discuss-gnustep
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Art backend Was: Package building


From: Sergii Stoian
Subject: Re: Art backend Was: Package building
Date: Mon, 9 Dec 2019 01:35:41 +0200
User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:68.0) Gecko/20100101 Thunderbird/68.2.1


On 08.12.19 14:57, Fred Kiefer wrote:
Am 08.12.2019 um 13:40 schrieb Sergii Stoian <stoyan255@gmail.com>:

On 8 Dec 2019, at 10:12, Fred Kiefer <fredkiefer@gmx.de> wrote:

Am 08.12.2019 um 03:07 schrieb Sergii Stoian <stoyan255@gmail.com>:

I suppose it’s not direct art problem. Perhaps art should be adopted to 
NSStringDrawing changed caching, I don’t know until I understand string drawing 
change. What was the idea of that rewrite? Commit message doesn’t explain why 
that change was made. For me, It doesn’t look as “cleanup”. That’s why I asked 
you as author of that change.
What I changed back then was to have the scratch item that gets used for the 
current drawing to be of the same structure as the cached items. The scratch 
item now even gets stored in the same array as the cached one, so the 
initialisation is identical. This really simplified the code and helped me 
understand it better to allow for further restructuring. Try to understand the 
code before and after that change and you may see the difference. The code is 
still complicated but at least the data structures are cleaner.
Thanks, now I understand your intention.

I also switched to using explicit types, for example for boolean values. But I 
may have made a mistake there. On line 501 (and again on line 674) I replaced 
the value 1 with NO and this looks wrong. As this stands for using screen fonts 
it would affect the art backend, being the one where we actually support screen 
fonts.
I cannot remember what was the reasoning behind that change. The difference 
between the actual drawing code and this bounds computation is that during the 
drawing we have the transformation to the screen and are able to determine 
whether this is close to identity, in which case we use screen fonts. The 
bounds computation may happen at any time, so we don’t know if screen fonts 
should be used. Maybe the art backend relies on these being used anyway? You 
could give it a try and use YES here instead. The lines in the current code are 
513 and 606
That’s it, thank you! I’ve changed lines you’ve mentioned and problem gone. 
What’s the point to change YES to NO? Was it intentional (some cairo backend 
specifics)?
I’ll prepare PR once I setup environment and compile master branch to test 
changes.
As I wrote, I cannot remember whether I did that on purpose or not. Both is 
possible. Both values are wrong, in some sense as I explained. That is why I 
might have changed them. But it is more likely that I just made a mistake here. 
Otherwise I would have documented the „improvement“ :-(

I've created a PR #36. It's plain simple revert NO to YES at lines you've mentioned.

However, another problem exists at master branch. One of the effects is shown on my WorkspaceMenu+HelpPanel screenshot: selection of index item shifted vertically. This problem appeared on 0.25.1 branch before NO/YES change, but got fixed after change. It's not the case with master.

I suppose something wrong with empty string height detection (LayoutManager?). I need to investigate further.

Sergii




reply via email to

[Prev in Thread] Current Thread [Next in Thread]