Hi,

(Jason, the bits I'd like you to weigh in on are a few paragraphs down.)

Enclosed is a patch pushing the roff_getstr functionality directly into 
libmdoc.  It works by testing against roff_getstr() in-band and splicing 
together a new buffer if necessary.

I thought about putting the entire mandoc_special() check in libroff, 
but don't want to cause yet another scan over the line buffer. 
check_text() needs to warn against '\t' and '\b' anyway.  This is an 
open question I'll answer later when I start looking at performance.

The reason I want to air it with you (I know it works: I've tested it 
across all manuals) is because it also removes the check for isprint(), 
using strcspn() instead.  As you can see, the rej filter is only for 
'\b', which we must prohibit else we boff output encoding; '\t' for 
non-literals (warning); and '\\' for the specials check.

I argue for lifting the ASCII-constraint because (1) there's nothing in 
mdoc/groff/etc that disallows non-ASCII (e.g., Latin-1) characters and 
(2) it makes the code much cleaner.

Thoughts?

Kristaps

PS, the patch doesn't mandate '\b': I just caught that now and will fix 
it later.