[R-pkg-devel] Native pipe in package examples

[Henrik Bengtsson]

On Thu, Jan 25, 2024 at 8:27 AM Duncan Murdoch <murdoch.duncan using gmail.com> wrote:
>
> On 25/01/2024 11:18 a.m., Henrik Bengtsson wrote:
> > On Thu, Jan 25, 2024 at 7:48 AM Duncan Murdoch <murdoch.duncan using gmail.com> wrote:
> >>
> >> On 25/01/2024 10:27 a.m., Josiah Parry wrote:
> >>> Hey all,
> >>>
> >>> I've encountered use of the native pipe operator in the examples for
> >>> 'httr2' e.g.
> >>>
> >>> request("http://example.com") |> req_dry_run()
> >>>
> >>>
> >>> Since r-oldrel (according to rversions::r_oldrel()) is now 4.2.3, can the
> >>> native pipe be used in example code?
> >>>
> >>> I do notice that the package httr2 requires R >= 3.6.0 which implies that
> >>> the code itself does not use the native pipe, but the examples do.
> >>
> >> I think that the package should state it requires R (>= 4.1.0), since
> >> that code won't work in earlier versions.
> >>
> >> I believe it's a syntax error before 4.1.0, but don't have a copy handy
> >> to test.
> >
> > Yes, support for the |> syntax was introduced in R 4.1.0;
> >
> > $ Rscript --vanilla -e "getRversion()" -e "1:10 |> sum()"
> > [1] ‘4.0.5’
> > Error: unexpected '>' in "1:10 |>"
> > Execution halted
> >
> > $ Rscript --vanilla -e "getRversion()" -e "1:10 |> sum()"
> > [1] ‘4.1.0’
> > [1] 55
> >
> >> That means the package won't pass R CMD check in those old
> >> versions.  If it wasn't a syntax error, just a case of using a new
> >> feature, then I think it would be fine to put in a run-time test of the
> >> R version to skip code that won't run properly.
> >
> > There's also the distinction of package code versus code in
> > documentation. If it's only example code in help pages that use the
> > native pipe, but the code in R/*.R does not, then the package will
> > still install and work with R (< 4.1.0).  The only thing that won't
> > work is when the user tries to run the code in the documented
> > examples.  I'd argue that it's okay to specify, say, R (>= 3.6.0) in
> > such an example.  It allows users with older versions to still use the
> > package, while already now migrating the documentation to use newer
> > syntax.
>
> Is there a way to do that so that R will pay attention, or do you mean
> just saying it in a comment?

As a "comment".

>
> I think you're right that syntax errors in help page examples will be
> installable, but I don't think there's a way to make them pass "R CMD
> check" other than wrapping them in \dontrun{}, and I don't know a way to
> do that conditional on the R version.

I think

$ R CMD check --no-examples --no-vignettes ...

would check everything else but examples and vignettes.

>
> I would say that a package that doesn't pass "R CMD check" without
> errors shouldn't be trusted.

Somewhat agree, but we still get some "trust" from the fact that the
package passes R CMD check --as-cran on R (>= 4.1.0).  Also, if the
maintainer documents something like "On R (> 4.1.0), the package
passes 'R CMD check --no-examples ...'; we use R (>= 4.1.0)-specific
syntax in some of the help-age examples", then there's additional
"trust" in it's working there.  But, yes, there's less "trust" here,
but I think it's okay for maintainers to declare "R (>= 3.6.0)" to be
backward compatible. Another way to put it, it would be extreme to
require "R (>= 4.1.0)" just because of a single "1:3 |> sum()" in some
example code.

/Henrik

PS. Personally, I'd skip the use of |> in examples to avoid these concerns.

>
> Duncan Murdoch