GSHorizontalTypesetter improvements: Add Fast Path for Simple Rectangular Containers

[IOOI-SqAR]

I asked the Kimi AI how to further improve GSHorizontalTypesetter. Here is what it came up with:

https://www.kimi.com/share/19d34cdb-8242-8fc9-8000-00001fd46e3b

Add Fast Path for Simple Rectangular Containers

The TODO comment mentions this optimization. For the common case of a simple rectangular text container without exclusion paths, bypass the expensive line fragment rect building:

- (BOOL) _useSimpleRectangularLayout: (NSRect)proposedRect
                          lineHeight: (CGFloat)lineHeight
{
    if (![currentTextContainer isSimpleRectangularTextContainer])
        return NO;
    
    NSSize containerSize = [currentTextContainer containerSize];
    NSRect simpleRect = proposedRect;
    
    // Single rectangle for the entire width
    if (lineFragmentCount == 0) {
        lineFragmentCapacity = MAX(lineFragmentCapacity, 2);
        lineFragments = realloc(lineFragments, sizeof(LineFragment) * lineFragmentCapacity);
    }
    
    lineFragments[0].rect = simpleRect;
    lineFragmentCount = 1;
    
    // Skip the expensive lineFragmentRectForProposedRect: loop
    return YES;
}

I would say we should address this after fixing the line wrap bug first (has been #382 which is closed now (but the issue is not fixed yet))

[rmottola]

Yes, we know this optimisation, but I prefer to implement on it only after #382 is done, as you write.

[fredkiefer]

The code here is incomplete and for most cases it will be no real optimization. In many cases the simple rectangle will result in just on line fragment per call. So we only end up with more code and no speedup. There are other places where knowing we have a rectangle could allow for other optimizations.

…

Am 03.04.2026 um 01:01 schrieb Riccardo ***@***.***>: rmottola left a comment (gnustep/libs-gui#409) Yes, we know this optimisation, but I prefer to implement on it only after #382 is done, as you write. — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>