Classify why a package installation failed

repro_install_deps()/repro_install_deps_docker() already capture the raw error text for a failed package, but the raw text alone does not say what KIND of failure it was — and the distinction matters for what a reader should do about it: a package genuinely gone from CRAN with no successor is a real, permanent problem (worth reporting as-is), a transient network hiccup is worth an automatic retry rather than treating as conclusive, and a missing compiler toolchain is an environment-setup gap, not a defect in the paper's own code. None of these change whether the install COUNTS as a failure (they still do, and still stay out of the traffic light — see the module's own dependency-unavailable handling), only how informative the report is about the subtype.

Usage

.repro_classify_install_message(msg)

Arguments

msg

the error message text (repro_install_deps()'s own message column for one failed row)

A package that Docker's second pass found in none of the places it looks (Bioconductor, the CRAN Archive, the list of known places, R-universe) is "source_unknown": the code loads it without saying where it comes from.

Value

a single category string: "source_unknown", "cran_unavailable", "compile_failure", "network", "transitive_dependency_missing", or "other"

Details

Matched against the install error message's own text with a fixed set of regexes, most-specific first, so a message that could plausibly match more than one category (rare, but not impossible) is not misassigned to a looser pattern. A message matching none of them is "other" — the raw message column is always still the authoritative text; this is advisory framing on top of it, not a replacement for reading the message.

install.packages()'s own standard CRAN-unavailability message — "package 'x' is not available for this version of R" — is CRAN's ONE generic wording for two different real causes (the package was removed/ archived entirely, or a current version simply never supported this R release) and does not itself say which one applies; confirmed directly against real R output rather than assumed. Splitting those into two categories from the message text alone would be a guess this classifier does not make — both are reported as "cran_unavailable", and the CRAN Archive retry's own via_archive result elsewhere already provides the real, non-guessed signal for the archived case (repro_install_deps()), since a successful Archive install proves that specific cause rather than inferring it from wording.