In D12509#253568, @rkflx wrote:

The maximum size available on all of / for that icon is 64px. Despite that, Gwenview is able to show a good looking thumbnail in one case, but not the other.

In D12509#253558, @rkflx wrote:

Ah, maybe I used unclear terminology for "thumbnail". My point was that for both the relative and the absolute patch the same icon is chosen (i.e. the GNOME foot) and rendered in the same quality, but in one case the preview thumbnail of the image painted over the icon is pixelated, while it is fine for the other case.

The ingredients are the same: Same image to show a thumbnail for, same folder icon. Only in one case the thumbnail is not rendered optimally.

In D12509#253527, @rkflx wrote:

Doesn't your approach mean that now the thumbnail quality will be worse for absolute paths?

In D12465#253539, @rkflx wrote:

This is caused by KIconLoader::loadMimeTypeIcon() failing for absolute path entries.

Have not looked into it too deeply, but I think an important question to ask here is if this difference in behaviour compared to KIconLoader::loadIcon() is by design, or if it is a bug. If it is the latter, maybe we should actually fix the library?

Set splitter sizes only if config value exists

@huoni Take the time you need. ;)
Thinking offline about this a bit... I just noticed a problem with a clean Gwenview config - there should be a default or prevent setting in this case. Otherwise the sidebar doesn't show.

In D12458#252812, @rkflx wrote:

@muhlenpfordt For now, just use --skip-dependencies.

Looks good. :)
Only arc patch fails:

Cherry Pick Failed!
Exception 
Command failed with error #1!
COMMAND
git cherry-pick 'arcpatch-D12300'
[...]

Added const

Dolphin also uses KIconLoader::loadIcon() - this version loads the same mime icons as KIconLoader::loadMimeTypeIcon() but correctly handles absolute path entries.

Removed unnecessary modifications

The "Save As" dialog always starts in the source file's directory now and not in the previously selected but failed one.
Is it desirable to restart in the previous folder?

In D12300#249693, @rkflx wrote:

This bug was the reason for adding Gwenview's check: dc2188907b54.
KFileDialog did not support its own confirmation until v4.2: KFileDialog::setConfirmOverwrite(). After introduction it seems the default was disabled.
QFileDialog's confirmation dialog is default now: QFileDialog::Options.

The markup doc: KI18n -> Predefined Tags

In D12302#inline-62294, @rkflx wrote:

Agreed, but please split this out into a separate commit on master (no need for a Diff).

In D12302#249282, @rkflx wrote:

In D12302#248599, @muhlenpfordt wrote:

There are two console warnings caused by wrong markup for xi18n*:

doc->errorString() is also used in GvCore::saveAs and in ImportDialogPrivate::deleteImportedUrls. Are you referring to those?

Removed Url shortening option

In D12246#inline-61983, @rkflx wrote:

Agreed, but please split this out into a separate commit on master (no need for a Diff).

Added const / Readded setting mCurrentMainPageId

Works good for me.
Only when I modify pixel size too small or too big the percentage values does not update below 1.00 % or over 999.00 %.

Guide on how to add markup containing arguments to xi18n*: KI18n -> Escaping

Tested in a VM with Kubuntu 17.10 and with xsm.

In D12105#246382, @rkflx wrote:

Hm, for me the testUndoRedo test does not fail when I have not applied the rest of your patch yet.

Remove 'cleanup' changes - TODO for other patch
Added Qt version check for QUndoCommand::setObsolete()

In D12187#246380, @huoni wrote:

This says FIXED-IN: 18.04.1, but is branched from master... in case you weren't aware.

Rebase to Applications/18.04
Changed comment

In D12105#245019, @muhlenpfordt wrote:

Added DocumentTest::testUndoRedo()

Added DocumentTest::testUndoRedo()

In D12105#244953, @rkflx wrote:

Meanwhile I hope you don't hate me for saying this, but to me it looks like that you should add an autotest or extend an existing autotest, so that Redo does not regress in the future… Could you try to work on that?

Looks excellent - aside from my color scheme... ;)
Tested with different raster and svg files, including these special ones posted above and found no issues. :)
Also tested the HiDPI patch but I'm not sure if I should see any difference without the corresponding hardware and/or system config?

Rebase to Applications/18.04

Is this a fix for 18.04.0 or better master only?

In D12059#243124, @rkflx wrote:

What colour scheme are you using?

Seems it's a combination of my background color and the selection frame color. Using a white background it looks totally different - probably there's no perfect color.

Seems to be really unnecessary. ;)
I can't spot anything unusual, no matter what Animations option is set.

Works good and the frame itself is drawn perfect. :)
But I would prefer keeping the gap because the frame is nearly invisible if the image has similar colors.
I think the summary is detailed enough and I can follow the explanation. ;)

Since April 5th the PlaceTreeModelTest on Jenkins fails. I can't reproduce this, on my system the test is ok.
Is this something we should worry about?

Rebase to Applications/18.04
Changed comment

Rebase to Applications/18.04

In D11715#242174, @rkflx wrote:

BTW, Redo is quite broken (unrelated to your patch, though)…

In D12023#242187, @rkflx wrote:

Ah, got a suspicion: Did you use git checkout -b <name> or arc feature <name> (which does the checkout for you, but also calls --set-upstream-to=…)?

Removed unnecessary semicolon

On one of my systems (Kubuntu 17.10) this bug happens only sometimes, so much for Comment 1. On two other it's always reproducible (Kubuntu 17.10 and 16.04).