Ignore NAs when pruning cache files

[hadley]

file.info() returns NA for files it can't stat, which happens when a cache file is deleted (e.g. by another process, or an OS temp cleaner) between when cache_info() lists it with dir() and when it stats it. These NAs propagated into to_remove:

• if (any(to_remove)) errors with "missing value where TRUE/FALSE needed" when there's an NA but no TRUE.
• Where there is a TRUE, info[!to_remove, ] turns NA-indexed rows into all-NA rows, corrupting later pruning passes.

cache_prune_files() now coerces NA entries in to_remove to FALSE, leaving such files in place.

🤖 Generated with Claude Code

[jeroen]

What leads to this is info being a zero-row data frame:

Browse[1]> info
[1] name  size  mtime
<0 rows> (or 0-length row.names)

Browse[1]> info$mtime + max$age
POSIXct of length 0

Browse[1]> info$mtime + max$age < Sys.time()
[1] NA

[hadley]

Hmmmm, but

.POSIXct(numeric())
#> POSIXct of length 0
.POSIXct(numeric()) < Sys.time()
#> logical(0)

^{Created on 2026-07-01 with reprex v2.1.1}

[jeroen]

Okay so here is the full reproducer for completeness:

unlink(tools::R_user_dir("gh", which = 'cache'), recursive = TRUE)

library(messydates) # Breaks date comparisons!
gh::gh('/repos/cran/manydata')

The bug is in the messydates package which when loaded changes behavior of date comparisons:

numeric() < Sys.Date()
# logical(0)

library(messydates)
numeric() < Sys.Date()
# [1] NA

[hadley]

Seems like this isn't httr2's problem to fix 😄