model-writtenrun 3304 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Michael Banck reported that a prepared statement over a heavily partitioned schema failed while serializing a parallel plan with a misleading out-of-memory error; disabling parallel workers avoided the failure, and his PostgreSQL 17 beta test did not reproduce it #1.
- David Rowley proposed improving enlargeStringInfo's diagnostic to identify the maximum string-buffer length, while noting similar misleading errors in encoding-conversion routines and pvsnprintf that were outside the attached patch #2.
- Daniel Gustafsson suggested more readable units, but David explained that rounding would obscure why the requested growth exceeded the limit; Daniel and Tom Lane agreed to retain exact counts and explicitly label them as bytes #3 #4 #5 #6 #7 #8.
- David posted the revised patch with "(%zu bytes)" in both backend and frontend diagnostics #9.
- Tom recommended waiting for PostgreSQL 18 because this diagnostic improvement was not a bug fix and would otherwise add translation work late in the PostgreSQL 17 cycle; Daniel agreed #10 #11.
- After branching, David committed the revised patch unchanged as 1029bdec2d and announced completion #12; the change clarifies the error but does not remove the buffer-size limitation behind the original report #1.