network 1.16-378: ERROR
martinamorris opened this issue · comments
Started because CRAN sent an email to @CarterButts saying network was generating errors with a new code-checking package rchk for the development version of R, and that these errors needed to be fixed by a specific date or else... See initial discussion in #15
Most of this exchange occurred in email, and I'm transporting it here for the record.
- Carter was unable to reproduce the errors using our regular build tools, and there is more than one version of
rchkaround, so it was not clear which one CRAN is using. - @krivit found no errors using
valgrind - @krivit and @chad-klumb implemented different versions of
rchk, each of which generated a long list of errors associated withprotectstacks.
- Chad's implementation came from R-hub
Pavel's implementation:
I rebuilt the singularity image based on the recipe in
https://raw.githubusercontent.com/kalibera/rchk/master/image/singularity.def
, so it's against the latest R-devel and versions of everything, as far
as I can tell.
and he also noted:
Regarding the protection depth, I found a ticket about that: kalibera/rchk#21 [github.com] ,
which suggests that the checker can't handle situations in which PROTECT() and UNPROTECT() are inside a loop.
Meanwhile, however, @CarterButts says that CRAN has removed the link to the rchk output file that they sent him, so it is not possible to compare what Pavel and Chad found to what CRAN is complaining about.
Bottom line for now: We'll submit an update for network with the fix to is.color(), and ignore the rest. If CRAN wants to complain about rchk results we'll follow up.
fixed with commit 746a532 and 1.16 release