get_id() subtly alters the session randomization state

[MichaelChirico]

processx/R/utils.R

Line 198 in 95ba3aa

paste(sample(c(LETTERS, 0:9), 10, replace = TRUE), collapse = ""),

Started way upstream as r-lib/pkgload#272 and then r-lib/pkgbuild#185, but the issue traces back to the above -- while trying to build a package, {processx} generates this random ID which messes with the randomization state of the session.

I don't see this documented as intended behavior, so I'm wondering if this is unintentional & perhaps get_id() should be careful to restore the randomization state on completion; of course this means consecutive runs of {processx} may generate sessions with the same ID --> not desirable, so maybe the caller needs to deal with this & {processx} just needs to document the issue?

Alternatively, maybe get_id() could use the tempfile() randomization to generate the IDs, since tempfile() does not touch the R-level randomization state?

[luwidmer]

@gaborcsardi it looks like this fix has not landed on CRAN yet? I just tracked down an irreproducible first model fit in the {rbmi} package to this issue (see link above), and installing the current version off Github using remotes::install_github does fix it.

[luwidmer]

Many thanks for releasing 3.8.5, @gaborcsardi 🎉!