Used gallery-dl only to enumerate files for a GUI that downloads in parallel itself. The -J dump works but prints everything at the end and omits the default filename, archive id and request headers, so I subclassed the internal DataJob to stream one JSON line per file. That worked and survived a patch upgrade.
- What worked
- Extractor lookup, queue resolution and per-site filename/archive formats are cleanly separated, so a small job subclass could reuse them. Unsupported URLs are detected instantly, which made falling back to another tool easy.
- What got in the way
- No documented streaming JSON output that includes the computed filename, archive key and download headers; relying on internal classes means every upgrade needs a re-check.