GitLab

Based on the following warning:
CMake Warning (dev) at /home/alex/kde/usr/share/ECM/modules/ECMFindModuleHelpers.cmake:112 (message):
  Your project should require at least CMake 3.16.0 to use FindKF5.cmake

There is no point in installing systemd tempfiles or startup
services in FreeBSD.

Plain <errno.h> on FreeBSD does not have this value defined,
so libc doesn't know about it (and similarly, fuse3 might not either).

After discussion with Favian Vogt, decide to **not** do any acrobatics
to preserve ENODATA in this one case, but switch the error code to
EIO instead.

steady_clock's epoch is the system start and such a default constructed
steady_clock::time_point might actually not be older than ATTR_TIMEOUT.
By setting it to min() explicitly, it should be sufficiently behind.

According to the man page of stat, st_blocks is always in 512 B units,
independent of st_blksize.

Bhumit Attarde authored Mar 17, 2021 and Fabian Vogt committed Mar 17, 2021

- Implemented opendir(), releasedir()
- Modified readdir() accordingly

Reject mounting slaves with this incompatible behaviour, it can't be mapped
to FUSE semantics completely.

BUG: 433300

- "nonexistant" -> "nonexistent"

Previously open-coded in various places, this commit simplifie those.
Additionally, this fixes a case where the url was logged directly.

The only purpose is basically to avoid the "QUrl()" around the URL in the log.

The KIO documentation for UDS_URL says:
`use UDS_TARGET_URL if you want "links" to unrelated urls.`
So the use of UDS_URL here was probably wrong.
Switching to UDS_TARGET_URL fixes mounting of some slaves such as baloosearch,
which set UDS_URL to the URL of the entry itself for some reason.

BUG: 431079

createNodeFromUDSEntry() can return a nullptr, which it does when
passed any URL from the baloosearch protocol.

BUG: 431079

The previous commit was not quite enough.

Let's not require a newer version than the RC, it's easily avoidable.

Fabian Vogt authored Dec 25, 2020 and Nate Graham committed Dec 26, 2020

First stable release!

Adapt and expand the README for DBus and systemd activation.

Fabian Vogt authored Aug 12, 2020 and Alexander Saoutkin committed Dec 23, 2020

Absolute symlinks have a different meaning when accessed though the regular
filesystem, they don't anchor to the mounted URL anymore:
sftp://server/tmp/symlink -> /home/user/
/mnt/sftp/server/tmp/symlink -> /home/user/
Rewrite absolute symlinks when reading or writing them to keep the target
consistent:
/mnt/sftp/server/tmp/symlink -> /mnt/sftp/server/home/user/

It is necessary that KIOFuseVFS is aware of the mountpoint now, to be able to
deal with absolute paths. This is not ideal as it doesn't account for bind
mounts or symlinks.

Use readlink in fileopstest to get the raw link target instead of an absolute
path.

BUG: 422090

Wraps KIOFuse in systemd service `kio-fuse.service` when dbus activated, while still working if systemd is not present, in the `background` slice as per https://systemd.io/DESKTOP_ENVIRONMENTS/

This provides systemd logging and resource control

This is to ensure that it doesn't end up creating a node with invalid name.

This avoids duplicating / and so sanitizeNullAuthority is no longer needed.

Instead of always creating an origin node at the authority level
and creating all path elements as directories, only create everything
up to the origin when mounting. Everything beyond that is created
on-demand when an application triggers readdir or lookup.

Some ioslaves like tar don't support access at the root level
(tar:/// doesn't exist), so for mounting those it has to traverse the
URL upward until a valid origin is found.

As mountUrl no longer creates all nodes to resolve the full URL, it
returns a string in the callback now instead of a node.

As there is no longer a concept of a protocol node, it was replaced
with a plain DirNode instead.

remoteUrl had to be fixed to deal with m_overrideUrl having a non-empty
(or '/') path, it didn't add '/' before.

The algorithm of localPathToRemoteUrl was changed to not need all path
elements as nodes. It stops traversing once an origin was found.
This means it behaves incorrectly/unexpectedly with nested origins.

BUG: 422091
CCBUG: 422090

Fails currently as directory symlinks aren't handled by those yet.

This makes the caller responsible for ensuring it has the proper type.

Those are just KIOFuseDirNodes with different name, no need to distinguish.

Convert some uses of the type member into dynamic casts or typeid
comparisons. Has slightly more overhead, but in the success case
the result is needed anyway.
Also remove some unnecessary Q_OBJECT macros while at it.