Incremental Update: Amit Kapila Responds, Pushes Back on Documentation Need
What's New
Amit Kapila has responded to the original thread with a counterargument that significantly reframes the issue. Rather than engaging with the documentation patch proposal directly, he argues the behavior is already well-understood by analogy with the explicit INCLUDE case, and therefore additional documentation for the EXCEPT case may be unnecessary.
Kapila's Technical Argument
Kapila demonstrates with a concrete example that the same OID-based behavior applies to explicit table inclusion — not just exclusion. When you CREATE PUBLICATION pub_inc FOR TABLE inc_t, then drop and recreate inc_t, the table silently falls out of the publication. The pg_publication_tables view returns zero rows.
His point: this is a symmetric behavior across both INCLUDE and EXCEPT publication types. Since the INCLUDE case has never been documented with an explicit OID-based caveat, singling out the EXCEPT case for special documentation treatment would be inconsistent. The OID-based tracking is a fundamental PostgreSQL-wide pattern, not something specific to the EXCEPT clause.
Shift in Framing
The original report framed the problem as an asymmetry between dynamic FOR ALL TABLES (auto-includes new tables) and static EXCEPT (doesn't auto-exclude recreated tables). Kapila reframes it as a consistent behavior — all catalog-based object references use OIDs, and drop/recreate breaks those references uniformly, whether the reference is "include this table" or "exclude this table."
Kapila's Disposition
He does not outright reject a docs patch — his closing remark ("feel free to propose a docs patch") leaves the door open — but his tone clearly signals he doesn't see this as a gap worth committer attention unless the reporter can demonstrate real user confusion. The implicit bar is: produce an actual patch and make the case that users are being surprised by this in practice.
Impact on Thread Trajectory
This response significantly lowers the likelihood of a documentation change being committed. Without a concrete patch and additional community support, the thread is likely to stall. The original reporter would need to either (a) submit a patch that documents the OID-based behavior for publications generally (covering both INCLUDE and EXCEPT), or (b) provide evidence of real-world user confusion to justify the change.