FTR, this is how SkParagraph determines whether if a cluster is a space (or break).
https://github.com/google/skia/blob/8b2f189cfa59d0c69b700bbe8587f52f5bd2ad5c/modules/skparagraph/src/ParagraphImpl.cpp#L357-L381

With the ICU4C backend, ::kPartOfWhiteSpaceBreak is
https://github.com/unicode-org/icu/blob/23d9628f88a2d0127c564ad98297061c36d3ce77/icu4c/source/common/unicode/uchar.h#L3388-L3425, so this list should probably be a bit different from the Zs category.

Also you could use regex for matching unicode properties, e.g.: RegExp(r'\p{Space_Separator}', unicode: true)

Also you could use regex for matching unicode properties, e.g.: RegExp(r'\p{Space_Separator}', unicode: true)

These 3 lines are no longer needed it seems?

Ah I see you needed A's horizontal advance. It's 14.0.

Yes. I checked the value at runtime and found it to be 14.0, but since it depends on the style of the painter, I calculated it.

nit: charset -> character

Thanks, I fixed at 1e1d000.

Can you also verify that each input character is indeed a space, by laying out the painter at maxWidth = 14, and checks if there's only one line, and the line width is exactly 14? If one of the characters is not considered a trailing space then there will be 2 lines (or lineWidth would be greater than 14 if the character is not a line breaker).

I added test case at 9c37308.

painter.width is not the line width. I think this should be:

final lines = painter.computeLineMetrics();
expect(lines.length, 1);
expect(lines.width, 14.0);

I added test case at 86adcc1.

All of these should be 1-line. If it is not then the character is not considered a space (during line-breaking) by SkParagraph, so the character must be excluded from the list.

nit: "character"

This is still checking the last code unit. I don't see the difference between this and the expression before the change?

Also I'm curious how the regexp fares with surrogate pairs, could you also test a surrogate pair ("\uD801\uDC37" for example)?

Ah I tested locally and it didn't throw. So up to you if you want to skip adding that test.

There's little point in using a switch expression now I think? This could just be an ||.

Readability nit: it's not immediately obvious what this is checking. This is verifying that:

1. character is indeed considered a trailing space by SkParagraph
2. TextPainter's implementation matches the SkParagraph implementation

So consider renaming this function to something like "verifyCharacterIsConsideredTrailingSpaceBySkParagraphAndTextPainter" (or turn the "by skparagraph and textpainter" part to a comment since the name is a bit long).