model-writtenrun 3828 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Bowen Shi reported that large scram_iterations values could leave backend password computation unresponsive to cancellation and proposed a backend-only CHECK_FOR_INTERRUPTS() in the scram_SaltedPassword() loop #1.
- Aleksander Alekseev suggested considering a lower iteration limit and requested a comment #2; Daniel Gustafsson opposed an arbitrary limit because users may deliberately harden passwords #3, while Tom Lane raised denial-of-service concerns and noted that frontend computation would remain uninterruptible #4.
- Bowen retained the interrupt-check approach and added the requested comment #5; Aleksander supported merging it because choosing a lower iteration limit could break existing users #6.
- Michael Paquier supported making the backend loop interruptible and proposed backpatching to PostgreSQL 16, where scram_iterations became user-settable #7, and Bowen agreed #8.
- Daniel defended configurable iteration counts, reported timings for default and high iteration counts, and supported stopping backpatches at PostgreSQL 16 because older releases used a low hardcoded value #9.
- Michael committed Bowen's revised patch with light include-placement, comment, and whitespace edits as 14f2f9eb1a, backpatched it to PostgreSQL 16, and confirmed completion while leaving frontend cancellation for \password as a possible future callback-based improvement #10.