Release v0.8.19 · karust/openserp

Changes

Chrome recovery, steadier Google results, structure-preserving extraction, and a container that reaps its own processes.

Browser recovery: if Chrome dies (killed, crashed, or replaced by a package upgrade), the next request relaunches it instead of failing until restart (#70). Concurrent failures share one reconnect, canceled requests stop waiting, and the temp profile is removed before exit.
Google: waits for result titles, not just their wrappers, which fixes false timeouts on slow pages. A no-results card is confirmed by page text, so a notice above results no longer returns an empty list.
Extraction: preserve_structure=true on /extract and /extract/batch converts the page's content container as-is instead of running article detection. Headings, code blocks and API signatures stay with their descriptions on <main> pages, DocBook, Texinfo, Apache httpd docs and man pages. Falls back to article detection when the page has no such structure; ignored with clean=false (#76).
Docker: the image runs under tini, so orphaned Chrome processes are reaped even without docker run --init. --init and compose init: true keep working.
Startup: serve exits non-zero on invalid proxy config, all engines disabled, or a listen error. Proxies are validated before Chrome starts.
API schema: meta.partial and meta.total_results are documented as optional fields.

Process supervisors and scripts now see a non-zero exit code for startup errors that used to log and exit 0.
--init in docker run is no longer required.
Google no longer logs the "Found N total results" line.

Thanks to @krustfurt for #70 and @ramgalin for #76.

Full Changelog: v0.8.17...v0.8.19