Hello,
For the record `org-match-substring-regexp' is a variation on:
"\\(\\S-\\)\\([_^]\\)\\(\\(?:\\*\\|[-+]?[^-+*!@#$%^_
\t\r\n,:\"?<>~;./{}=()]+\\)\\)\\)"
I think it is a bit convoluted and therefore difficult to predict. For
example, as recent bug report showed, you may tend to interpret
a_b[fn:1] as
a_{b}[fn:1]
but, in fact, it is equivalent to
a_{b[fn}:1]
Of course, we can prevent this by forbidding "[" and "]" in the last
part of the regexp. But I wonder if there's something better to do.
The idea behind this regexp is that we should be able to write simple
sub/superscript, including numbers and entities, without requiring curly
braces (see `org-use-sub-superscripts' docstring for details). Maybe
something like the following could be an interesting alternative:
"\\(\\S-\\)\\([_^]\\)\\(\\*\\|[+-]?\\(?:\\w\\|[0-9.,\\]\\)*\\(\\w\\|[0-9]\\)\\)"
That is, without braces, either an asterisk or any combination of word,
number, dot, comma and backslash characters, which may start with either
a plus or a minus sign but cannot end with either a dot or a comma.
I find it arguably more predictable (no inverted class). Also, we "gain"
the following:
a^3.14. <=> a^{3.14}.
At the moment, a^3.14. <=> a^{3}.14.
What do you think?
Regards,
--
Nicolas Goaziou
