Towards#5468. This is a small change that greatly speeds up searches for regexes likea+ that start with some letter/string/character class followed by a+ quantifier (or any other quantifier requiring at least one repetition). Because this loop must be matched at least once, we can enter the repeated subpattern and look for the first position a letter/string/character class in the subpattern can match.

While working on this, I noticed that I didn't think the implementation oftext_regex::should_search_match_capture_groups() through well enough: I designed it to use relative coordinates for expected submatches, but this isn't so helpful when one wants to ensure that the whole match is in a particular position. So I changed the implementation to use absolute coordinates from the start of the matched string. Luckily, no test seems to have relied on the previous behavior, meaning all of them just matched the start of the input string anyway.

Benchmark

Running on my machine: