SmokingHistoryGenerator

7.0.0 (unreleased)

Version numbering

• Package releases now use standard R semver (starting at 7.0.0). The former {CLI engine}-{R wrapper} form (e.g. 6.5.3-1.0.1 on CRAN) is retired.

Memory safety (CRAN gcc-ASAN / valgrind)

• Fix seed string alloc/dealloc mismatch in RunWebVersion (new[] / delete[] consistently).
• Fix uninitialized error message when the legacy error file cannot be opened.
• Resolve fast_itoa ODR conflict between R wrapper and simulation engine.

Path layout (shorter defaults)

• Default table subfolders renamed from smoking/ and mortality/ to smok/ and mort/ under inst/extdata/2018/ and in factory defaults.
• Test fixtures moved from tests/testdata/NHIS-1965-2018/ to tests/testdata/2018/.
• shg_load_params() still accepts legacy smoking/ / mortality/ layouts in downloaded zips and normalizes them to smok/ / mort/ in the cache.

Race defaults

• Legacy config files may omit RACE= (defaults to 0).
• CSV parameter tables may omit a RACE column (header starts with SEX; all rows treated as race 0).
• shg_run() already defaulted omitted race and sex to 0 in config lists.

Tooling and docs

• Add ./tools/check-docker-asan.sh for local CRAN-faithful gcc-ASAN checks (Docker).
• README: CRAN-first install and minimal shg_run() quick start with bundled extdata.
• Maintainer docs no longer reference shg-cli sync as source of truth.

6.5.3-1.0.1

Breaking: split parameter bundles (shg-params)

• Removed params_bundle_source and single-zip shg_load_params(url = ...).
• Required separate smoking and mortality releases: smok_params_source, mort_params_source, and mort_params_type ("acm" or "ocm"; was params_mortality).
• shg_load_params() downloads/caches each zip, merges params/ tables into engine layout smoking/ + mortality/.
• Portable YAML and getConfig() / getReproConfig() use the new provenance field names.

6.5.3-1.0.0 (2026-05-18)

• Synced core engine from shg-cli v6.5.3.
• Windows: open bundled CSV tables in binary mode so ftell/fseek dimension scans work; strip CR/LF on all CSV lines in CPD, initiation/cessation, and mortality loaders so missing "." fields parse correctly with CRLF.

6.5.2-1.0.0 (2026-05-11)

Bundled inputs and YAML

• Bundled inputs: Default factory input_data_folder is system.file("extdata", "2018", package = "SmokingHistoryGenerator") (NHIS-1965–2018 csv-partial with cohort columns 1940, 1950, 2010). Removed transitional inst/extdata/2016/, NHIS-1965–2016 test fixtures, and tests/fixtures/2016/ XML goldens; tests and docs use the 2018 tree only. Regenerate the 2018 partial from tests/testdata/2018/csv-complete/ using Rscript tools/refresh-nhis-2018-csv-partial.R.
• Layout: Under each year folder, smoking tables live in smoking/, mortality in mortality/; factory defaults use relative paths (smoking/initiation.csv, …).
• Portable / written YAML groups params_bundle_source, params_mortality, and optional folder paths under a params: map; shg_load_config / shg_apply_config accept nested or flat keys.

Reproducibility

• getReproConfig() / portable YAML omit num_threads (effective segment count and seeds define the run; thread count defaults to auto on reload).
• getReproConfig() exports package_repro as r_package_version only (full install metadata remains internal for fingerprint checks).
• Run bundles with attach_run_info = TRUE enrich repro_config with a nested results block (content_md5, compact summary) and a single repro_digest (engine settings plus R session). Legacy flat results_* / repro_engine_md5 / r_session_md5 keys in YAML are merged or dropped on load.
• Summaries omit sentinel -999; age_at_death is split for never vs ever death-age stats (mean, sd, n_obs). Top-level ever_smokers holds count, fraction, and integer cpd_mode (most common rounded CPD among ever smokers). Keys count (never/ever totals) and n_obs (rows contributing to each mean/sd) replace bare n (YAML 1.1 reserves n). Initiation and cessation means use ever smokers only; age_at_death$ever_smokers is only death-age statistics (not the same list as top-level ever_smokers).
• shg_save_config(..., results = ) optionally writes those verification fields (including content_md5 and compact results summary metadata when results is supplied).

Configuration

• CLI sync metadata: src/shg-cli-info.txt (YAML map shg-cli: with MostRecentTag, CommitHash, SrcHash; listed in .Rbuildignore so it is not shipped in CRAN source tarballs); python tools/shg-sync.py update-description refreshes it from the sibling shg-cli checkout. R merges these into the packageDescription() list as SHGMostRecentTag, SHGCommitHash, and SHGsrcHash when the file is present (for example devtools::load_all() from a checkout). The old RWrapperVersion field is dropped.

• shg_load_params() (URLs): shg.params.download.timeout_sec (default 600) and shg.params.download.connect_sec (default 60) when httr2 is installed; clearer HTTP/network errors; HTML and non-zip responses detected before unzip.

• shg_reset_defaults() / shg$reset_to_factory_defaults() restore engine fields to the same defaults as a fresh SHGInterface.

• shg_apply_config(shg, config) resets defaults, then applies a sparse or full named list via useConfig(), so partial YAML/intent configs do not inherit stale instance state.

• shg_apply_config() with params_bundle_source now calls shg_load_params() the same way as shg_load_config() (clears derived paths, restores the bundle). Without a bundle, explicit input_data_folder / table filenames in the list are still applied.

• shg_load_config() now starts from factory defaults before applying the YAML bundle (via reset_to_factory_defaults() in the bundle applier).

• shg_write_config_yaml(config, path) serializes any config list: drops audit keys, and strips redundant table paths when params_bundle_source is present (shape-driven “portable” output).

• Config lists and YAML may use mortality as an alias for params_mortality. Normalization uses [[ only so it does not partially match mortality_filename.

• Factory defaults: mortality file is mortality/acm.csv, matching shg_load_params(..., mortality = "acm") bundle layout.

Simulation return value

• shg$runSimFromFixedValues(..., attach_run_info, original_config) and shg$runSimFromDataFrame(..., attach_run_info, original_config) (6-argument forms): when attach_run_info is TRUE, the return value is a list with results, original_config (sparse intent; default for fixed cohort = repeat/race/sex/cohort_year), repro_config (post-run getReproConfig(FALSE)), and run_info (host/software/audit).
• The 4- and 1-argument forms keep the previous behavior: a data.frame of simulation output (attach_run_info = FALSE).
• shg_run() / shg$runSim() accept attach_run_info (default TRUE; set FALSE for data-frame-only return).
• shg_run() / shg$runSim(): if repeat, individuals, and N are all omitted, repeat defaults to 1000.

Note

Direct shg$useConfig() without shg_apply_config() still overlays on the current instance (legacy). Prefer shg_apply_config() for defaults-first semantics.