[Rd] Feature suggestion: %!in%

Fri Dec 5 07:30:56 CET 2025

> Date: Tue, 2 Dec 2025 10:13:11 +0000
> From: Taras Zakharko<taras.zakharko using uzh.ch>
> 
> I wonder what would be the formal relationship between %in% and %notin%, which will be relevant if these operators are made generic in the future.  Is `%notin%` defined as a boolean negation of `%in%`, is it the other way around, or are they entirely independent definitions that just happen to have complementary semantics?
> 
> I would be in favor of defining %notin% as a sugar for !x %in% y
> 

'match' and transitively '%in%' are already generic-like, because 'match'
converts classed arguments using generic function 'mtfrm', for which anyone
can write methods if the default one does not suffice.  For that reason
(among others), I would not hold my breath for a generic base::`%in%`.

Or maybe you refer to generic functions named '%in%' defined in contributed
packages, masking base::`%in%`?  But then, due to namespacing, methods for
those generic functions would not be dispatched by base::`%notin%` defined
as you propose.

I might as well mention that since svn rev 89099 we have

     > dput(base::`%notin%`)
     function (x, table)
     match(x, table, nomatch = 0L) == 0L

Mikael

> Best,
> 
> Taras
> 
> 
>> On 1 Dec 2025, at 21:23, Kevin Ushey<kevinushey using gmail.com> wrote:
>>
>> Another useful data point: a large number of CRAN packages also define
>> their own %nin% / %notin% operators, e.g.
>>
>> https://github.com/search?q=org%3Acran+%25nin%25&type=code
>> https://github.com/search?q=org%3Acran+%25notin%25&type=code
>>
>> I think the broad usage of the operator, and the consensus over its
>> implementation, makes it a strong candidate for inclusion in R itself.
>>
>> I imagine a similar justification was used when %||% was added to base
>> R as well (which I was very glad to see!)
>>
>> Best,
>> Kevin
>>
>> On Fri, Nov 28, 2025 at 3:12 AM Duncan Murdoch<murdoch.duncan using gmail.com> wrote:
>>> On 2025-11-27 6:09 p.m., Simon Urbanek wrote:
>>>> Given that the args of tools:::%notin% don’t match %in% I'd say it was just a local use more than any deep thought about general use.
>>>>
>>>> Personally, I really like the idea of %notin% because it is very often that you start typing foo[foo %in% and then realise you want to invert it and the preceding negation is then cognitively sort of in the wrong place (reads like "not foo"). I also like %notin% better than %!in% because I think a salad of special characters makes things harder to read, but that may be just subjective.
>>> I agree with both points.  I generally use inefficient and unnecessary
>>> parens, i.e. `foo[!(foo %in% baz)]`.
>>>
>>>> And to your 'why bother' question - I do think it’s better to standardise common operators in core rather than have packages re-define it each time. And certainly just importing something that trivial from another package is a bad idea given the dependency implications.
>>> If someone is willing to put up with the fallout from the "masked"
>>> messages, then I'd also be in favour.  (And I'd choose %notin% rather
>>> than %!in% or %nin%, but whoever is willing to do the work should make
>>> that choice.)
>>>
>>>> (On the flip side: if you start using it you need to depend on recent
>>> R which may not be feasible in some environments, but then if that was
>>> always the argument we’d never add anything new :P).
>>>
>>> Or depend on the backports package.
>>>
>>> Duncan Murdoch
>>>> Cheers,
>>>> Simon
>>>>
>>>>
>>>>> On 28 Nov 2025, at 08:24, Duncan Murdoch<murdoch.duncan using gmail.com> wrote:
>>>>>
>>>>> On 2025-11-27 11:58 a.m., Marcelo Ventura Freire wrote:
>>>>>> If it is not a rhetorical question about a closed issue (if it is, tell me and I will shut up), this inclusion [1] would be useful (since it was exported and rewritten so many times by so many people and will keep being), [2] would create an uniformization (since it was and will be written under so many names before), [3] would not break stuff (since it is not altering the interface of any already existing function nor it is overwriting any symbol with a diverse use), [4] would not be neither a complex nor a tiringsome inclusion (even I myself could do it in a single 1-line pull request, hypothetically speaking) and [5] would benefit users all around.
>>>>>> I am not naive to the point of believing that an alteration to the R core would have few repercussions and surely there must be reasons why it was not done before.
>>>>> I don't know why it was added to tools but not exported, but here is my guess:
>>>>>
>>>>> - A member of R Core agrees with you that this operator is useful. This appears to have happened in 2016 based on the svn log.
>>>>> - It already existed in some contributed package, but base packages can't import anything from non-base packages, so it needed to be added.
>>>>> - It wasn't exported, because that would break some packages:
>>>>>     - the ones that export something with that name would now receive a check message about the conflict.
>>>>>     - if those packages stopped exporting it, then any package that imported from one of them would have to stop doing that, and import it from the base package instead.
>>>>> - It is very easy to write your own, or to import one of the existing ones, so a lot of work would have been generated for not very much benefit.
>>>>>
