model-writtenrun 1234 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Jan - Feb 2024 · 3 entries
- David G. Johnston opened the thread by proposing that COPY ON_ERROR ignore explicitly identify row skipping, anticipating a separate null-substitution behavior; Yugo Nagata preferred retaining ignore and naming the new action set_to_null #1 #4.
- jian he submitted an initial ON_ERROR 'null' implementation that rejected tables containing NOT NULL columns, but Atsushi Torikoshi and Yugo argued that this would unnecessarily exclude ordinary tables with primary keys #5 #6 #12.
- The next patch adopted set_to_null, allowed ordinary constraint enforcement during insertion, and reported affected rows; Jim Jones tested it, and David explained why even entirely null input records should remain eligible for insertion #13 #14 #15.
Aug 2024 - Jan 2025 · 8 entries
- After a six-month gap, jian simplified substitution to operate directly within the field-conversion loop, dropping earlier row-level bookkeeping; a subsequent refresh added option-validation tests, and Jim reported successful testing #17 #19 #20.
- Kirill Reshke took over updating the patch, rebasing it and incorporating Jim's wording corrections; Fujii Masao then requested notices, proper completion, clearer documentation, and a defined interaction with REJECT_LIMIT #21 #22.
- Kirill added documentation and completion changes, identified that substituted nulls could bypass domain constraints, and introduced a second input-function call to validate them, with failed substitutions handled through REJECT_LIMIT #23 #26.
- Atsushi Torikoshi and Yugo debated whether REJECT_LIMIT should count all malformed rows or only failed substitutions, and Atsushi also identified that completion incorrectly offered SET_TO_NULL as a standalone COPY option #27 #29 #33.
- Two competing revisions followed: jian made domain null-rejection fatal to align with column constraints, while Kirill independently retained the REJECT_LIMIT approach and added affected-row counting and notices #34 #35.
- The authors established that ordinary column constraints and domain input constraints already followed different error paths, and Kirill noted that making general constraint failures skippable would require broader executor work #36 #37.
- jian resolved the immediate complexity by removing REJECT_LIMIT support for set_to_null and making rejected null substitutions fatal; Kirill endorsed the reduced scope as more likely to be committed #39 #40.
- An update briefly counted substituted rows in tuples_skipped, but the authors recognized that those rows were retained, so jian removed that progress reporting and deferred a separate metric #41 #42 #43.
Mar - Apr 2025 · 5 entries
- After another rebase, Jim tested CSV input, verbose reporting, and existing IGNORE behavior; NOTICE severity was retained for consistency, while his terminology feedback changed 'erroneous' to 'invalid' #45 #46 #47 #48 #49.
- Vignesh identified the completion error and requested documentation cleanup; jian fixed completion and also supplied an alternative ON_ERROR NULL grammar patch, but continued with set_to_null after Vignesh accepted that approach #50 #51 #52.
- Vignesh's testing exposed misleading column information in constraint-error context, prompting jian to clear stale context and align verbose notices with the existing IGNORE behavior #52 #53.
- Masahiko Sawada proposed the shorter set_null name, matching enum naming, richer notices, and improvements to domain validation; jian adopted the naming and reporting changes #54 #55.
- Further review from Masahiko led to cached constrained-domain flags, revised error handling, and benchmarks showing the benefit of avoiding unnecessary repeated domain input calls #56 #57.
Jul 2025 · 1 entry
- Atsushi reviewed the revised implementation and suggested keeping affected-row counting within the SET_NULL branch; jian reorganized conversion handling around shared verbose reporting and explicit loop control, then maintained the patch through later rebases #58 #59 #60 #61.
Jan - Mar 2026 · 7 entries
- Matheus Alcantara resumed detailed review after a fresh rebase, checking FORCE_NOT_NULL interactions, allocation helpers, notices, and progress documentation; the discussion concluded that FORCE_NOT_NULL processes input markers before ON_ERROR handles conversion failures #63 #64 #65 #66 #67.
- Matheus supplied a small follow-up diff for comments, diagnostics, and monitoring documentation; jian incorporated the diagnostic ideas while retaining IGNORE-only tuples_skipped reporting and explaining why a proposed null-input notice branch was unreachable #69 #70.
- Matheus tested and approved the updated patch, after which Peter Eisentraut requested final wording and comment improvements and clarification of domain CHECK constraints that reject null values #71 #72.
- jian supplied the final revision and explained that substituted nulls must satisfy all domain constraints; Peter committed it with light documentation, comment, message-grammar, and formatting edits as 2a525cc97e, crediting jian and Kirill as authors #73 #74.
- Fujii then noticed that the new option also worked through file_fdw but lacked regression coverage there, and submitted a separate test checking that malformed fields become null #75.
- Yi Ding endorsed the test and suggested explicit IS NULL checks; Fujii explained that the suite's configured null display already verified the result, and Yi accepted that explanation #77 #78 #79.
- Fujii committed the file_fdw test unchanged as d8879d34b9 and announced completion, leaving both submitted changes landed while REJECT_LIMIT support and dedicated progress reporting for SET_NULL remained deferred #75 #80 #65 #43.