feat(xml): read an xml file as its source, not as one very long line

[andiwand]

🤖 Generated with Claude Code

An xml file opened as text_file and rendered through the numbered line list. For the files anyone opens on purpose — content.xml, document.xml, anything a writer emitted rather than a human — that is one line several megabytes wide.

It now opens as xml::XmlFile and renders as a source view: reindented, highlighted, and foldable through <details>/<summary> with no script and no external resources.

Stages 1 and 2 of src/odr/internal/xml/PLAN.md, which is updated to say what landed and what did not.

What is in here

• internal/xml/xml_file.{hpp,cpp} — XmlFile, an abstract::TextFile over a text::TextFile mirroring JsonFile, plus parse_source holding the parse flags (parse_full | parse_ws_pcdata_single) both callers need.
• internal/html/xml_file.{hpp,cpp} — create_xml_service and the writer. xml_css and write_xml_style in frontend.cpp.
• util::xml::read_declared_encoding — the declaration sniff, in the shared xml helper rather than a second one with the same name.
• open_file_as gains an xml branch; open_file's unknown-type path tries it after svg and before the text fallthrough, so anything with a more specific reading keeps it. The table row flips to {.open = true, .translate_html = true}.
• html::translate(const DecodedFile &) sends FileType::xml to the new service, next to the csv branch and for the same reason. Not the TextFile overload: a caller who asks to translate a text file is asking for the text rendering, and still gets it.

Decisions worth a second look

• Encoding comes from the declaration. pugixml resolves UTF-8/16/32 from a BOM or the <?xml byte pattern and nothing else, so encoding="ISO-8859-1" would be read as UTF-8 and yield invalid UTF-8 in the node strings, silently. The declaration is read off the head of the file, mapped through text_encoding_by_name, and the bytes are transcoded before pugixml sees them. An encoding we can name but not decode throws UnsupportedTextEncoding — there is no "let the browser sort it out" once bytes are inside a parser — and the file falls back to the line list.
• Mixed content is not reindented. An element holding any text child keeps all of its children on one line, untouched. The plan said "any non-whitespace text child"; with parse_ws_pcdata_single the two differ only for <a> </a>, and the looser rule is the one that leaves it alone.
• No line numbers. They would number our lines, not the file's, which for a reindented minified document is actively misleading. The gutter carries the fold handles instead.
• A parse failure falls back to text, so a malformed file renders exactly as it does today.
• No DTD processing, which is pugixml's behaviour and closes XXE and entity expansion by construction.

Behaviour changes to be aware of

• A .xml that answered text_file / text/plain now answers xml / application/xml. No binding work — the enumerator already existed — but a caller switching on file_type() sees it.
• A flat-xml ODF, .xhtml, .rels, .plist and rss feeds become source views rather than line lists. Correct for a source viewer; not to be mistaken for flat-ODF support.

Not in here

Stage 3 (a node budget for the tens-of-megabytes case, which wants a measurement first) and stage 4 (routing archive entries through translate). The test-data samples stage 1 asked for are still owed — the tests are inline string literals in test/src/internal/xml/xml_file_test.cpp.

Test plan

• ./test/odr_test --gtest_filter='Xml*' — 13 new tests: detection, svg still winning over xml, malformed falling back to text, the declared encoding decoding, the declaration sniff itself, translating it as a text file still writing the line list, reindentation, mixed content, whitespace-only content, declaration/doctype/PI/comment/CDATA all surviving, folding, and attribute quoting.
• Full suite: 848 passed, 0 failed, 8 skipped (the same 8 as on main).

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 0dec71d5f3

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".