GitLab

use unique_ptrs instead. more modern this way, also less chance
of getting it wrong

they are entirely unused

they hide unhandled enum values when we want and need to know that they
are missing! also the return explicitness was sometimes lost in the process
(when emitting finished or error we must return as otherwise assertions inside
the base class may fail)

This error message is already quite good, but can be made even better by
explicitly mentioning the need to allow access, which is typically done
by tapping "Allow" on the popup that appears on the device's screen when
you plug it in.

readChunks returns an int, use const auto instead of size_t to be able to
compare against -1.

Specifying only "none" as compression method causes the exchange to fail if
the server doesn't support "none".

Use the same options as OpenSSH does by default. The Compression=yes/no option
in the config file overrides them if specified.

Setting the options explicitly can be dropped in the future after libssh
adjusts its built-in defaults.

this causes the kdirmodel in the UIs to reload when devices change and
thus pick up changes without having to hit f5

removal is on a 5s delay because removing causes the models to switch
out of the directory, so we must only emit removal after we are sure the
device disappeared rather than temporarily getting reconnected with
storages

one can but wonder why that was an int to begin with :(

this is always true and never called with false, it serves no purpose
but increasing complexity

James John authored Jan 05, 2022 and Harald Sitter committed Jan 05, 2022

MTP storage needs to be real-time. After storage permission is
allowed, storage count is still empty, causing you to relaunch
Dolphin to be able to access devices.

I have been having this issue for as long as I can remember.
This commit is refreshing device list before querying device list
in case new device information has arrived. This ensures we always
have the latest storage information.

BUG: 447319

using a vendor specific mimetype and 0x0 access (the thing can't be
edited!) and with explicit erroring when trying to stat (which should
for the most part prevent KIO from opening the file by doing a mimetype
test on it)

this facilitates smb browsing of printers in printer-manager as it can
simply filter for the vendor mimetype now

By creating a macro for this, we avoid copying the same bit of boilerplate all over the place.

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"

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"

Martin Tobias Holmedahl Sandsmark authored Nov 23, 2021 and Harald Sitter committed Nov 24, 2021

The Qt X11 integration checks if the variable set, not whether it is
empty, so we need to unset it to avoid a warning about failing to
connect.

Martin Tobias Holmedahl Sandsmark authored Oct 29, 2021 and Martin Tobias Holmedahl Sandsmark committed Nov 23, 2021

Similar to multimedia/ffmpegthumbs!11

This avoids having to update the list every time a new image format is
added to e. g. kimageformats (now it's missing heic etc.).

This way thumbnailer plugins don't need to be loaded using the KServiceTypeTrader
mechanism, but can use embedded json metadata.
KServiceTypeTrader is planned to go away in KF6.

This also simplifies the code once the KServiceTypeTrader code path is dropped, because
we can then load the metadata from the binary path and don't have to use
KServiceTypeTrader to get the desktop file for the given binary.

By using KPluginInfo to convert the metadata to json on the fly, we don't need to
do another query for the file on disk and then a reparsing of the desktop file.
This is ~2.5 times slower and even requires a few more lines to implement.

Counterpart of frameworks/kio!630

Task: https://phabricator.kde.org/T12179

With the standard header, styles etc as provided by kdoctools.

7d11a093 accidentally moved the
memory pointer increase out of the '+'-handling branch, which would
either result in bad number being read or the pointer getting behind
the end and resulting in garbage memory processing, potentially ending
in a crash.

Also changing to using dedicated own variable "sign" instead of reusing
function-global variable "j" with meaningless name.
global

Fixed-in: 21.08.3
BUG: 443983

Error reporting was changed in KCoreDirlister, to match the original behaviour
we need to explicitly disable auto error handling.

This fixes a crash while searching in Dolphin, kio_filenamesearch intrenally
uses a dirlister, and a nested event loop, if there is an error while listing
some dir it tries to report it using a job ui delegate, but the parent job
could well be dead by that time, which causes a crash:
"QWidget: Cannot create a QWidget without QApplication"

Note that this crash only happens if KDE_FORK_SLAVES is _not_ set.

BUG: 438187
FIXED-IN: 21.08.3
(cherry picked from commit 5dff395e)

this was previously wildly duplicated. createUDSEntry has been refitted
to support both calls from stat() and from listDir().

createUDSEntry now takes an incoming SFTPAttributesPtr (taking ownership
of the attribute) and fills the udsentry accordingly. for listDir
efficiency's sake we pass in the entry by reference as well as the
details such that we can avoid needless work where possible.
the function now returns a conclusive Result, callers may naturally
ignore it.

listDir no longer aborts on readlink errors. indeed any errors coming
out of udsentry creation shouldn't impact listing. just because a file
fails to list doesn't mean the entire listing should stop 🤷

listDir now ignores readlink problems. there's no reason why failing to
read links would abort the dir listing

it's poor form and gets in the way of readability

divide openConnection in the wrapper function openConnect and the
internal openConnectionWithoutCloseOnError. the former calls the latter
and automatically calls closeConnection if an error was returned. the
internal function no longer calls closeConnection at all

also cleaned up style oddities while stepping through the code (fewer
line breaking, fewer useless var assignment, no longer `else if` after
break/return)

it's way too large, way too complicated, way to repetitively concerned
with cleanup. fingerprinting is a fairly isolated bit, so split it into
a helper function. this also helps reduce the in-scope variables and in
turn makes it less likely that we forget to clean up something

also use scopeguards for cleanup in the new function. also to reduce
repetition and ensure cleanup