Comparing changes

* perf(css): number properties for the declaration order check * perf(css): number properties in the longhand fold walk * perf(css): find an :is() to unwrap during the pseudo name fold * perf(css): keep source map anchors in one flat array * perf(css): size a taken piece's span list to what it holds * perf(css): read only the merge entries a block's properties lead * perf(css): print each node type in a function of its own * perf(css): look for a universal selector before walking the parts * perf(css): spell each small source map delta once * docs(css): keep the _unwrapLoneIs description to two sentences

…22825) * test: split the syntax unit tests into tokenizer, parser and printer Each language's tests now follow its modules: a tokenizer, a parser and a printer, whose file holds both print modes and the minifier that drives the printer in its minify mode. - css/, html/: syntax splits into tokenizer, parser and printer; the SourceProcessor tests go to the parser when they walk the tree and to the printer when they rewrite what it writes - javascript/: the tokenizer blocks leave parser for tokenizer, minify (jsMinify) joins printer, and webpack-parser becomes module-parser, after `module.parser.javascript`, the layer it tests - xml/, json/: minify becomes printer Every test keeps its name and every snapshot its key; each new file requires only what its tests use. Claude-Session: https://claude.ai/code/session_017TkBWTk3wcyUcJYDp5ZhaP * test: keep JavascriptParser's tests with webpack's classes javascript/ holds the syntax: tokenizer, parser, scope and printer. JavascriptParser is webpack's layer over it, the one plugins tap, so its tests go back to the top level next to NormalModule's and the others. The layout docs give tooling/ its own sentence. Claude-Session: https://claude.ai/code/session_017TkBWTk3wcyUcJYDp5ZhaP * test: cover the CSS tokenizer from its own tests The tokenizer suite now reaches every function of the tokenizer and all of its statements and branches but five that no input can reach: the end-of-input forms of `@`, `#` and numbers, a bad url running to the end, TokenStream's mark/restore, whitespace and prelude skipping, and the name and range helpers. Claude-Session: https://claude.ai/code/session_017TkBWTk3wcyUcJYDp5ZhaP * test: cover the HTML tokenizer from its own tests The tokenizer suite now reaches every function of the tokenizer and all of its statements and branches but sixteen that no input can reach: the content modes a start tag or fragment context switches to, NUL in every text state, out-of-range character references, and tokenizeXml's tags, attributes, comments, CDATA, processing instructions and DOCTYPE. Claude-Session: https://claude.ai/code/session_017TkBWTk3wcyUcJYDp5ZhaP * test: share the HTML tree helpers between the HTML suites The tokenizer, parser and printer suites each carried their own copy of the helpers that materialize the struct-of-arrays tree (childrenOf, materialize, parseHtml, child, html, body) and of the Mat* typedefs. They now live once in test/helpers/materializeHtmlTree.js. Claude-Session: https://claude.ai/code/session_017TkBWTk3wcyUcJYDp5ZhaP * test: cover the JavaScript tokenizer from its own tests Table-driven comparisons with acorn's tokenizer across ecmaVersions and options (punctuators, numbers, strings, templates, regexps, comments, identifiers, keywords and regexp-vs-division contexts), plus the callbacks, error locations and eof behaviour. Every tokenizer function is reached; what is left is a readNumber guard no input reaches and the context-stack fallback of inGeneratorContext. Claude-Session: https://claude.ai/code/session_017TkBWTk3wcyUcJYDp5ZhaP * test: share the JavaScript parse helpers between its suites The tokenizer and parser suites each carried the same `parse` (the options a build hands the parser) and `plain` (a tree, token or error as comparable data). They now live once in test/helpers/javascriptParse.js. Claude-Session: https://claude.ai/code/session_017TkBWTk3wcyUcJYDp5ZhaP * test: keep the XML printing tests with the XML printer The beautify and minify output, the embedded and async renderers and _xmlCharacterData sat in the HTML parser suite; they move, with their snapshots, to xml/printer. Claude-Session: https://claude.ai/code/session_017TkBWTk3wcyUcJYDp5ZhaP

#22870)

* fix: return the value of every module type from importModule() Build-time execution reads the executed module as JavaScript in its own chunk graph instead of each generator counting importModule() as a JavaScript consumer, so the real bundle gets no extra code. Code generation there also follows codeGenerationDependencies. * test: execute a module after its css referrer in the importModule case * test: keep the inline module type intact on windows * test: check out the importModule case's text fixtures with LF

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>

Navigation Menu

Repository navigation

Comparing changes

Open a pull request

Commits on Oct 23, 2026

This comparison is taking too long to generate.

Uh oh!