Defer innerHTML serialization in the "Grabbed" debug log

[fivefilters]

What

grabArticle() logged the grabbed article with:

$this->log('Grabbed: ' . $articleContent->innerHTML);

PHP evaluates function arguments eagerly, so innerHTML serialized the entire article subtree into a string on every parse — even when no logger was configured and debug was off. log()'s early return then discarded it. The serialization work (and a large transient allocation) happened before log() was ever entered.

This changes the call site to pass a closure, and teaches log()'s formatter to resolve any \Closure argument — but only after the enabled check that guards the method:

$this->log('Grabbed:', fn (): string => $articleContent->innerHTML);

When logging is off, the closure is never invoked and innerHTML is never built. When logging is on, the emitted message is byte-for-byte identical to before.

Why a closure, not a change to the element formatter

The formatter's \Dom\Element branch is shared by ~12 other call sites (Candidate:, Looking at sibling node:, setNodeTag, Cleaning Conditionally, …), many inside hot per-node loops. They rely on the compact <div class="..."> representation. Emitting innerHTML for all elements would flood those logs with unreadable full-subtree dumps. Deferring via a closure scopes the change to this one line and leaves every other element log untouched. The formatter now also resolves closures generically, so other sites can defer expensive values the same way.

Testing

Verified with a standalone script (composer/phpunit unavailable in this environment):

• Behavior unchanged: a real parse() with a PSR-3 logger emits exactly one Grabbed: message containing the serialized innerHTML.
• Deferral works: reproducing log()'s structure with a counting closure, the closure is invoked 0 times when logging is off and once when a logger is present.

php -l passes.

🤖 Generated with Claude Code

https://claude.ai/code/session_01BXHSUavFZ8SkC5kX6j59BV

---

Generated by Claude Code