Negative/huge Alignment values in ASS Style lines

[astiob]

I stumbled upon this while doing #261, but this is probably best discussed separately.

• When an ASS (but not SSA) script contains a Style line with the Alignment field set to a reasonable negative value, VSFilter ignores all alignment override tags in lines with this style. If we want to emulate this, we need either a new bit in alignment values or a separate new flag in the ASS_Style struct. Both options seem to break the ABI, although I don’t know what promises if ever we’ve given about the alignment values or what assumptions libass users make about them. Even if we tried to add a hidden field to ASS_Style like we did with ASS_Image in ASS_ImagePriv, we would still need to bump sizeof(ASS_Style) because we have an array of them in the public ASS_Track.

• When an ASS (but not SSA) script contains a Style line with the Alignment field set to (the value of) INT32_MIN, VSFilter tries to take abs of this value, which of course returns the same value, and then does several inconsistent things when aligning the event. I have described this in detail in the relevant commit in Integer overflow fixes #261. Do we want to emulate this?

  If so, then we probably also want to store alignment in int32_t with the appropriate wraparound for values larger in magnitude than 2³¹, to ensure we handle those values the same as VSFilter as well. This (and another parsing quirk whereby all integers other than in override tags should accept the &H and 0x prefixes like colours do) applies to other fields in style/dialogue lines too, so it would perhaps make sense to do it for all of them simultaneously. The question of INT32_MIN Alignment still stands though.

Do such files exist?

[astiob]

Personally, I don’t recall ever seeing one or seeing anyone mention these issues. Then again, my collection isn’t huge and there are entire communities (e. g. Arabic or Chinese subtitle authors/users) that I don’t interact with at all.

[realfinder]

right now? I dont think there are files like these, the vast majority do it in aegisub with vsfilter

there are very few do subtitles using aegisub with libass, and they advise using MPV, I know one of them stop doing that and he now seek for some aegisub automation to make his files work with vsfilter

In my opinion, I hope libass be more compatible with vsfilter so vsfilter can dead in Peace, but since there are many issues not fixed since 2014 I think vsfilter will live as long as libass still not work as vsfilter (like by not fixing #148 and #71)

if libass can be compatible with vsfilter in more than 95%, and people start use it even in windows and left vsfilter this issue will no longer be issue