model-writtenrun 8169 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
May - Sep 2020 ยท 17 entries
- Peter Eisentraut opened the thread by proposing documentation for the existing factorial() function as a prerequisite to deprecating the postfix factorial operator and eventually removing postfix-operator support #1.
- Robert Haas proposed retaining a specially recognized postfix ! operator, but Tom Lane and Vik Fearing opposed the compatibility restrictions that would impose on other uses of !; the discussion ultimately favored removing postfix operators altogether #7 #11 #12 #14.
- Stephen Frost advocated tying deprecation to an implemented removal, while Tom insisted on demonstrating a concrete user benefit first: removing postfix ambiguity alone still left conflicts involving type names and aggregate syntax #15 #16 #20.
- Tom posted a grammar prototype that removed postfix productions and split troublesome unreserved keywords into a new category, allowing the remaining unreserved keywords as column labels without AS #21.
- Tom, Alvaro Herrera, and Robert developed the alternative of making bare-label eligibility independent of existing keyword categories, with a grammar list and consistency checks, while Peter demonstrated that removing postfix operators also eliminated some precedence hacks #23 #26 #27 #28 #29.
- After a six-week gap, Mark Dilger expanded Tom's prototype into a working series with infrastructure and tests, then adopted the orthogonal keyword design so that 411 of 450 keywords could serve as labels without AS #31 #32.
- John Naylor's review led Mark to remove the redundant keyword category, split postfix removal from keyword-label support, and expose label eligibility through pg_get_keywords() #33 #34.
- John subsequently found a pg_get_keywords() crash, missing backend cleanup, and the need for a pre-upgrade check; Mark's next version fixed those issues, merged the keyword-related patches, and demonstrated detection of user-defined postfix operators during an upgrade #35 #38.
- John supplied a separate precedence-cleanup delta, which Mark incorporated and extended while separating the keyword row-estimate adjustment; John then recommended deferring optional precedence refactoring until after the functional changes #39 #40 #49.
- John opened a documentation-only deprecation branch, and Robert extended it to cover both factorial operators after Tom advocated standardizing on factorial(); Robert then noted that the deprecation work had moved to another thread #43 #45 #50 #51.
- Mark's next series removed both factorial operators and numeric_fac(), incorporated documentation changes against the separately committed deprecation notices, and deferred precedence cleanup; John's review then produced a final revision that restored postfix dumping with a warning and removed obsolete documentation and redundant tests #52 #53 #54.
- Robert questioned whether pg_dump should reject postfix operators and requested a separate commit for removing the built-in operators; Tom favored warning while preserving dumps usable on old servers, and the discussion settled on the existing warning behavior without adding generalized compatibility options #56 #59 #61 #62 #63.
- Tom extracted the factorial-removal portion of Mark's final series into 76f412ab31, preserving factorial() and making only test and catalog-version adjustments #54 #64.
- Tom committed the remaining postfix-support removal as 1ed6b89563, retaining old-server support in pg_dump and psql, refining parser and deparser code, and standardizing documentation terminology #54 #65.
- For the keyword-label change, Tom replaced implicit-alias terminology with bare-column-label terminology and expanded pg_get_keywords() to provide both boolean and descriptive outputs; 06a7c3154f landed that revised portion of the final series, including a row estimate of 500 instead of 450 #54 #65 #68 #71.
- Finally, Tom revisited John's deferred cleanup and committed 28a61fc6c5, removing precedence assignments for GENERATED, NULL_P, PRESERVE, and STRIP_P while retaining IDENT's existing position because moving it could affect window-name parsing #39 #73.
- By the final message, the functional removal, expanded keyword-label support, and conservative precedence cleanup had landed; the broader precedence rearrangement remained deliberately unadopted, with no further patch submission pending in this thread #73.