[kvlonge]

Adds three file inputs to kind = "files" sources. SQLLoad reads all of them itself, on every destination.

• Excel: format = "xlsx" with xlsx = { sheet, header, skip_rows }. Cells keep their stored types: int, float, bool, date, timestamp_ntz and time. Empty cells and error cells load as NULL. Cells are streamed with calamine's cell reader rather than a dense range, and a sheet is capped at 20 million values.
• XML: format = "xml" with xml = { record_path, lists }.
  • Each element at record_path becomes a row. Its attributes and child elements become columns, matched by local name.
  • A repeated or listed child loads as a JSON array, and a nested element as a JSON object.
  • xsi:nil loads as NULL. Namespace declarations and xsi: attributes are not columns.
  • These fail with an error: a truncated document, a UTF-16 document, duplicate attribute names, and a record_path whose parent elements the document doesn't have.
• ZIP: compression = "zip", with a members glob. Each matching member loads as a file of its own, keyed <archive>/<member> and versioned by its archive.
  • Only the archive's central directory and the members being read are fetched, using range requests. The archive is never downloaded whole.
  • Members are limited to 1 GiB each. An archive with duplicate member names fails.
  • CSV, xlsx and XML members are all supported, through one source per format over the same path.

Dependencies: this adds calamine 0.38, quick-xml 0.42 and zip 9 (with deflate). calamine turns on zip's deflate, which switches flate2 to the zlib-rs backend across the workspace, so gzip output bytes differ from 0.34. No test or feature depends on byte-identical gzip output; tests decompress before comparing. CsvOut::Gzip is boxed to satisfy clippy's large_enum_variant.

Docs: a new "Excel, XML and ZIP files" section in sources/files.mdx, marked "From 0.35.0 (not released yet)".

Testing:

• make check and make test-unit pass.
• The new test_read_containers cases cover typed xlsx cells, sheets and skip_rows, header clashes, XML shapes, namespaces and errors, ZIP members and globs, xlsx and XML inside a ZIP, and an archive larger than one 4 MiB range block.
• A CLI end-to-end run on synthetic data loaded one ZIP holding CSV, xlsx and XML through three sources into DuckDB, and an append rerun loaded 0 rows.

Closes CHI-1238.

[cloudflare-workers-and-pages]

Deploying sqlload with    Cloudflare Pages

Latest commit:	da3ba1a
Status:	✅  Deploy successful!
Preview URL:	https://2fe47224.sqlload.pages.dev
Branch Preview URL:	https://feat-xlsx-xml-zip.sqlload.pages.dev

View logs