>>>>> Kurt Hornik writes:
Jeroen,
I now took a closer look.
The check code by default uses tools:::.check_packages_used() on the
code in the installed package, and hence cannot pick up top-level calls
as these got processed when installing. In principle one can also run
tools:::.check_packages_used() on the actual source code, which for
HDBRR nicely gives
R> tools:::.check_packages_used(dir = "/data/rsync/PKGS/HDBRR")
'::' or ':::' import not declared from: ‘bigparallelr’
'library' or 'require' calls not declared from:
‘MASS’ ‘bigparallelr’ ‘bigstatsr’ ‘numDeriv’ ‘parallel’
'library' or 'require' calls in package code:
‘MASS’ ‘bigparallelr’ ‘bigstatsr’ ‘numDeriv’ ‘parallel’
Please use :: or requireNamespace() instead.
See section 'Suggested packages' in the 'Writing R Extensions' manual.
Not so nicely, apparently there are quite a few packages with similar
issues, but even worse, running tools:::.check_packages_used() on the
actual source code fails for several packages, so I will need to look
into fixing this.
Also, the code in installed package variant knows to look for S4 things
but not for code in reference classes (from methods or R6).
Now, in any case tools:::.check_packages_used() would only pick up calls
related to package use, and other top-level non-assignment calls can
have undesirable side effects when installing. One could do something
like
R> x <-
tools:::.find_calls_in_package_code("/data/rsync/PKGS/HDBRR",
which = "code")
Filter(length,
lapply(lapply(x, tools:::.call_names),
setdiff,
c("<-", "=", "if")))
$`HDBRR/R/HDBRR.R`
[1] "require" "bigparallelr::set_blas_ncores"
to find such calls.
(One might wonder why the require(bigparallelr) etc without having
bigparallelr as a dependency make no problem: that's because it is a
recursive strong dependency.)
Best
-k
>>>>> Jeroen Ooms writes:
> Thanks. Let me try to (find the time to) take a closer look (including
> reminding myself of how the package dependency check code works ...)
> Best
> -k
>> I encountered a few CRAN packages that do surprising things in their
>> R/ files, that perhaps R could warn against.
>> WRE states "Ideally, the R code files should only directly assign R
>> objects and definitely should not call functions with side effects
>> such as require and options. One example of a package that does
>> exactly these things:
>> https://github.com/cran/HDBRR/blob/master/R/HDBRR.R#L1-L15 .
>> Interestingly, it calls require(bigparallelr) and then a function from
>> bigparallelr, without even declaring this dependency in the
>> DESCRIPTION, which goes unnoticed in CMD check.
>> I suspect the author wants this to run this on package-load time,
>> while in fact this happens during install time. Hence it does not have
>> the intended effect and leads to strange behavior on build-servers.
>> ______________________________________________
>> [email protected] mailing list
>> https://stat.ethz.ch/mailman/listinfo/r-devel
> ______________________________________________
> [email protected] mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel
______________________________________________
[email protected] mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel
