model-writtenrun 9836 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Ertuğrul Kahveci reported a DROP TABLE failure with “vsnprintf failed: Invalid argument” on PostgreSQL 11.1 for Windows, and Andres Freund and Tom Lane requested a retry with untranslated messages to isolate the problem #1 #2 #3.
- Tom Lane traced the failure to an extra space in a Turkish translation's format specification: POSIX allowed the space flag, but PostgreSQL's snprintf implementation rejected it; he proposed correcting the translation, displaying failing format strings, and finding a way to validate translations against PostgreSQL's supported syntax #4.
- Stephen Frost suggested testing translations or strengthening validation, and Tom Lane described his experimental validator that suppressed argument fetching and output, proposing stricter rejection of unknown format codes as groundwork for translation checks #5 #6.
- Alvaro Herrera fixed the reported Turkish translation for PostgreSQL 11.2, while Tom Lane announced the diagnostic and format-validation improvements #7 #8; these landed directly from the discussion as d2b0b60e71, with the pvsnprintf diagnostic improvement backpatched and the snprintf parsing changes confined to development.
- Alvaro Herrera recommended integrating translation checks into the regularly run pg-make-po script; Tom Lane favored reusing C parsing code to avoid maintaining a separate Perl implementation and explained why ordinary snprintf calls could crash without correctly supplied arguments #9 #10 #11 #12.
- Tom Lane posted a validator patch after finding many translations with extra spaces, but reconsidered committing it because PostgreSQL's move to its own snprintf everywhere could expose compatibility problems in third-party code; he instead floated implementing the space flag and accepting two other flags as no-ops #13.
- Peter Eisentraut closed the thread by reporting corrections to all erroneous space flags in translations for upcoming minor releases and explaining that gettext checks argument types rather than formatting flags; the validator remained uncommitted in the supplied record, and reliable prevention of similar mistakes remained unresolved #14.