invert matching regex

Issue raised by @andy and @Mattijs - could be useful to have an inverted regex matching. Eg,

   "(?!).*processor.*"

to match anything that does not contain 'processor'

We already handle the PCRE-prefix "(?i)" - so relatively simple to extend the prefix handling.

cross-ref: EP1748

assigned to @mark

In the original versions (POSIX), the regex was just a wrapped pointer (8 bytes). Adding an invert-flag will automatically bump that to 16 bytes. Not really a problem since we mostly use C++11 regex versions anyhow. However, the plain C++11 regex is around 32 bytes. After tacking on basic validity flags etc, this grows to 40 bytes.

This means that wordRe is 72 bytes vs 32 bytes for a plain word. Might reconsider attaching the regex portion of wordRe to a unique_ptr.

changed the description

Hi @mark, as discussed in EP1748, the processorCyclic might be not excluded. So the regex should be updated to "include" the processorCyclic.

The regular expressions currently have limited abilities which makes it difficult to combine both include and exclude logic into the same expression. Perl-compatible versions have various types of negative lookback and lookahead, but I don't know how well the C++ versions manage that. Some c++ regex had problems with ECMAScript grammar, so we currently default to posix. The OSspecific POSIX version will be even more limited still.

The expressions that we would/could devise would probably be quite unreadable. Better to have explicit allow/deny lists with wordRes (cf. filtered copying).

mentioned in commit 770f821f305bc21d0e3961f7e83f140186513739

mentioned in commit f602d318b8e0cd4b0be9cae8adea6414f4a0fc6f

changed milestone to %v2112

mentioned in commit e0929809

closed