Capture Groups

(pattern) captures the substring matched by pattern for later retrieval.

Numbering: groups are numbered left-to-right by their opening parenthesis. Group 0 is always the whole match.

Named captures (Python (?P<name>...), JS (?<name>...)):

Implementation in NFA: each capture group adds two "save" pseudo-states:

• One at the open paren — saves the current text position as group-N start
• One at the close paren — saves the current text position as group-N end

When the simulation transitions through a "save" state, it does NOT consume input but updates the per-state save list. The accepting state's saves become the captures.

This adds O(N*K) overhead where K = number of groups. Still linear in text.

In backreferences \1, the engine compares the next text characters against the previously captured text — this is what makes them non-regular and forces backtracking.