Wow. I wasted hours in FrameSvg trying to find this. Does this also fix the panel mask in non-composited mode, e.g. round corners with Oxygen? (Quickly browsing the bug numbers suggests it does \o/)

In D13215#270878, @broulik wrote:

Wow. I wasted hours in FrameSvg trying to find this. Does this also fix the panel mask in non-composited mode, e.g. round corners with Oxygen? (Quickly browsing the bug numbers suggests it does \o/)

It has nothing to do with compositing. ;-)

I noticed when I was working on this diff that paintCorner is not working _really_ correct. I'll check it with Oxygen.

@broulik I tested panels with Oxygen(with compositing on and off) and didn't see any problems.

(Quickly browsing the bug numbers suggests it does \o/)

Not sure about that. All of those bugs are "Breeze-related"(i.e. bug reporters use Breeze desktop theme).

+1

Doesn't make a difference in Oxygen's rounded corners on panels but doesn't make it worse either. Please also profile whether it causes any needless recalculations on Plasma startup

In D13215#271983, @broulik wrote:

+1

Doesn't make a difference in Oxygen's rounded corners on panels but doesn't make it worse either. Please also profile whether it causes any needless recalculations on Plasma startup

I'd like to dig deeper why margins do not correspond to enabled margins. Maybe, because of

if (!d->repaintBlocked) {
   d->updateFrameData();
}

FrameSVG misses to update margins, and so on.

It would be great to get opinion of @mart and @aseigo . Maybe, there is a better way to fix this.

In D13215#271994, @zzag wrote:

Maybe, because of

if (!d->repaintBlocked) {
   d->updateFrameData();
}

FrameSVG misses to update margins, and so on.

OK, I'm wrong about this part.

I think I got it.

FrameSVGPrivate::updateFrameData updates only frame, maskFrame is updated lazily in alphaMask method, but the alphaMask method ignores previous state of enabledBorders. So, we end up with using stale maskFrame.

E.g.

FrameSVGPrivate::updateFrameData() {
if (!fd) {
    ...
    old key = "10_216_768_1_1__dialogs/background"
    old borders = QFlags<Plasma::FrameSvg::EnabledBorders>(BottomBorder|RightBorder)
    new key = "8_216_768_1_1__dialogs/background"
    new borders = QFlags<Plasma::FrameSvg::EnabledBorders>(RightBorder)

Now, when we call alphaMask,

FrameSVGPrivate::alphaMask() {
    if (maskPrefix.isNull()) {
    } else {
         ...
         oldKey = "10_216_768_1_1_mask-_dialogs/background"

+1 on this! Love it!

Answer to my inline question: updateSizes can be omitted (but I would prefer to leave it). Also, maskFrame->enabledBorders = frame->enabledBorders; above is redundant (maskFrame->enabledBorders is already equal to frame->enabledBorders; it if wasn't the case, then cacheId works incorrectly or FrameData constructor doesn't copy enabledBorders, which isn't the case).

Also,

if (s_sharedFrames[q->theme()->d].contains(oldKey)) {
    s_sharedFrames[q->theme()->d].remove(oldKey);
    s_sharedFrames[q->theme()->d].insert(newKey, maskFrame);
}

ignores refcount, which may cause problems. I won't fix that, because that would be unrelated change.

Yet another "also":

if (maskFrame->cachedBackground.isNull()) {
    return QPixmap();
}

is redundant.

Anyway, I think I finished my work on fixing bugs that are listed in the summary and would like to proceed with review.

In D13215#271983, @broulik wrote:

Doesn't make a difference in Oxygen's rounded corners on panels but doesn't make it worse either.

Could you please share a screenshot? I still don't know what's wrong with Oxygen rounded corners.

Ping.

What can we do to help this patch move forward @zzag ?

In D13215#270880, @zzag wrote:

In D13215#270878, @broulik wrote:

Wow. I wasted hours in FrameSvg trying to find this. Does this also fix the panel mask in non-composited mode, e.g. round corners with Oxygen? (Quickly browsing the bug numbers suggests it does \o/)

It has nothing to do with compositing. ;-)

I noticed when I was working on this diff that paintCorner is not working _really_ correct. I'll check it with Oxygen.

it has nothing to do with compositing, but it *should* fix the non composite mask never the less, as it was coming from a misrendering of the framesvg

In D13215#274782, @mart wrote:

it has nothing to do with compositing, but it *should* fix the non composite mask never the less, as it was coming from a misrendering of the framesvg

Could someone please show me what's wrong with Oxygen corners? (I still don't know what's wrong with them, sorry)