Patterns should not include the separator (unless escaped), even
as part of a character class. In conformance with Posix, the Cray
`sed' rejects `s/[^/]*$//': use `s,[^/]*$,,'.
[...]
Portable `sed' regular expressions should use `\' only to escape
characters in the string `$()*.0123456789[\^n{}'.
Ah, I didn't read those two conditions as applying simultaneously. Since
the second condition essentially says that you must only ever escape a special
character to make it non-meaningful or a normal character to make it special,
maybe the first condition say something a bit more like ...
Patterns should not include the separator (not even escaped, unless
you fancy having to use a semantically significant metacharacter for
your separator), even as part of a ...
No. The problem is exactly when you use a semantically significant
metacharacter for your separator.
\* => unportable, a\* may become either a* or a\*
\> => unportable, a\> may become either a> or a\>
\$ => unportable, a\$ may become either a$ or a\$
\, => unportable, part of \{a,b\}
\| => unportable, a\|b may become either a|b or a\|b and the latter
may be parsed as alternation
\/ => in theory unportable in practice it would be insane to introduce
a special sequence \/
\@ \: etc. => in theory unportable, in practice they should be fine.
All of these are forbidden by the wording in Autoconf manual.
One could add / to the list in the second paragraph, and saying
something like "only / should be used as a separator if you wish to
escape it, to avoid inadvertent introduction or escaping of regular
expression operators".
Or even better, it should be left as is.
Paolo
