model-writtenrun 9776 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Jul 2016 · 3 entries
- Fabien COELHO opened the thread in July 2016 with \into, allowing pgbench scripts to retain single-row query results in explicitly named variables and use database values in subsequent commands #1.
- Pavel Stehule argued for compatibility with psql's \gset, while Tom Lane objected to a following metacommand collecting results retroactively across several queries; Fabien defended compound-query support as important for realistic latency measurements #2 #7 #14.
- Fabien's first substantial rewrite associated variable targets with individual queries, permitted \into inside unfinished SQL, and added escaped-semicolon counting to the shared scanner #16.
Sep - Oct 2016 · 2 entries
- Amit Langote's testing exposed mismatches between empty statements and returned results, prompting scanner changes to count empty statements and suppress comments, along with naming and diagnostic improvements #19 #20 #22.
- Amit considered the patch ready, but an upstream execution-state-machine refactoring immediately required another rebase, which both Fabien and Amit tested successfully #25 #27 #28.
Dec 2016 - Apr 2017 · 8 entries
- After several months, Tom rejected the nonterminating \into design and demonstrated that stripping comments could incorrectly concatenate SQL tokens, requiring a simpler command-ending approach #31.
- Fabien replaced \into with \gset and \gcset, using result-column names and optional prefixes; he removed the problematic comment handling and documented that empty queries inside compound commands remained unsupported #33.
- Rafia Sabih reviewed and exercised the new version, leading to indentation and argument-diagnostic fixes before she marked the revised patch ready #37 #39 #40 #41.
- Tatsuo Ishii found that valid non-ASCII SQL column names could not become pgbench variables, and the resulting compatibility concern deferred the main feature to another development cycle #42 #44 #47.
- Fabien started an independent non-ASCII variable-name patch by adapting psql's validator and extending pgbench's expression scanner, choosing a small compatibility change rather than merging the programs' different variable infrastructures #52 #53.
- Tatsuo's tests found that the new validator still left SQL variable substitution broken, so Fabien extended parseVariable() and subsequently clarified the scanner synchronization comments #55 #56 #58 #59.
- Meanwhile, Fabien renamed \gcset to \cset and removed the main patch's ASCII-only documentation because the independent compatibility patch now addressed that limitation #57.
- Tom committed the non-ASCII feature from #59 as 9d36a38660, adjusting parseVariable() to use the same character test as the validator and relocating the documentation #64.
Aug 2017 - Jan 2018 · 2 entries
- Following a further gap, the command feature was rebased over changes that deferred query-mode processing, preserving its own delayed SQL-command initialization #62.
- Fabien added TAP coverage for successful assignments and error cases, then addressed Pavel's questions about compound-query assembly with clearer comments while maintaining the patch through documentation and source rebases #66 #68 #69.
Mar - Apr 2018 · 1 entry
- Stephen Frost's April 2018 review improved command-field names, ownership explanations, and diagnostics; the one-row requirement remained consistent with psql, but Stephen ran out of time before the PostgreSQL 11 feature freeze #75 #76 #77 #78.
Aug 2018 - Jan 2019 · 10 entries
- The feature continued through another release cycle with rebases, including a resend to resolve MIME-related line-ending problems, and Fabien eventually invited another committer to take it forward #81 #84 #94.
- Alvaro Herrera resumed substantive review in November, asking for separate explanations of \cset and \gset and replacing manual command-string manipulation with PQExpBuffer; Fabien implemented both changes #96 #98 #99 #102.
- Alvaro supplied an additional documentation and cleanup delta, although Fabien's next full attachment only rebased the preceding version; discussion then focused on clearer terminology for individual queries within compound commands #103 #104 #105.
- In January 2019, Alvaro posted a revised full patch with a public scanner accessor and command-structure cleanup, and Fabien corrected its query-count indexing and implemented exponential prefix-array growth #107 #108.
- Fabien expanded the documentation examples and supplied the same changes as a delta so Alvaro could retain his ongoing edits #109 #110 #111.
- Alvaro's final delta added lazy prefix allocation, metacommand enum handling, response-handler cleanup, and command destruction; Fabien integrated it, refined buffer and prefix ownership, and reported additional successful testing #112 #113.
- Alvaro committed the resulting \cset and \gset feature as 6260cc550b from #113, with documentation corrections and a small empty-prefix allocation optimization #114.
- Immediately after landing, Tom challenged \cset's usefulness and its assumption that counting escaped semicolons predicts returned results; Alvaro acknowledged that removing it would simplify the implementation and left time for further arguments #115 #117 #118.
- At this thread's final messages, both features had landed, but \cset's retention remained disputed: Fabien defended the ability to combine queries and capture their results in one network round trip #119 #120.
- The later resolution belongs to thread #40065, where a newer removal patch produced 25ee70511e, partially reverting the command feature by removing \cset and its scanner-counting machinery while retaining \gset.