>>>>> R Core members try to be careful not to generate work for others unless there's enough of a net benefit to the community.  They are very busy, and many authors of contributed packages who might be affected by this change are busy too.
>>>>>
>>>>>
>>>>>> But, in the end, this inclusion would be just a seemingly unharmful syntax sugar that could be shared, like it was with "\" for the reserved word "function", but with waaaay less work to implement.
>>>>> The difference there is that it added new syntax, so as far as I know, it didn't affect any existing package.  Personally I don't see that it really offered much of a benefit (keystrokes are cheap), but lots of people are using it, so I guess some others would disagree.>
>>>>>> If it is not a dumb proposal, I can just include it in the wishlist of features in Bugzilla as prescribed in the contributor's page or I can do that PR myself (if you propose more work to others, the sensible thing to do is at least to offer yourself to do it, right?). In either case, I create more work to the dev team, perhaps to different people.
>>>>> It's hard for you to do the coordination work with all the existing packages that use a similar operator, so I don't think that's really feasible.
>>>>>
>>>>>> Thanks for taking your time to answer me.
>>>>> No problem.  I'm sitting in an airport waiting for a plane, so any distraction is a net benefit for me!
>>>>>
>>>>> Duncan Murdoch>
>>>>>> Marcelo Ventura Freire
>>>>>> Escola de Artes, Ciências e Humanidades
>>>>>> Universidade de São Paulo
>>>>>> Av. Arlindo Bettio, 1000,
>>>>>> Sala Paulo Freire (Sala Coletiva 252), Prédio I1
>>>>>> Ermelino Matarazzo, São Paulo, SP, Brasil
>>>>>> CEP 03828-000
>>>>>> Tel.: (11) 3091-8894
>>>>>> Em qui., 27 de nov. de 2025 às 14:15, Duncan Murdoch <murdoch.duncan using gmail.com <mailto:murdoch.duncan using gmail.com>> escreveu:
>>>>>>     The R sources already contain an operator like that, though it is not
>>>>>>     exported.  tools:::`%notin%` is defined as
>>>>>>        function (x, y)
>>>>>>     is.na<http://is.na>(match(x, y))
>>>>>>     Several CRAN packages export a similar function, e.g. omnibus, mefa4,
>>>>>>     data.table, hutils, etc. So I think if it was exported by R that's a
>>>>>>     better name, but since it is easy to write yourself or import from some
>>>>>>     other package, why bother?
>>>>>>     Duncan Murdoch
>>>>>>     On 2025-11-27 9:19 a.m., Marcelo Ventura Freire via R-devel wrote:
>>>>>>> Hello, dear R core developers
>>>>>>>
>>>>>>>
>>>>>>> I have a feature suggestion and, following the orientations in
>>>>>>> https://contributor.r-project.org/rdevguide/chapters/
>>>>>>     submitting_feature_requests.html <https://contributor.r-project.org/
>>>>>>     rdevguide/chapters/submitting_feature_requests.html>,
>>>>>>> I have searched in Bugzilla to the best of my capabilities for
>>>>>>     suggestions
>>>>>>> like the one I have in mind but found no results (however, I can
>>>>>>     be wrong).
>>>>>>> My idea is including this line
>>>>>>>
>>>>>>> `%!in%`  <- function(x, table) match(x, table, nomatch = 0L) == 0L
>>>>>>>
>>>>>>> between lines 39 and 40 of the file "src/library/base/R/match.R".
>>>>>>>
>>>>>>> My objective is to create a "not in" operator that would allow us
>>>>>>     to write
>>>>>>> code like
>>>>>>>>     value %!in% valuelist
>>>>>>> instead of
>>>>>>>>     ! value %in% valuelist
>>>>>>> which is in line with writing
>>>>>>>>     value1 != value2
>>>>>>> instead of
>>>>>>>>     ! value1 == value2
>>>>>>> I was not able to devise any reasonable way that such inclusion
>>>>>>     would break
>>>>>>> any already existing heritage code unless that operator would be
>>>>>>     defined
>>>>>>> otherwisely and it would improve (however marginally) the
>>>>>>     readability of
>>>>>>> future code by its intuitive interpretation and by stitching
>>>>>>     together two
>>>>>>> operators that currently stand apart each other.
>>>>>>>
>>>>>>> So, if this suggestion was not already proposed and if it is seen as
>>>>>>> useful, I would like to include it in the wishlist in Bugzilla.
>>>>>>>
>>>>>>> I would appreciate any feedback, be it critic or support, and I
>>>>>>     hope I have
>>>>>>> not crossed any communicational rule from the group.
>>>>>>>
>>>>>>> Many thanks!  😄
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> Marcelo Ventura Freire
>>>>>>> Escola de Artes, Ciências e Humanidades
>>>>>>> Universidade de São Paulo
>>>>>>> Av. Arlindo Bettio, 1000,
>>>>>>> Sala Paulo Freire (Sala Coletiva 252), Prédio I1
>>>>>>> Ermelino Matarazzo, São Paulo, SP, Brasil
>>>>>>> CEP 03828-000
>>>>>>> Tel.: (11) 3091-8894
>>>>>>>
>>>>>>>       [[alternative HTML version deleted]]
>>>>>>>
>>>>>>> ______________________________________________
>>>>>>> R-devel using r-project.org <mailto:R-devel using r-project.org> mailing list
>>>>>>> https://stat.ethz.ch/mailman/listinfo/r-devel <https://
>>>>>>     stat.ethz.ch/mailman/listinfo/r-devel>
>>>>>>
>>>>> ______________________________________________
>>>>> R-devel using r-project.org mailing list
>>>>> https://stat.ethz.ch/mailman/listinfo/r-devel
>>>>>
>>> ______________________________________________
>>> R-devel using r-project.org mailing list
>>> https://stat.ethz.ch/mailman/listinfo/r-devel
>> ______________________________________________
>> R-devel using r-project.org mailing list
>> https://stat.ethz.ch/mailman/listinfo/r-devel
> 
> 
> ------------------------------