Input hash of one LANDIS-II replicate — landis_input

A digest of what a replicate was run from: the MD5 of every dependency file (keyed by path), the base seed, the replicate index and the scenario file name. tar_landis() writes it to <rep>/log/input_hash.json after a run and compares it before the next, via landis_rep_is_current().

Usage

landis_input_hash(
  dep_files,
  base_seed,
  rep_index,
  scenario_file,
  collation = NULL
)

Arguments

dep_files: Character. The dependency files staged into the replicate.
base_seed: Integer. The base random seed.
rep_index: Integer. The replicate index.
scenario_file: Character(1). The scenario file name.
collation: NULL (the default) for byte order. Otherwise an LC_COLLATE locale to sort in, which reproduces the hash landisutils 0.0.149 and earlier wrote in that locale; "" means the locale the process's environment selects, as when R started.

Value

Character(1), a SHA-1 digest; NA_character_ if collation cannot be set.

Details

Paths are ordered by bytes (sort(method = "radix")), so the hash does not depend on the collation locale of the R process computing it. R collates with ICU in UTF-8 locales and by bytes under C/POSIX; a locale-sorted hash differs between a worker started with and without LANG.

R also collates by bytes, whatever Sys.setlocale() says, while the environment variable LC_ALL (or, when that is unset, LC_COLLATE) is C. For a named collation the sort therefore runs with LC_ALL unset and LC_COLLATE set to collation, as in a process started in that locale, and both are restored afterwards.