model-writtenrun 4195 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Daniel Verite proposed EBCDIC ordering as a practical ICU tailoring example for mainframe migrations, explaining that database-level rules could preserve existing queries while asking how to present a long rule in the documentation #1.
- Joe Conway and Jonathan S. Katz supported documenting the real-world example, while Jeff Davis requested explanatory material and questioned the importance of database-level support given his technical concerns about that feature #2 #3 #4.
- Daniel explained that removing database-level support would require explicit COLLATE clauses throughout migration tests, then submitted documentation using a PL/pgSQL generator so users could adapt character ordering for different EBCDIC variants, with cross-links from CREATE COLLATION and CREATE DATABASE #5.
- Peter Eisentraut demonstrated that ICU rules accept whitespace and argued that the generator introduced another mechanism without explaining the rules language; Jeff Davis endorsed the clearer direct example #6 #7 #8.
- Daniel replaced the generator with a direct CREATE COLLATION example, compressed consecutive characters into ICU ranges and corrected the caret position relative to Peter's inline example #9.
- Peter committed the revised patch as 17ec2c5dfa with light editorial changes, moving the existing simple tailoring explanation into the new section before the more complex EBCDIC example #10.
- After the commit, Jehan-Guillaume de Rorthais shared a customer-provided libc locale for IBM500 EBCDIC ordering and asked whether custom libc locale creation also deserved documentation; that separate proposal remained unanswered when Daniel closed the thread by thanking Peter for the committed ICU documentation #11 #12.