tidync

• Zero file I/O construction: supply coordinate values directly with concat_dim = list(name = "time", values = dates) and only the first source is opened (as the template). Values can be numeric, Date, or POSIXct, and hyper_filter() operates on them directly. This is aimed at file-database workflows (for example raadfiles) where per-file coordinates are already known. This construction assumes one step per source; files holding more steps on the concat dimension are detected at read time and reported as an error rather than silently subset.
• Read-time validation is armed automatically for fast mode and for values-supplied construction: shared dimensions and the per-source concat dimension length are checked against the consolidated view as each file is opened, with clear errors naming the file.
• Optional parallel reads via mirai: when mirai::daemons() are active, per-source reads in hyper_array() run in parallel with mirai::mirai_map(), falling back to sequential reads when mirai is not installed or no daemons are set. Parallelism is entirely under user control; tidync never starts daemons itself. Errors raised during parallel reads (including validation failures) propagate normally.
• Sources are kept in the order supplied; the consolidated index on the concat dimension is file order. Value-based filters work regardless of ordering, but supply sources in coordinate order for a monotonic view.
• tidync() on a tidync_data object (round-trip) preserves the source table and concat_dim.

Known limitations

• Non-contiguous selections *within* a single source still read a bounding slab and are not supported at hyper_array() time, consistent with single-source behaviour. Non-contiguous selections that span whole sources (for example dropping intermediate files) are supported and read only the files required.

Fixes

• Fixed hyper_tibble() producing constant dimension values when drop = TRUE (the default) collapses arrays to bare vectors; expansion is now based on length() rather than prod(dim()).
• Fixed the print method showing NA for min/max of Date/POSIXct coordinate columns; non-numeric coordinates are now converted with as.numeric() (Date as days since epoch) rather than replaced with NA.
• Repaired mangled pipe operators in the package overview help examples.
• Removed commented-out browser() calls.
• Update to CFTime as_timestamp(), thanks to @fabern.
• Support for CF time metadata via package CFtime thanks to @pvanlaake, see https://github.com/ropensci/tidync/pull/124
• Suppress ncdf4 warnings on open, issue #119.

Dependencies and internals

• Minimum R version is now 4.1.0 (native pipe |>). Removed forcats, magrittr, and purrr dependencies; %>% replaced with |> throughout and purrr::safely() replaced with tryCatch().
• ncmeta (>= 0.5.0) is now required, for extended (CFtime) metadata and for correct handling of sources with no variables (such as compound-type L3 bin files), which errored uninformatively with earlier ncmeta under current dplyr.
• Minimum dplyr version is 1.1.0 (multiple argument in joins) and minimum tidyr version is 1.0.0 (cols argument in unnest()); all version-gating conditionals removed.
• mirai added to Suggests.
• print.tidync now returns its input invisibly, as always claimed.