model-writtenrun 7279 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Aug 2021 - Feb 2022 · 11 entries
- Peter Eisentraut proposed SQL-standard hexadecimal, octal and binary literals in #1, covering both SQL scanners and integer input functions while identifying numeric input and other integer parsers as possible follow-up work.
- John Naylor recommended avoiding repeated radix detection in #2, leading to separate scanner rules with an explicit base in #3; Zhihong Yu's test suggestion in #4 then produced stronger examples in #8.
- Vik Fearing requested underscore separators in #5, and Tom Lane warned in #6 that the proposed literals already parsed as numbers followed by aliases; Peter addressed that ambiguity by rejecting adjacent identifier text in #10.
- Peter split the work into six patches in #11, separating scanner cleanup, integer-conversion refactoring, trailing-junk tests and rejection, and the non-decimal feature.
- John's review in #13 led to a wording correction in the scanner cleanup, which landed as fb7f70112f, while the unused-header component landed unchanged as 89d1c15d64.
- Peter explained in #15 that the draft standard's schedule pushed the feature toward PostgreSQL 16 while preparatory work could enter PostgreSQL 15.
- The updated series in #16 resolved the conversion-function naming conflict through intervening refactoring, added removal of pg_atoi(), and included an underscore prototype; #17 strengthened parameter-junk and ECPG error handling.
- Robert Haas and Tom Lane questioned extending integer input syntax in #21 and #23, while Peter argued in #24 that casts and input routines should remain aligned with literal syntax.
- John reviewed the rebased series in #26, identifying a redundant scanner rule and ECPG comment-placement problem and supplying an underscore-scanner addendum to restore the no-backtracking property.
- The conversion refactoring from #18 landed as cfc7191dfe with an added errno reset, and its pg_atoi() caller replacement and removal landed unchanged as b45fa79340 and 73508475d6.
- The trailing-junk tests from #18 landed unchanged as 13d129333e, followed by rejection as 2549f0661b with John's scanner cleanups; Peter confirmed the preparatory landings and deferred the remaining features in #28.
Oct 2022 - Jan 2023 · 12 entries
- After an eight-month gap, Peter resumed the non-decimal remainder in #29, leaving underscores for later and noting that numeric_in() support and ECPG interactions remained concerns.
- John's request for overflow tests in #33 exposed lost radix prefixes when scanner integers became Float nodes, prompting the parser fix and expanded boundary tests in #35.
- David Rowley proposed faster non-decimal conversion in #37 and supplied the delta in #38, but Peter questioned its readability and inconsistency with decimal parsing in #41.
- Dean Rasheed found that negative non-decimal literals failed in make_const() in #40, and Peter fixed sign handling and added negative boundary tests in #44.
- David benchmarked the optimization in #45, but Dean found incorrect signed accumulation and incomplete overflow checks in #46; their subsequent discussion in #47 and #50 developed unsigned accumulation with separate overflow checks.
- David announced the decimal optimization's move to a separate thread in #51; the complete patch in #521580 in thread #47027 landed with light edits as 6b423ec677, credited to David and Dean, so that commit belongs to the separate thread.
- David also reported that bare radix prefixes were accepted by integer input functions in #48; Peter's #53 added post-loop digit checks and adapted non-decimal conversion to the improved accumulation algorithm.
- Tom noted in #52 that PostgreSQL 15's trailing-junk rejection had already removed the compatibility obstacle, and the final attachment in #53 landed unchanged as 6fcda9aba8, announced in #54.
- Dean supplied separate numeric-type and underscore patches in #55, extending non-decimal values beyond bigint limits and improving ordinary decimal-input performance by avoiding unnecessary special-value comparisons.
- Following Peter's positive response in #56, Dean chose in #57 to land numeric support first and merge underscore handling into the separate underscore thread; 6dfacbf72b landed the first component with additional comments and tests.
- Joel Jacobson demonstrated a faster byte-array-to-numeric conversion and suggested a built-in function in #58; Dean recommended gauging interest in a separate thread in #59.
- Ranier Vilela proposed freeing numeric temporaries before soft-error returns in #60, but Dean explained in #61 that memory-context cleanup was intentional and broader immediate cleanup would complicate input code; Ranier accepted the explanation in #62, leaving the integer and numeric features landed, underscores continuing elsewhere, and this cleanup uncommitted.