Skip to content

Time to revisit ifelse ?

12 messages · Antoine Fabri, Ben Bolker, Avraham Adler +6 more

#
Dear r-devel,

`ifelse()` has a lot of issues, and for these reasons it has been redone in
`dplyr::if_else()` and `data.table::fifelse()`, which are both great. Yet
it's an important base R function, it's really hard to program in base R
without it and scores probably as high as it gets in the most_used *
most_problematic metric.

Obviously we can't change it without breaking a ton of code, but with all
the experience we now have with it and the dplyr and data.table alternative
maybe it might not be absurd to have a good alternative, say `if.else` in
base R, that we can document on the same page and recommend for future use.
It would require a common type in yes/no, not return logical() for all zero
length input, work with dates, datetimes and factors, handle a na condition
etc. The test suites of dplyr and data.table probably tell us everything
about the edge cases we want to look at. Maybe the old ifelse could even
warn when called from the top level, to incite us to work with the new one.

It feels wrong to me to be stuck with ifelse() forever just because it has
been like this for a long time. I'm sure some of you learnt your way around
it but I work with R every day and after 10+ years of R it still bites me
all the time, I'm probably not alone, at least chatGPT called it a
"footgun", and we don't want that :).

Thanks,

Antoine
#
Rather than asking others to do this, why don't you create a tiny 
package containing nothing other than an ifelse() replacement?  I 
wouldn't want to depend on dplyr or data.table just to get their 
versions, but depending on your tiny package wouldn't be an issue.

Duncan Murdoch
On 2025-07-08 6:12 a.m., Antoine Fabri wrote:
#
It's not about asking others to do it really, that was a harsh assumption.
I'd be happy to propose a version if it helps, I'd be also very happy if it
were just a copy of if_else or fifelse (both MIT FWIW).
It's a low level building block and it's broken, IMO it's way better to
have it available and documented in base R and incite everyone to use it,
so not only we don't suffer from it in the code we write, but also in the
code we use or inherit from.

Le mar. 8 juil. 2025 ? 13:25, Duncan Murdoch <murdoch.duncan at gmail.com> a
?crit :

  
  
#
I think Duncan's point is that R-core are (reasonably) very, very, 
very conservative about adding things to base R. It would be useful to 
the community, and would indeed further the discussion, to make a tiny 
package containing just that function. (Even just copying it from some 
other package might require some work to disentangle it from 
dependencies: for example, a quick glance at dplyr::if_else shows that 
it uses functions from rlang, vctrs, ...)

   I'd be happy to accept a pull request in `gtools`, which is a 
zero-dependency (except base R) package for small utility functions ...

   cheers
    Ben Bolker
On 7/8/25 07:36, Antoine Fabri wrote:

  
    
#
I think the point is not that there needs to be a smaller package for yet
another if-else (https://xkcd.com/927/). It is that if the R-language, as a
whole, had a performant if-else in the base of the language would benefit
**everyone** such that a data.table or dplyr or gtools etc. alternative
would not be necessary.
On Tue, Jul 8, 2025 at 5:09?AM Ben Bolker <bbolker at gmail.com> wrote:

            

  
  
#
On Tue, Jul 8, 2025 at 10:55?AM Josiah Parry <josiah.parry at gmail.com> wrote:
While that may be true, Josiah, R Core's time is very limited.
Following Duncan's idea, if a small, simple package were created and
was proven to dominate the performance of standard ifelse without
causing any issues with the ten thousand plus packages in the R
environment, that would make R Core's decision much simpler, whether
or not to use the existing, proved performant code. Asking R Core to
do the research and testing for something which currently _works_,
albeit not in the most efficient way possible, is pretty much a
non-starter. Do as much work as possible for R Core to have even a
possibility of consideration. For something similar, albeit much less
core (pun intended) to R's code, see this discussion [1] from June
2012 on Kendall's tau, where the code already existed but was deemed
unimportant enough to add to base R.

[1] https://stat.ethz.ch/pipermail/r-devel/2012-June/064351.html

Thanks,

Avi
#
Avi, appreciate the puns!

I don't think anyone is suggesting R-core dedicate all of their time to
this problem.
To me, the thread is about consensus making (as there is no formal way to
do that).

Quoting OP here:

"It's not about asking others to do it really, that was a harsh assumption.

I'd be happy to propose a version if it helps, I'd be also very happy if it

were just a copy of if_else or fifelse (both MIT FWIW)."


The initial email, IMO, was to show that there already are community
implementations of faster
and type-safer of ifelse (notably dplyr and data.table and I'd add
kit::iif, too) and perhaps now
is the time to add this enhancement to the language.

