Skip missing/non-finite coordinates when encoding, with a warning

[neon-ninja]

Problem

encode() passed NaN coordinates through an undefined double -> int cast in encode_polyline() (src/googlePolylines.cpp). On x86 the cast yields INT_MIN, so a NaN pair emitted invalid polyline characters (bytes outside the valid 63–126 range, e.g. 0f ce 76 d7) and — because the format is delta-based — corrupted every subsequent point too. decode() on such a string threw a raw C++ exception (basic_string::at: __n >= this->size()). R's NA_real_ is a NaN at the C level, so NA triggered the same behaviour. Relatedly, empty sf POINTs (whose coordinates are NaN) encoded as "??", which silently decodes to the real coordinate (0, 0).

Changes

• encode_polyline() now skips coordinate pairs where lon or lat is not finite (NA/NaN/Inf); every entry point (encode(), encodeCoordinates(), the sf/sfc path) warns with a count of pairs that could not be encoded.
• encode(byrow = TRUE) returns NA for such rows instead of skipping (and decode() already maps NA strings back to NA coordinates).
• Empty sf POINTs now encode to "" (with the warning) instead of "??"/(0, 0); the test pinning the old behaviour was updated.
• RcppExports regenerated with Rcpp::compileAttributes(); roxygen docs updated (including the previously undocumented precision argument of decode()) and man pages regenerated.
• New testthat coverage for NA/NaN/Inf inputs, byrow, and empty points; NEWS entry; version bumped to 0.8.8.9000.

Testing

Full testthat suite (with NOT_CRAN=true): 161 pass, 0 fail. R CMD check (no vignettes/manual) shows nothing beyond pre-existing environment notes.

🤖 Generated with Claude Code

https://claude.ai/code/session_01UYymBnTDZbXyuus9FY97pd

---

Generated by Claude Code