voku / simple_html_dom / 31980434841

97%

push

github

Add agent-loop workflow tooling and explicit PHP 8.4 Html5DomParser (#150)

* Add agent-loop workflow tooling and opt-in PHP 8.4 HTML5 parsing

WHY

Two things at once, on purpose: the governed workflow is the tooling, and the
first task run through it is the PHP 8.4 HTML5 parser. A previous attempt at
the parser (PR #146) bridged "\Dom\HTMLDocument" back into "\DOMDocument"
node by node, lost more performance than it gained, and was reverted.

Tooling

voku/agent-loop needs PHP >= 8.3 while this library still supports PHP >= 7.1,
so it is installed as the isolated Composer tool project tools/agent-loop and
reached through the bin/agent-loop wrapper. The generated client projections
(.claude, .codex, .agents, .github/skills, .github/agents) stay out of Git; the
Claude and Codex hook bundles are host-owned copies under docs/agents, because
the package-owned hooks resolve only the root vendor/autoload.php and otherwise
claim this repository's src/ as the package source root.

Board and workflow state live under .agent-loop: SHD-1 is this change, planned,
approved, validated, reviewed and closed through the workflow; SHD-2 .. SHD-8
are the follow-ups it produced, five of them backed by validated findings about
the workflow itself.

Parser

"HtmlDomParser::useHtml5Parser()" parses with the PHP >= 8.4 HTML5 parser and
bridges the result into the "\DOMDocument" this library hands out. There is no
zero-copy handoff - PHP refuses to pass a node of the new DOM implementation to
the old one - so the bridge serializes to XML and re-parses. "Dom\HTML_NO_DEFAULT_NS"
keeps the tree out of the XHTML namespace, and an "xmlns" attribute from the
input is parked under a placeholder for the transport, so the generated XPath
of this library keeps matching.

The default path is unchanged and costs one boolean check. The HTML5 path falls
back to it - without an error - on PHP < 8.4, with useKeepBrokenHtml(), and for
input the XML transport cannot carry.

Validation

  php ... (continued)

83 of 85 new or added lines in 2 files covered. (97.65%)

2266 of 2341 relevant lines covered (96.8%)

308.16 hits per line