- Instead of Continuation(), write just Continuation
- All that futzing with overloads and tag-classes isn't needed
since the whole point is to output some constant string. Leave
cleverness for later, if it's needed.
- Using project() to set up the version is idiomatic for CMake
and more standardised than doing it by hand. Do retain the
RC flag, because that's used in other parts of versioning.
- When environment is empty, use default values from spec
- Search in application-named subdirs first (but keep previous
behavior of also searching directly in the named dirs)
- Don't consider empty XDG_* elements
- Settings XDG_DATA_DIRS=":" would yield an empty list of extra
directories to check; don't bother setting haveExtraDirs for that.
- This initial bit of code re-uses the *dont-chroot* setting;
it may need to be made independent.
- This branch will use `isSetupMode()` to adjust user-visible
strings to match the intended use.
SEE #1100
- Load instances list in a separate method
- Load sequence list in a separate method
- Don't rely on QASSERT, explicitly throw if something is missing.
This commit creates a _KiB operator for future use by the partition
module.
It also fixes a typo in one instance of MiBtoBytes(), requiring a couple
extra fixes.
Signed-off-by: Arnaud Ferraris <arnaud.ferraris@collabora.com>
- The test wants to read settings.conf, but by default it's run
from way inside the build dir, where there is no such file.
Go looking for one (but not too far).
- Write out Int, Double
- Special-case empty lists
- Do objects (not lists of objects) correctly
Now passes the tests for all the example config files.
- The code in loadYaml was refactored out of the module-descriptor
loading code, but the variable names in the implementation were
not changed and still strangely specific to the prior task.
In some cases, e.g. when calamares is used as an "initial setup" tool,
we may want to user to go through all the configuration steps in order
to end up with a usable system.
Therefore, disabling the "Cancel" button can be useful in this case.
This commit adds an option to settings.conf which disables this button
when set to "true". If the option is not present in the settings file,
the default behavior ("Cancel" button enabled & visible) is enforced.
Signed-off-by: Arnaud Ferraris <arnaud.ferraris@collabora.com>
Keeping std::initializer_list around is fraught. Causes segfaults
because I'm not keeping the underlying temporary array around
properly. Switch to vectors because those initialize from the
underlying array.
TODO: look into making this sufficiently constexpr -- perhaps
just use std::array and make find() work on that.
For (all?) those cases where we have configuration with
a value followed by a unit, introduce a class that
uses the NamedEnum properties to make parsing and split-up easier.
- Use DATA for the qml and branding directories (looks for qml/
and branding/<name>/ in those directories).
- Use CONFIG for the global settings and module settings (looks
for settings.conf and module/<name>/ in those directories).
FIXES#941
- Use YAML-CPP API for finding out if a node has a value at all.
- Asking for Type() of an undefined or NULL node throws an
exception, so the existing code didn't **actually** catch
cases where a required setting wasn't set at all.
Introduce the notion of emergency modules and emergency jobs.
Initial use will probably center around the preservefiles module,
and possibly umount.
FIXES#928
- After a failure, skip non-emergency jobs.
- After running all emergency jobs, then emit failure message.
- In log, distinguish emergency and non-emergency jobs.
Any job can be an emergency job; emergency modules spawn
emergency jobs (but conversely, a non-emergency module
can spawn an emergency job explicitly).
The substitution of @@ROOT@@ should happen when running in the
host, not in the target, system. Also only complain about it
if @@ROOT@@ is actually used in the commands.
FIXES#954
QLocale::name() doesn't include script information, and if it did
it would probably use sr_RS@Latin; when searching for translation
files it won't consider dropping just the country.
- Don't insert a space before the output of a process
- To do this, suppress space and quoting on the output, and to do
that move the labeling-output for warnings and errors into
the constructor (so that an idiomatic .nospace() does the right thing).
- Settings is just a settings class, no UI involved, so
move to libcalamares where it can be used also from
system helpers.
- YAML utilities are useful at a lower level of the stack, too.
- Use DebugRow for one-row-at-a-time output with continuations.
- Use DebugList for one-item-per-line with continuations.
- Use DebugMap for one-row-at-a-time output of a QVariantMap.
- Don't bother with the address of the main thread
- Do put a marker on restart into the log file
- Do put the Calamares version into the log file (previously, the version
was printed through cDebug() before the log file was opened, so it was
lost to the on-disk log).
