[billdenney]

Summary

R_ut_format() formatted units into a 256-byte buffer and warned only when ut_format() returned exactly sizeof(buf). But ut_format() truncates to fit and returns the length it needs. Any unit whose string form was longer than 255 bytes therefore came back silently cut short, and callers re-parsed the truncated string. For a product of about 52 or more user-defined base units, convert_to_base() then failed with syntax error, cannot parse '...' or cannot convert ... into ....

Fix

• When the buffer is too small, format again into a buffer of the length ut_format() reports.
• Stop with an error on a negative return instead of returning an unset buffer. That path is currently unreachable, because r_error_fn throws first.
• Remove the dead == sizeof(buf) warning.

Tests

• New tests cover formatted lengths of exactly 255, 256 and 257 bytes, and 300 bytes, through both ud_parse() and convert_to_base(). They fail on main and pass with the fix.
• R CMD check: Status OK; tests 596 pass, 0 fail.

Relationship to #443

Both PRs touch the formatting helper, so whichever lands second merges the other. The conflicts, in src/udunits.cpp, NEWS.md and tests/testthat/test_udunits.R, are mechanical:

• keep that PR's R_ut_format() and move this retry into its format_unit();
• keep both NEWS entries and both tests, watching the shared closing }).

A trial merge gives 594 expectations and 0 failures.

🤖 Generated with Claude Code