[edalzell]

Comb's boolean mode (the default) checks +word with an unanchored lookahead, #(?=.*zakat)#iu. On a record without the word, it retries from every position and scans to the end each time: quadratic in record length.

On a real site (2.4k–3.7k documents, 5–16 MB per index), +zakat -ramadan took 9.7s, 18.5s and 26.9s on three indexes, versus ~0.5s with non-matching documents filtered out first.

Fix: one preg_match('#'.$word.'#iu', $record) per required word, a linear scan each. Anchoring the old pattern instead hits pcre.backtrack_limit on records over ~1 MB and silently drops them.

Words are still not preg_quoted, same as before. One intentional change: required words on different lines of a multi-line field now match (. didn't cross newlines).