Subtitles without spaces do not break to a new line when overfull

[svenper]

by way of mpv-player/mpv#7158 :

Expected behavior

Subtitle lines without spaces (Chinese, Japanese, etc.) wrap automatically when too long.

Actual behavior

Such lines extend outside the window, with characters cropped.

See e.g. at 43:00 in sub.ass with a large font size:

Dialogue: 0,0:43:00.01,0:43:06.16,Default,,0,0,0,,任何一个有远大抱负的男人都不会因为区区一个女人

(Many lines in the subtitles do indeed have spaces between short phrases, which is an exeption to the rule in East Asian scripts.)

[siikamiika]

Found this function while investigating this myself:

libass/libass/ass_render.c

Line 1588 in 6a670b4

wrap_lines_smart(ASS_Renderer *render_priv, double max_text_width)

I'd implement line breaking so that first priority would be for spaces as listed here, and if the line would still be too long I would break before any CJK character that goes above max_text_width.

However, I see the problem that the function loops the text char by char instead of unicode characters. I'm not sure if it can be assumed to be UTF-8.

[ghost]

It's definitely UTF-8, except when it's not. The latter could happen if an .ass file contains invalid UTF-8 sequences, but then libass doesn't need to render anything sane. But it shouldn't crash or so.
Due to the properties of UTF-8, looping by char is not a problem. If you need to skip codepoints that consist of multiple bytes, then I guess the current code structure is going to be very annoying.

[ghost]

PS: if you haven't seen it yet, ass_utf8_get_char should be useful to compute the length of UTF-8 sequences (although there are faster methods if you don't need the codepoint itself).