It is tough to get a sense of total usage of each though but some code
searching on GitHub:

- data.table::fifelse:
https://github.com/search?q=%22fifelse%28%22+language%3AR&type=code
- dplyr::if_else:
https://github.com/search?q=%22if_else%28%22+language%3AR&type=code
- kit::iif:
https://github.com/search?q=%2F%28%3F-i%29iif%5C%28%2F+language%3AR+&type=code

On Tue, Jul 8, 2025 at 8:23?AM Avraham Adler <avraham.adler at gmail.com>
wrote:

  
  
#
On Tue, 8 Jul 2025 at 17:23, Avraham Adler <avraham.adler at gmail.com> wrote:
I might have missed it but the suggestion wasn't to replace the
existing ifelse code.
There are at least two existing proven performant code implementations.
One on a package (data.table) with no additional dependencies.
What would another implementation add?
Yet recently it was suggested by one R-core member as a possible
improvement for R suggesting that patches via bugzilla would be
appreciated [1]. Of course, this doesn't mean it would be the case for
this suggestion but seeing the interest of the community and how it
can help many useRs I think exploring how to integrate such a if.else
function on base R could be helpful for all (even if it is only added
to R source later).

[1]: https://github.com/r-devel/r-dev-day/issues/87
#
Since you and Antoine are volunteering to do the work, why not start in 
the way I suggested?  Write up a comparison of the known ifelse 
implementations, and either pick the best one, or choose the best parts 
of each.  Put the result in a package containing nothing else, and 
invite comment from the wider community.

My only comment in advance is that the package should have no 
dependencies other than base packages, for two reasons:

1. The hope is to have it adopted in base R, and for that it can't have 
any other dependencies.

2. If it's never adopted by R Core, I might still want to use it, but I 
don't want to add extra dependencies for just one little function.

Duncan Murdoch
On 2025-07-08 12:46 p.m., Josiah Parry wrote:
#
A package with only one function, what a concept!

But then it becomes tempting to also create a function like if_else_else() and if_else_default() and of course if_not_else() ....

Joking aside, plenty of functionality in extendible languages like R were written long enough ago that they might be done quite differently today. I don't mean just different code, but different arguments it takes and different defaults. Who really wanted you to type stringsAsFactors=TRUE every time, for example.

But it is late enough that making changes now can backfire. Look at the announcement that updating ggplot2 to use S7 objects will break some packages that depend on it including many in the Bioconductor world. I would imagine if R magically had started with S7 and skipped over S3 and S4 and perhaps others that have not been widely used, we might have had more consistency. But that is not how it happened and we likely will be stuck for a long time with both R core and all kinds of packages needing to be able to handle older kinds of objects. And, not to compare, but a language like python made a very different approach to objects long ago so that pretty much everything is an object and there generally is no need to create a new form as you can use all kinds of existing ways to tailor your objects to your needs.

Given the budgets and other constraints we are dealing with, I suspect that there is a long list of possible changes that are currently not being seriously considered and since there are several work-arounds available in existing packages, there is less urgency.

And, there is a reality that although an ifelse() has a certain generality, for specific purposes, it may be trivial enough to fashion your own variant using anything from explicit to implicit loops. But, certainly, if some people work on a variant and show it is compatible and benchmarks suggesting how much faster, then a minimal package with no odd dependencies is a way to go, and might eventually be taken into the core. I would be cautious about naming the package/function as whatever is chosen, ...

Avi  (too)

-----Original Message-----
From: R-devel <r-devel-bounces at r-project.org> On Behalf Of Duncan Murdoch
Sent: Tuesday, July 8, 2025 3:06 PM
To: Josiah Parry <josiah.parry at gmail.com>; Avraham Adler <avraham.adler at gmail.com>
Cc: r-devel at r-project.org
Subject: Re: [Rd] Time to revisit ifelse ?

Since you and Antoine are volunteering to do the work, why not start in 
the way I suggested?  Write up a comparison of the known ifelse 
implementations, and either pick the best one, or choose the best parts 
of each.  Put the result in a package containing nothing else, and 
invite comment from the wider community.

My only comment in advance is that the package should have no 
dependencies other than base packages, for two reasons:

1. The hope is to have it adopted in base R, and for that it can't have 
any other dependencies.

2. If it's never adopted by R Core, I might still want to use it, but I 
don't want to add extra dependencies for just one little function.

Duncan Murdoch
On 2025-07-08 12:46 p.m., Josiah Parry wrote:
______________________________________________
R-devel at r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-devel
#
Thanks Antoine for starting this discussion. It would indeed be great to see
an improved `ifelse()` in base R.

I also agree with Duncan's suggestion that the way to proceed would be to
create a package where the improved version could be drafted, discussed and
refined so that R Core would have a concrete proposal to consider in the end.

Some initial thoughts on what should be considered:

Performance has been mentioned a few times. While it would of course be nice
to see improvements there I think the main goal is in the API. The goal for
performance should rather be that it doesn't deteriorate unacceptably.

While data.table's and dplyr's ifelse variants may serve as a good starting
point for identifiying the improvements needed, I don't think either is a good
candidate for simply copying as the base R candidate. A function in base R
should adhere to the conventions in base R; neither of the packages does that.
They instead have their own stricter requirements. For example:

* Incompatible lengths: Base R recycles with a warning, both packages error out.
* Different classes: Base R coerces loosely, dplyr uses stricter coercion rules
  based on vctrs, and data.table doesn't allow any coercion.

