Re: Side-effect of change to scan-code.l

[Paul Eggert]

I think Paul Hilfinger was on the right track.
It's better to use lex and regular expressions
to specify rules for tokens, rather than to use
ad hoc C code.  The proposed ad hoc code would
treat identifiers differently after "$", which
sounds like a recipe for trouble.

Under the previous rules, an identifier could not
begin with a number, because that is confusing.
So, for example "$35.x" does not contain the
identifier "35.x"; instead, it is "$35" followed
by the identifier ".x".

We should be consistent, and continue to do this.
An identifier should not be allowed to begin with a number,
even a negative number.  So, for example, "$-35.x" should
not contain the identifier "-35.x" because "-35.x" should
not be an identifier: it should be "$-35" followed
by the identifier ".x".

We can do this by using something like the following:

 letter  [.abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ_]
 id      -*(-|{letter}({letter}|[-0-9])*)
 ref     -?[0-9]+|{id}|"["{id}"]"|"$"

and then "id" would be treated consistently everywhere.