model-writtenrun 4787 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Peter Eisentraut proposed SQL-standard underscore separators for numeric literals, implementing lexer and integer input support while leaving numeric and float input support unfinished #1.
- Tom Lane and Justin Pryzby opposed extending datatype input functions because of performance and input-validation concerns, while Andrew Dunstan argued that accepting unquoted values but rejecting quoted equivalents would confuse users #2 #3 #4.
- Dean Rasheed favored consistent literal and datatype input syntax, including COPY input, and supplied revised integer parsing loops that avoid underscore checks for ordinary digits; he also proposed centralizing parsing through soft-error conversion routines and identified a redundant lexer quantifier #5 #6.
- Dean implemented that rework, replacing lexer-side underscore stripping with shared integer conversion routines and adding underscore handling and tests for numeric input #7.
- Peter approved the updated patch, and he and Dean agreed to leave float input unchanged while it relies on strtof() and strtod(), avoiding additional performance costs #8 #9 #10.
- Dean committed #7 as faff8f8e47 with documentation cleanup and a named 32-bit exponent bound, then confirmed completion; integer and numeric support landed, with float input deliberately left unchanged #11.