Another point to consider is the handling of attributes for the result.
data.table copies from the first non-NA input from left to right, while dplyr
delegates to vctrs again for merging the attributes gracefully. This matters
for example for factors, where data.table special-cases them to require the
same levels, wherease dplyr merges them. For a base R solution, it would make
sense to delegate the attribute handling to `c()` somehow, as that's conceptually
what should be happening; we're combining values from the `yes` and `no` objects.

I'm sure there are many other points to consider, but as I said this is what
comes to mind at first. Best of luck with the effort.

Kind regards,

Mikko
On Tuesday, 8 July 2025 at 21:41, avi.e.gross at gmail.com <avi.e.gross at gmail.com> wrote:

            
Message-ID: <3znc9Ny2KoAlUEgqWJBsTpUdfaMXZrhSbrkPUFxQLlX82nfHZE0npbR4eFRpi7rXzjQoWz-hLSaUr9pnX9HYywDzzc3TGRi2FtNNLGFz9IE=@protonmail.com>
#
> Thanks Antoine for starting this discussion. It would indeed be great to see
    > an improved `ifelse()` in base R.

    > I also agree with Duncan's suggestion that the way to proceed would be to
    > create a package where the improved version could be drafted, discussed and
    > refined so that R Core would have a concrete proposal to consider in the end.

    > Some initial thoughts on what should be considered:

    > Performance has been mentioned a few times. While it would of course be nice
    > to see improvements there I think the main goal is in the API. The goal for
    > performance should rather be that it doesn't deteriorate unacceptably.

    > While data.table's and dplyr's ifelse variants may serve as a good starting
    > point for identifiying the improvements needed, I don't think either is a good
    > candidate for simply copying as the base R candidate. A function in base R
    > should adhere to the conventions in base R; neither of the packages does that.
    > They instead have their own stricter requirements. For example:

    > * Incompatible lengths: Base R recycles with a warning, both packages error out.
    > * Different classes: Base R coerces loosely, dplyr uses stricter coercion rules
    > based on vctrs, and data.table doesn't allow any coercion.

    > Another point to consider is the handling of attributes for the result.
    > data.table copies from the first non-NA input from left to right, while dplyr
    > delegates to vctrs again for merging the attributes gracefully. This matters
    > for example for factors, where data.table special-cases them to require the
    > same levels, wherease dplyr merges them. For a base R solution, it would make
    > sense to delegate the attribute handling to `c()` somehow, as that's conceptually
    > what should be happening; we're combining values from the `yes` and `no` objects.

    > I'm sure there are many other points to consider, but as I said this is what
    > comes to mind at first. Best of luck with the effort.

    > Kind regards,

    > Mikko

    [..........]

    >> -----Original Message-----
    >> From: R-devel r-devel-bounces at r-project.org On Behalf Of Duncan Murdoch
    >> Sent: Tuesday, July 8, 2025 3:06 PM
    >> To: Josiah Parry josiah.parry at gmail.com; Avraham Adler avraham.adler at gmail.com
    >> Cc: r-devel at r-project.org
    >> Subject: Re: [Rd] Time to revisit ifelse ?
    >> 
    >> Since you and Antoine are volunteering to do the work, why not start in
    >> the way I suggested? Write up a comparison of the known ifelse
    >> implementations, and either pick the best one, or choose the best parts
    >> of each. Put the result in a package containing nothing else, and
    >> invite comment from the wider community.
    >> 
    >> My only comment in advance is that the package should have no
    >> dependencies other than base packages, for two reasons:
    >> 
    >> 1. The hope is to have it adopted in base R, and for that it can't have
    >> any other dependencies.
    >> 
    >> 2. If it's never adopted by R Core, I might still want to use it, but I
    >> don't want to add extra dependencies for just one little function.
    >> 
    >> Duncan Murdoch

    [................]

Thank you, Mikko, Antoine, Duncan, etc
I'm trying to summarize the things I agree / or find important.
Note that we had ifelse() discussions in the past (on this
mailing list and/or possibly on R-help); I did get involved and
spent many hours on coding myself, with no convincing result
IIRC,  but I do vaguely remember I got very convinced we should
*not* plan to replace ifelse() but add a second version, say
if.else() (as  "if_else" is already taken by dplyr).

1) Antoine Fabri proposed that base R should get *another*
  version of ifelse() *in addition* to ifelse().   The issue
  hence is *NOT* replacing ifelse() by something incompatible.

2) Duncan Murdoch's points are *very* much to the point, most
   importantly:
   
    Propose (with discussion / RFC / ...) a function in a (single
    function) package which only depends on R's base package. 

 I'd add to that that you should probably use the GPL-2 licence
 or are willing to donate it with that licence to R and do say so;
 e.g., we cannot add MIT-licenced things to R.

3) Ben Bolker's offer to "host" such a function in his 'gtools'
   package (w/ 0-dependency) would also be acceptable to me,
   even though it is against DM's "2. If it's never adopted by R Core, .."

Best,
Martin

--
Martin Maechler
ETH Zurich  and  R Core team