model-writtenrun 8441 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Apr - May 2020 ยท 19 entries
- Tom Lane opened the poll after publishing an initial function/operator-table redesign, seeking a layout that accommodated longer explanations and multiple examples without becoming unreadable in PDF or narrow windows #1.
- Robert Haas found the initial layout harder to read, Andrew Dunstan identified redundant function names, and David G. Johnston proposed combining examples with their results instead of allocating separate result columns #2 #9 #11.
- Tom's first attached prototype combined signatures, descriptions and examples into table cells with explicit line breaks; horizontal rules added excessive spacing and lacked a satisfactory PDF implementation #19.
- Jonathan S. Katz proposed removing the separate function-name column and letting signatures span the available width, while Robert advocated considering definition lists and questioned unnecessary operator descriptions #22 #23 #33.
- Tom compared two layouts on the string and enum tables, preferred the spanning-signature alternative, and retained tables because their boundaries distinguished per-function details from surrounding commentary #45.
- Pierre Giraud demonstrated a single-row layout with an indented description/example block, leading Tom to implement role-based hanging indents in HTML and PDF instead of spanning table rows and columns #47 #49 #53.
- Robert and Tom settled on omitting a separate operator-name column while retaining explanatory descriptions; Tom also rejected David's operator-first notation because signatures should show actual invocation syntax #56 #57 #58 #63.
- The hanging-indent implementation from #53 landed as 737d69ffc3, with additional date/time-function conversion and editorial improvements beyond the posted patch.
- Operator discoverability remained controversial after that landing: Tom tried a separate name column, Victor Yegorov reported difficulty finding operators, and Alvaro Herrera favored reconsidering the name-column alternative #67 #75 #78.
- Peter Eisentraut challenged the implementation's semantic markup and maintainability, but Tom declined to abandon the readability improvements and offered to replace the markup while preserving the agreed appearance #71 #81 #82.
- Alvaro identified entries that needed multiple signatures, and Tom proposed separate paragraphs with signature roles so each signature could receive its own indentation #85 #86.
- Tom posted a paragraph-based prototype that preserved the visual layout, exposed a website paragraph-margin override, and revealed unwanted PDF whitespace around index terms #87.
- Jonathan supplied a targeted CSS override removal that Tom verified, while Tom concluded that JSON documentation made multiple signatures essential and separately investigated index-term whitespace in FO output #89 #90 #94 #95.
- Set-returning JSON functions required multiline programlisting results, so Tom expanded the website rules and supplied a separate JSON conversion draft for testing; Jonathan then revised and tested the CSS against that draft and older documentation #96 #98 #99 #101.
- Tom found that the condensed CSS failed in Safari, Jonathan restored explicit margin and padding declarations, and the resulting website styles were tested and pushed before the PostgreSQL markup conversion #102 #103 #104 #105.
- The prototype from #87 became 4ad047a6ea, extending paragraph-based markup across previously converted tables, supporting multiple signatures and program listings, and removing obsolete formatting machinery.
- Tom subsequently reported that Chapter 9's conversion was complete, with contrib documentation still needing attention, and proposed replacing returnvalue markup with a custom arrow entity because DocBook restricted multiline results and embedded type markup #106.
- Jonathan offered further website spacing and list-formatting improvements, while Dean Rasheed preferred explicit function overloads and Tom explained why enumerating every numeric operator variant would be impractical #107 #108 #109 #110 #112.
- The thread ended with Tom's uncommitted comparison-table spacing experiment responding to Oleksandr Shulgin's report; Tom found that it only partly improved alignment and considered more elaborate markup unjustified, leaving this experiment and the remaining markup questions unresolved #111 #113.