Use !anyNA() over all(!is.na(.))

[MichaelChirico]

all(!(.)) is logically equivalent to !any(.); in this case, that means we can use the specialized anyNA() instead as well

[gavinsimpson]

There's also an instance in R/ordConstrained.R

vegan/R/ordConstrained.R

Line 248 in 48b7543

if (all(!is.na(wa)))

Does this make a practical difference? - Are there examples of this being a performance bottleneck?

[MichaelChirico]

to me it's about readability as much as performance -- I find "aggregate + negate" easier to process than "negate + aggregate".

Came across it because our (very) old version was tripping the previous error from if (!is.na(wa)) when length(wa)>1, not for performance reasons.

[gavinsimpson]

OK - I'll deal with this and the extra instance I found (plus a more thorough check for other instances than the GH search I did to find the ordConstrained one).

Thanks :-)

[MichaelChirico]

FWIW this will eventually be available as the OuterNegationLinter:

r-lib/lintr#884

the problem with writing all those linters is now I can't help but see them everywhere :)

(PS in the meantime, I can run it manually and file PRs on any package if you'd like)

[gavinsimpson]

Thanks @MichaelChirico

[jarioksa]

Just a historic note: anyNA was introduced in R version 3.1.0 in April 2014, and the code you changed is from 2001. I think this is "syntactic sugar", although the docs say that

The default method for anyNA handles atomic vectors without a
class and NULL. It calls any(is.na(x)) on objects with
classes and for recursive = FALSE, on lists and pairlists.

It is more readable, though. The release note says that it may be faster for atomic vectors (without a class).

[gavinsimpson]

Yep; I triggered the GH Actions checks for the PR to insure that we weren't inadvertently bumping the minimum version of R we require.