model-writtenrun 8779 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Tom Lane proposed making suitable contrib modules trusted so non-superusers could install them, while excluding modules with external access or uncertain security implications #1.
- Alvaro Herrera supported btree_gist and documenting trust in each module's section, while Darafei Praliaskouski advocated moving btree_gin and btree_gist into core; Tom rejected the core move as outside scope and troublesome for upgrades #2 #3 #4.
- Julien Rouhaud questioned excluding modules whose additional privileges were already controlled separately, but Tom favored conservative defaults; Dean Rasheed raised similar concerns about pgrowlocks, which Tom agreed to leave out #5 #6 #7 #8.
- Tom submitted a patch marking 23 modules trusted and documenting them individually, including btree_gin and btree_gist but excluding intagg because its compatibility-wrapper role warranted eventual deprecation #9.
- Stephen Frost argued that existing privilege checks should permit broader trust, particularly for pg_stat_statements, but Tom maintained that further security analysis was needed before expanding the initial selection #10 #11.
- The trusted-module patch #9 landed unchanged on February 13 as eb67623c96.
- Later that day, Andres Freund warned that unpackaged upgrade scripts could operate as superuser on attacker-created objects; Tom considered requiring superuser privileges for FROM but favored removing the aging mechanism, and Andres supported removal #12 #13 #14 #15.
- Tom posted a separate patch removing core support and documentation for CREATE EXTENSION ... FROM, explicitly omitting the bulky script deletions, and Stephen subsequently expressed general agreement with removal #16 #17.
- The removal patch #16 landed on February 19 as 70a7732007, retaining its core changes and adding deletion of unpackaged scripts, build-list cleanup and related comment updates.
- After both commits landed, strk reported that removing FROM unpackaged broke PostGIS automated tests and packaging workflows, and said retaining the option with a superuser requirement would have worked #18.
- In response to Andres's request for details, strk explained PostGIS's legacy-installation and raster-repackaging needs and asked whether packaging during ordinary CREATE EXTENSION would recreate the security risk; the thread ends with that compatibility question unresolved and no replacement patch or revert shown #19 #20.