model-writtenrun 175 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Zsolt Parragi proposed extensible HBA options because global validator GUCs could not express different OAuth configurations per authentication rule, later identifying multitenancy, identity-claim selection, and per-database client restrictions as concrete uses #1 #4 #8.
- Jacob Champion favored a general solution, suggesting a PGC_HBA GUC context or eventual configuration unification, while VASUKI M advocated a smaller OAuth-specific option-registration interface #2 #3.
- Jacob explained that existing inline includes and backslash continuations already addressed long HBA lines, resolving that part of Zsolt's original request #6 #8.
- Zsolt's first patch implemented a generic HBA parsing hook requiring shared preload, but Jacob objected to requiring cluster restarts for validator security updates #10 #11.
- After discussion of library-loading order, Zsolt replaced the hook with per-HBA GUC settings, applying the matched rule's settings during authentication and rejecting unresolved variables after session preload #14 #15 #16.
- Jacob rejected unloading libraries inside the postmaster for configuration checks, requested support for existing per-session GUC contexts, and argued that database and role settings should override HBA defaults #19 #21.
- Zsolt explored validation in child processes, but Jacob recommended separating that work and accepting connection-time validation initially; namespace collisions and interaction with SNI configuration made the broader GUC approach risky for PostgreSQL 19 #22 #23.
- Zsolt expanded the GUC series with prefix ownership enforcement, broader supported contexts, and revised precedence, while reporting that a separate validator-variable API prototype required too much boilerplate #28.
- Alvaro Herrera and Nikolay Shaplov suggested reusing the relation-options framework, but Zsolt concluded that a general solution needed repeated validation across loading phases and would not transition cleanly into PGC_HBA #29 #31 #33.
- After the discussion went quiet, Jacob supplied an OAuth-only string registration and lookup API intended as a small interim solution for PostgreSQL 19, deferring the broader GUC design and rejecting extension hooks during HBA parsing; he also included a separate failure-DETAIL prerequisite #34.
- Zsolt requested validator-specific prefixes and earlier authentication-method checking, but accepted the fixed validator.* namespace after Jacob explained its collision avoidance and deliberate distinction from GUC semantics #35 #37 #39.
- Jacob completed the user documentation and incorporated the method-check ordering fix, then resent only the custom-options patch after dropping the prerequisite, with no further changes to that patch #41 #42.
- Zsolt approved the final patch, and Jacob retained repeated getter calls in tests to exercise their supported behavior; b977bd308a then landed the implementation from #42 unchanged #43 #44.
- In October, Zsolt reported that pg_hba_file_rules omitted the newly supported validator options and submitted a separate fix that preserves their input order and dynamically sizes the options array #45.
- At the thread's end, the custom-options feature had landed but the view-display fix remained under discussion: Daniel Gustafsson questioned unbounded allocation and suggested a cap, while Zsolt proposed ArrayBuildState instead, leaving the implementation choice unresolved #46 #47.