Skip to contents

Building a slot's index records

index_records.Rd

A slot's PACKAGES* files hold one record per tarball in the slot. The tarballs are binaries, but their records are read from each package's CRAN DESCRIPTION, fetched from the GitHub mirror of CRAN: one small request per package instead of downloading and unpacking a binary that can run to tens of megabytes. A package the mirror does not have falls back to the DESCRIPTION inside its tarball.

Details

Rebuilding every record on every run would mean ~25k requests per slot, so the previously published index is the cache: a record is reused while its tarball is still listed with the same ETag, and only new or replaced tarballs are read.