20260616/#413 double check errors in r

[Yicheng-Zhao1999]

Summary

Robustness fixes for the R globalmacrodata package, from the cross-language audit in #318. Every change was live-verified with R 4.5.3 against the live S3 backend (download paths, bug repros, and a full regression after each change).

Two audit items are intentionally not changed here:

• Listed versions returning 404 (Shixuan #1): root cause is server-side data, not R code. Re-verified that all advertised vintages now return data on S3, so no code change is warranted.
• Doubled Error: prefix (Update NAMESPACE #3): could not be reproduced; left as-is.

Bug fixes

#	Problem	Fix
2	gmd(version = NA) crashed with a cryptic missing value where TRUE/FALSE needed	Validate version is a single non-NA character string (or NULL) at the top of gmd(), before any download
4	gmd(iso = TRUE, country = "USA") only warned, then silently discarded the extra args	Now errors (stop) on conflicting inputs; the guard also covers sources/cite
5	variables was case-sensitive while country/sources were not (variables = "rgdp" failed)	Case-insensitive matching, normalized to the canonical casing (e.g. rgdp → rGDP); invalid codes still error
6	Empty character vectors (country = character(0), …) downloaded the full ~21 MB dataset before failing	Reject empty country/variables (and empty/NA cite) up front, before any network request
7	start_year / end_year were not implemented (unused argument)	Added both parameters plus an apply_year_filter() helper applied across the main, raw, and sources paths; validated up front

Quality / robustness improvements (audit item #8)

• Dependencies: added Depends: R (>= 3.5.0) and version floors (readr (>= 2.0.0), httr2 (>= 1.0.0), haven (>= 2.0.0)); migrated from the superseded httr to httr2 (request/req_perform/resp_body_*, with req_error() disabled to preserve the non-200 → NULL behaviour).
• CSV parsing: replaced type guessing with explicit col_types on the fixed-schema CSVs (versions, varlist, isomapping, source_list, bib) and the stable id columns of the raw CSV (ISO3/countryname/year); dynamic source columns remain inferred.
• Diagnostics: .gmd_safe_get now logs the HTTP status code (or the error message) on failure, so users can distinguish a 404 from a timeout. Still returns NULL, so control flow is unchanged.
• Return type: all data-returning paths now return a plain data.frame (previously a mix of tbl_df / spec_tbl_df / data.frame); the vars list keeps its custom gmd_vars print class.

Testing

• Verified against the live S3 backend: all 8 download paths (text CSV + binary .dta), each bug's before/after repro, and a 13-point regression re-run after every change.
• start_year/end_year year windows, case-insensitive variable/source names, and the empty-vector/NA guards all behave as expected; existing happy-path calls are unchanged.

🤖 Generated with Claude Code