model-writtenrun 1668 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Michael Paquier proposed SQL regression tests for injection-point name, library and function length limits, using the newly available attachment function to exercise those checks #1.
- Chao Li pointed out that rejecting a 64-character name while reporting a maximum of 64 was confusing, and Michael agreed that the reported limit should exclude the terminating zero byte #2 #3.
- Xuneng Zhou supplied a two-patch revision that corrected the error messages and updated Michael's expected test results; it also proposed reporting a private-data maximum of 1023 bytes #4.
- Daniel Gustafsson considered the tests unlikely to catch many bugs but accepted their low cost, and Michael subsequently reported that the revision worked for him #5 #6.
- The backend portion of #4 landed as 6e1535308c, with Michael additionally allowing exactly 1024 bytes of private data because that area needs no terminating zero byte.
- The test portion of #4 landed separately as fb9bff0454, adding private-data overflow coverage and making argument values consistent in both new tests and existing NULL-input tests.
- Michael announced the private-data correction and backpatching through PostgreSQL 17 to keep injection-point behavior consistent across supported branches containing the facility #7.
- Xuneng and Chao expressed agreement after both commits had landed; the thread ended with both components committed and no outstanding work identified #8 #9.