I really, really hate these things.
I've been testing this patch more than one month without any problem.
And now, after pushing it, I got crashes everytime I start dolphin without parameters (user home).
But not if I start it in / and keep changing directories, including user home.

Incorporated Stefan code.
Fill again the extra data like User/Group.
I don't see any change and I don't want to break anything more.

Added a comment: // readlink doesn't append a null byte to linkTargetBuffer.data()
Using target path to read ACL in a non broken symbolic link.
Do not fill remaining data in case of broken link, doesn't affect dolphin.

Renamed isSymLink to isBrokenSymLink.

Commas at the end.

Remove the friend non-exist class.

Added KIOCORE_DEPRECATED

Now better.

But ForwardingSlaveBase::prepareUDSEntry in KIO itself, uses insert() to replace UDS_URL, and in fact all of ForwardingSlaveBase should use replace(), as well as users of ForwardingSlaveBase like kio_desktop.

Call items.append inside the loop.

In D10342#286023, @hein wrote:

This patch doesn't make any sense. It's setting up a cache for something computed from data that's subject to change, and it's never evicting it when that data changes.

In D13813#286313, @aacid wrote:

Did we break how KIO works and we're now going after the users of KIO and fixing them or was this a "bad usage" from starters that worked by chance?

Undo the right number of steps from a quick experiment. :-)

In other words, the fix for that can probably be removed now.

I prefer not to do it (even I've tried without any crash), because QVariant is not even reentrant.

Can not use the name lessThan because of: error: use of ‘lessThan’ before deduction of ‘auto’

In D13814#285691, @bruns wrote:

I assume you are sorting by name with "natural sorting". There may be another possibility for speedup here:
QCollator::sortkey(), http://doc.qt.io/qt-5/qcollator.html#sortKey

It might be possible to generate the sortkey in e.g. https://phabricator.kde.org/source/dolphin/browse/master/src/kitemviews/kfileitemmodel.cpp$1334

Remove the _ prefix.

In D13814#285686, @bruns wrote:

In D13814#285660, @markg wrote:

I don't know if that's still an issue or if your patch re-introduces whatever the problem was (race conditions?). You could look back in the commit log when that was added to figure out more about it.

To answer that myself, it was done in this commit: https://cgit.kde.org/dolphin.git/commit/src/kitemviews/kfileitemmodel.cpp?id=d9680ead8099df9a2b06bfed61a62923778996f2
And doesn't explain anything :)

If you follow the link in the comment, i.e. https://bugs.kde.org/show_bug.cgi?id=312679, it mentions date sorting and KDateTime being non-reentrant.

Implement only the windowUrlFromMetadata cache.
It avoids a 100% plasma cpu usage when using vivaldi in private navigation.

Use 3 entries instead of replace().
Now it will work with any kio version.

Updated the patch with context.

Moved the comment.

Removed comment and added autotest.

Using NormalizePathSegments.

Your suggested code works.

You're right :-)

In D13475#278098, @lbeltrame wrote:

I'm backporting it. Before I push: will this work also with previous KIO versions?

In D13475#277938, @elvisangelaccio wrote:

Should we backport this on the stable branch? Otherwise affected people will have to either downgrade kio or compile kio-extras from master...

Change @since to 5.48

More numbers:
The patch reserves 67 buckets in the QHash table vs. 37 by default.
In my test, pasing the content of https://www.boe.es/legislacion/documentos/ConstitucionCASTELLANO.pdf from Okular,
the number of items in the cache is a maximum of 24 members when inserting.
But the first undo uses up to 2300 members (4099 buckets), and later undoes only use a maximum of 43 members.

I misread what was after the goto. :-(
Modified also the similar method createUDSEntry.

I'm sorry, I was missing a -dev package and therefore I was not compiling kio sftp.

In D12897#276182, @dhaumann wrote:

@mwolff To me this looks ok - do you see an issue with this? E.g. that KTextEditor will require much more memory for almost no gain?

If there is a more constrained limit, this should be changed, of course.

In D12696#275105, @martinkostolny wrote:

Hi! Probably after this commit sftp slave crashes when showing a directory with links. Please see my code comment. Can you also reproduce or is it on my side only?

I'm sorry, I can't reproduce it with dolphin. (symbolic links and hard links and sftp://127.0.0.1).
Could you get a backtrace with more information, at least who is calling udsentry.cpp, line 107? Probably you can get it running the program inside gdb or under valgrind. Or you should compile the program and dependencies with -DCMAKE_BUILD_TYPE=RelWithDebInfo

Changed the structure QListBinaryHash to QMap
Changed from KFileItems pointers to Values (it caused memory problems).

The change in kdirmodel is not needed anymore.
The methods that were const are const again.
Better documented the new reinsert method.

Now passes the tests and its performance for non invalid items is not degraded too much (same +3ms inserting).