GitLab

The undo didn't happen because of the early exit due to
"no files to delete". This was forgetting directories.

* It's async
* It's an interface in KIOCore, which means one less hurdle for moving
  FileUndoManager to KIOGui (https://phabricator.kde.org/T11821)

Technically this would break expectations of any code that reimplements
UiInterface::confirmDeletion(), no longer called, but nobody was doing
that. I wrote UiInterface ages ago so Dolphin could show error messages
inline, but confirmDeletion only got there for the unittest.

* You write a small idea in thesis.txt
* You realize this should be called ideas.txt so you rename the file
* You write your entire thesis into thesis.txt
* You press Undo accidentally, the file gets renamed back
  with Overwrite enabled and lose your thesis (!)

I traced the overwrite flag all the way back to Simon's very first
implementation of the class, at a time I still had hair,
so this is not reverting an intentional bugfix.

Discovered when testing the error handling in FileUndoManager.

Matches what Dolphin does and makes it easier to spot which entries
are actually hidden when selecting "show all"

For example, if you open http://foo.com/bar/myfile.pdf you don't want to
do the trick of changing dir to http://foo.com/bar and opening
myfile.pdf, because if we do that the next time you open the file dialog
it will try to list http://foo.com/bar and won't work, which is kind of
confusing, so just do that for listable protocols, where it makes sense
that if i open a file from /home/albert/documents/kde/myfile.pdf the
next time i open the file dialog it's at /home/albert/documents/kde

GIT_SILENT

GIT_SILENT

- Un-nest Private class and make d-ptr a unique_ptr, KF6 task:
  https://phabricator.kde.org/T13924
- Remove Q_PRIVATE_SLOT() usage where possible and remove the _k_ prefix
  from slot names
- Prefix member variable names with m_

NO_CHANGELOG

In C++20 there are new classes that have member functions named emit().
c.f.:
https://lists.qt-project.org/pipermail/development/2020-February/038812.html
https://en.cppreference.com/w/cpp/io/basic_osyncstream/emit

I don't see why we wouldn't do the "if this is an absolute path, try
opening that and not what tokenize gave us" when in the files+directory
mode, so changed to not do it when in the onlyDirectoryMode

BUGS: 431454

In case of conflict in i18n, keep the version of the branch "ours"
To resolve a particular conflict, "git checkout --ours path/to/file.desktop"

This is a first step towards using the async way of confirming deletion.

Maybe one day we'll end up stat'ing before renaming...

GIT_SILENT

KUrlNavigatorButtonBase and its sub-classes only ever have a KUrlNavigator
parent. This also makes it easier to connect to KUrlNavigator::requestActivation
using the new signal/slot syntax, with compile time type checking and all
(tip from dfaure).

NO_CHANGELOG

KDirModel: replace Qt::UniqueConnection by guarding the connect() call with
a member bool (tip from dfaure).

NO_CHANGELOG

QAbstractItemModel::modelReset() is a private signal, users of the class
can connect to it but can't emit it; these two connections couldn't have
worked AFAICS. Confirmed by the failed compilation after switching to
PointerToMemberFunction signal/slot syntax.

NO_CHANGELOG

- Remove code that's been commented out since 2014
- Use new signal/slot syntax
- Code style (spaces around '=' ...etc)
- Remove redundant includes

This will prevent clang-format from wrapping these
enums on a single line. Also it will create cleaner diffs,
because the line above does not need to be modified when
a new enum value is added.

TransferJob: disconnect from connection to a lambda by using a
QMetaObject::Connection member.

Remove some code that's been commented out for a long time.

NO_CHANGELOG

- Make the d-ptr a std::unique_ptr
  KF6 task:https://phabricator.kde.org/T13924
- use new signal/slot syntax
- rename "speedBar" to "placesPanel" everywhere, that is its current name
- fix unit test
- remove most of Q_PRIVATE_SLOT usage and the _k_ prefix from slots.
- prefix member variables' names with m_
- declare and initialize variables at the same time
- replace some disconnect/connect logic with QSignalBlocker or
  QObject::blockSignals()
- remove an unused member variable
- don't use bit-fields with bool member variables, in this particular case
  we can achieve the same optimisation by declaring the bools one after the
  other (sizeof(bool) is 1, we have 8 bools)

KFileWidget: make the d-ptr a std::unique_ptr

KF6 task: https://phabricator.kde.org/T13924

- Forward declare classes where possible; but don't remove #includes from
  headers so as not to needlessly break any KDE apps' compilation that
  depended indirectly on a header being included; instead add a KF6 TODO
- Remove redundant ones

NO_CHANGELOG

NO_CHANGELOG

So as to un-overload the undoAvailable signal; use isUndoAvailAble()
instead.

Port away from undoAvailable() method; remains one QOverload usage in the
unit test, since the undoAvailable() method is still compiled at the
moment.

In case of conflict in i18n, keep the version of the branch "ours"
To resolve a particular conflict, "git checkout --ours path/to/file.desktop"

It's a bit weird asking "what do you wish to do with this executable file?"
when the file in question is actually a .desktop file. Now the label just
says "this file".

Adapt the unit test to the changes.

In case of conflict in i18n, keep the version of the branch "ours"
To resolve a particular conflict, "git checkout --ours path/to/file.desktop"

GIT_SILENT

As discussed in [1], benchmark unit tests are useful when benchmarking
performance, but not if no one is looking at the results (either when run
locally or on the CI); change such tests to be always compiled, so as not
to bit-rot, but don't run them automatically when using e.g. `ctest` or
`make test`.

This was first propsed by Meven in https://phabricator.kde.org/D26468

[1] !301

It seems CI cannot exec files from the source directory during tests.
This patch copies the file into a temporary directory and adjusts
permissions explicitly.

This reverts commit 6093b218.

Breaks compilation.

Rather than using Q_OS_* defines to determine what fields are available
in struct stat and how to handle them, use C++ SFINAE to detect which
fields there are and call accordingly. This will help systems
like OpenBSD, which **do** have nanosecond-accuracy timestamps,
but don't set them right now because they are not listed in the
`#ifdef` list for using nanosecond-accuracy.

This introduces a handful of SFINAE machinery that could possibly
be used elsewhere, but I have not looked very hard.

Cast directly to the type used in the QList, uint.

- Un-nest the Private class, and rename it to KDirListerPrivate, also make
the d-ptr a unique_ptr.
- Prefix the private members with m_

Written because I thought we had a bug there, but it turns out
everything is fine.