size of 107 so that the number-complete seems busy:
the whole larger-or-smaller chunk size doesn't really matter.
- The progress-report was missing the weight of the current
module, so would report way too low if weight > 1. This affects
ArcoLinux configurations where one entry is huge and one is a
single file, so weights 50 and 1 are appropriate.
When there are multiple entries, the overall weight of the
module is divided between the entries: currently each entry
takes an equal amount of space in the overall progress.
When there are multiple entries which take wildly different
amounts of time (e.g. a squash-fs and a single file) then
the progress overall looks weird: the squash-fs gets half
of this module's weight, and the single file does too.
With the new *weight* key for entries, that division can
be tweaked so that progress looks more "even".
