parSim

• parSim_dt(write = TRUE) now writes to a temporary file when name is missing (instead of erroring), reports the path with a message, and returns the results invisibly instead of NULL.
• In both parSim() and parSim_dt(), design conditions passed via ... whose names collide with a function argument (e.g. replications, nCores, progress) now trigger a warning, since they would otherwise silently become a crossed design factor.

New features

• parSim_dt() gained the packages argument of parSim() (packages to load on the parallel workers).
• New seed argument in parSim() and parSim_dt(): one L'Ecuyer-CMRG RNG substream is derived per simulation condition, so a seeded simulation is fully reproducible and gives identical results for any value of nCores. The caller's RNG state and RNG kind are restored afterwards.

Bug fixes

• User result columns named id, error or message are no longer silently overwritten: they are renamed to <name>_result with a warning. A warning is also given when a result column duplicates a design-condition name (the join then suffixes the columns .x/.y).
• export/env now also work when running sequentially (nCores = 1): the simulation expression (and exclude) are evaluated with the caller's environment as enclosure, so objects local to a calling function are found without being in the global environment. Previously this only worked in parallel runs (the completion of the fix for GitHub issue #10), making "works with 8 cores, fails with 1 core" a confusing failure mode.
• BREAKING BUG FIX: the exclude argument of parSim_dt() did the opposite of its documentation -- it *kept* only the rows matching the exclusion expressions instead of removing them. Matching rows are now removed. Multiple expressions are combined with OR (any match removes the row); previously they were combined with AND.
• parSim_dt() no longer adds a spurious rn column when the simulation expression returns a data frame (row names are dropped), so list and data-frame results now produce identical output schemas.
• Removed LazyData: true from DESCRIPTION (the package has no data directory), fixing an R CMD check NOTE.
• The deprecated cores/save arguments are now only intercepted when they look like the deprecated scalar arguments; a design condition legitimately named cores (e.g. cores = c(1, 2, 4)) is kept as a condition. nCores is validated to be a single integer >= 1.
• Clear, early error messages for invalid input: zero-length simulation conditions, replications < 1, and designs left empty after exclude (previously crashed later with unrelated-looking errors).

Documentation

• The supercomputer vignette is no longer written around one specific cluster. It now describes a generic Slurm-based cluster, and the dead link to the decommissioned Lisa Computing Cluster has been removed.