The install-bits branch commit 83639b182b
dropped .so-versioning for libcalamares and the creation of the Python-
support symlink. This broke KDE Neon dev-unstable because the embedded
Python can no longer find libcalamares.
Installing unversioned .so's straight to LIBDIR is also not a good thing
(according to Debian), so revert to the original scheme with versioned
.so and a Python-support symlink.
Medium-term fix is to install unversioned straight into LIBDIR/calamares
and fix up the RPATH for the executable.
- Applies to libcalamares and libcalamaresui.so, install with no
version, just the bare .so. Since Calamares doesn't do versioning
anyway, and its plugins should be re-compiled for any change,
putting them in lib as unversioned .so's should make Calamares
happy and silence lintian.
- The Python testmodule script can end up calling in to System
methods (via System::instance()). This is unusual, and the
System instance has not been created at that point.
Now, create an instance and warn about it.
- Move logging-levels to an enum
- (re-)Order logging-levels so that the normal debug statement is
not the most-important (lowest level).
- Drop using namespace std;
These additional pointers were introduced for translations,
and needed their own tricks to get lupdate to recognize the
strings. Using QCoreApplication::translate() removes the
need to a QObject to provide context. Drop the now-unneeded
parameters.
Instead of using tr and some macro hacks to get lupdate to
recognize the translation, instead use QCoreApplication::translate()
which takes its own context for translation.
- Replace plain StringList with a list of <String, timeout> pairs,
and run that instead. All code paths still use the default 10sec
timeout and there's no way to change that.
- Copy stdout from timed-out process into the output variable,
instead of just dumping it into the log file. This will
improve the user experience, too, because they will get some
feedback / explanation of what the process has done.
- Remove some superfluous intermediate defines
- baseFactory was not used (always Calamares::PluginFactory)
- Move DECLARATION and DEFINITIONS apart
- CALAMARES_PLUGIN_FACTORY_DEFINITION was redefined (identically)
- CALAMARES_PLUGIN_FACTORY_DECLARATION was redefined (identically)
- __VA_ARGS__ was constant
- Move CommandList so it can be used from more modules than
just ShellProcess
- Allow a CommandList to run itself. This centralizes
code for executing one or more commands and simplifies
the ShellProcess module.
Various small cleanups:
- mention instance id in log message
- code formatting / style
Back targetEnvCommand() with a more general runCommand()
that takes an argument selecting the location to run
the command in. This allows us also to use the same
API for running processes in the host during install,
as we do for running them in the target system.
One reason for this change is wanting to run (user-specified)
commands and independently from the global dontChroot setting,
run those commands in the live system or the target.
This changes the ABI of the DLL, since targetEnvCommand()
is no longer exported. Plugins will need to be recompiled.
- refactor targetEnvCommand() into more general runCommand().
- While here, allow host system commands to run even if
there is no global storage.
- provide convenience accessors for ProcessResult members
- Move explanation of process errors out of ProcessJob
- Move from ProcessJob to ProcessResult, so it can be
reused outside of ProcessJob (e.g. from ShellProcessJob).
- Add some convenience functions, too.
Use dict methods, in particular d.get(k, v), to retrieve
the pretty_name() function (or None if it isn't there).
Using getattr() on a dict will not return values in the
dict.
For called processes, replace the not-very-useful type message
with the command-explanation, and replace the value (previously
command-explanation) by the stderr of the failed command.
FIXES#865
This is why the whole refactoring started: to get the process output
and exit code in one spot so we can attach the process output
to the (Python exception) CalledProcessError in all the code
paths, not just those that are explicitly looking for output.
- Add a more general targetEnvCommand() that returns both
error code and process output.
- Change existing targetEnvCall() and targetEnvOutput()
to use general form while discarding some data.
- Refactor, internal _handle_check_target_env_call_error doesn't need
to be in header or visible.
- Add optional output (of the command) to the Python exception.
- This is only a partial solution to warnings caused by third-party
code, since #including the headers from other sources won't apply
the warning-suppressions.
- Flags are not applied when building the source as part of a larger
target, but are on re-building just one object (it seems -- CMake
issue to track down).
Thanks to Kevin Kofler for pointing out what I'd forgotten about
source-file flags. While at it, introduce a generic mechanism for
suppressing warnings in third-party code.
Mostly reverts 4930484931
