model-writtenrun 7179 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- John Naylor announced the adaptive radix tree template, authored with Masahiko Sawada, using adaptive node capacities, 64-bit keys, combined pointer/value storage, and macro specialization for value types and shared or local memory #1.
- The initial implementation omitted general path compression because of its complexity and lazy expansion because it requires storing full keys; future work would use the infrastructure for TID storage #1.
- Tom Lane raised patent concerns, John initially cited existing open-source implementations, and Alvaro Herrera suggested contacting the paper's authors; John then reported doing so #2 #3 #4 #5 #6 #7.
- Andrew Dunstan reported failures on the drongo and fairywren buildfarm members #8.
- John reported that the paper's authors confirmed they had not patented adaptive radix tree, then turned to the Windows failures #9.
- David Rowley supplied a PGDLLIMPORT fix that restored his MSVC build, John preferred it to his impending workaround, and David separately supplied a typo correction for inclusion alongside it #10 #11 #12.
- John committed both attachments unchanged together, fixing the Windows linkage declarations and the test-module comment, with diagnosis credited to Masahiko Sawada ab6ae62603.
- The thread ended with John confirming completion and explaining that he had accidentally committed his workaround too, but had reverted that first attempt #13.