model-writtenrun 7893 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Brar Piening reported that TIMELINE_HISTORY advertised bytea while returning unescaped file bytes, which broke his attempted bytea decoding for Npgsql; he also identified the row descriptor as potentially incorrect in #1 and #2.
- Bruce Momjian proposed labeling the content as text while documenting that no encoding conversion occurs in #3, but Michael Paquier pointed out that user-supplied recovery target names can introduce non-ASCII characters in #4.
- Bruce retained the text-type change and proposed backpatching it in #6; Tom Lane opposed changing stable-branch behavior in #7, while Brar explained the client-decoding risk of an unqualified text label in #9.
- Tom rejected bytea escaping because it would break existing clients, recommended documenting raw file contents, and insisted that the encoding caveat apply only to the relevant fields rather than all text results in #10, #13, and #16.
- Bruce supplied separate branch variants in #17: development would label content as text and explain the raw bytes, while stable branches would preserve their existing type labels and receive documentation plus a source comment.
- Bruce committed the development variant as 66a8f09048 and the stable-branch clarification as fea2b70dd4, each matching its respective attachment in #17, and announced completion in #18.
- Jeff Davis reopened the descriptor question in #19 and supplied a patch using binary-format bytea for both filename and content in #21; that message also relayed Andres Freund's objection to unsolicited binary results and his alternative of enforcing ASCII.
- Jeff switched to enforcing ASCII restore-point names in #23 and #24; Andres requested a shared string helper in #25, and Michael identified existing duplicate implementations in #27.
- Jeff withdrew the ASCII patch in #29 after finding that users may add arbitrary comments to history files, so restricting restore-point names would only partially address the encoding problem; he also declined to pursue binary descriptors further against the objections.
- The two November commits remained the thread's landed outcome, with both later approaches abandoned; Michael recalled Japanese text in history files in #30, and Kyotaro Horiguchi ended the discussion in #31 with a related unresolved observation about backup-label encoding.