Feature #14594
closedRethink yield_self's name
Added by zverok (Victor Shepelev) over 6 years ago. Updated over 5 years ago.
Description
I feel really uncomfortable raising the question again, but...
In several months since 2.5 release I've written a lot of code with yield_self
(using backports
gem with earlier versions of Ruby when necessary), and explained it several times to students, and colleagues (and in this blog post which have gained pretty decent attention).
I should say that I am still assured the name chosen is really not optimal. Reasons:
- it is just too long for such a basic operation;
- it does not say "what it does", but rather "how it is implemented"; it is like having
each_returning_block_result
instead ofmap
; -
self
is really misguiding and obscure in situations like this:
class MyClass
def some_method
@path.yield_self(&File.method(:read)).yield_self(&Parser.method(:new)) ...
end
end
Intuitively, word "self" inside instance method is read like it somehow related to current context's self
(e.g. instance of MyClass
), which it is absolutely not. In other words, "self" in caller's context has nothing to do with "self" implied by method's name.
After reconsidering a lot of options, my current proposal is: #then
.
Reasons:
- despite being a keyword,
something.then(something)
is not a conflicting Ruby syntax, and allowed by current Ruby; - it is short!
- it shows intention pretty well, and reads natural, in both cases: when receives block and when returns Enumerator:
File.read(filename).then(&JSON.method(:parse))
rand(10).then.detect(&:odd?)
In many languages, .then
or .and_then
is useful construct, meaning the same (calculate next value from the result of the previous operation), just in a narrower context of futures/promises. I believe that even when/if Ruby will have those as a language feature, that syntax will play well:
value.then(&:computation) # => value
promise.then(&:computation) # => promise
PS: For historical reasons, here is huge list of previous proposals I've gathered for this method name.
Updated by shevegen (Robert A. Heiler) over 6 years ago
I agree to your statement that yield_self is not a good, descriptive
name.
Though I have to admit that I never used yield_self so far. I can
not even say what it does, either. :)
I like yield and self. I don't like the name yield_self so I could
not use it so far. I use only a subset of ruby that makes sense
to me and I find appealing. Thankfully one can use ruby just fine
without having to use everything.
After reconsidering a lot of options, my current
proposal is: #then.
I dislike that as well. I also do not think that "then"
makes a lot of sense.
For example:
File.read(filename).then(&JSON.method(:parse))
This is regular method chaining. But it sounds like
you are using a conditional there.
"then" is already a keyword in ruby isn't it?
https://docs.ruby-lang.org/en/2.5.0/keywords_rdoc.html
I think it is confusing to have keywords and method names,
even more so when they do different things.
In many languages, .then or .and_then is useful construct, meaning
the same (calculate next value from the result of the previous
operation), just in a narrower context of futures/promises.
Realistically ruby should strive for intrinsic consistency first,
not for what features or anti-features other languages may or
may not have.
I believe that even when/if Ruby will have those as a
language feature, that syntax will play well:
value.then(&:computation) # => value
promise.then(&:computation) # => promise
I don't think it reads nicely really.
yield_self is not a good name but your proposal is also
not good, in my opinion. But it's just a personal opinion,
feel free to ignore it. At the end of the day you only have
to convince matz. :)
The name yield_self is however had indeed not a good name.
Finding good names is quite difficult. Single words are
also almost always better than combined names, even though
one is a bit limited with single words alone. There are
exceptions though. For example .each_with_index or
.each_index are good names, IMO.
Updated by cmoel (Christopher Moeller) over 6 years ago
I also agree yield_self
isn't the best name but I'm also skeptical of then
. Why not use map
, which we already have for collections? Generally speaking, the map
function isn't just about collections (though that's usually how it's used in Ruby). map
is more about putting an object in a context (a block in Ruby's case), modifying the object, and returning the modified object.
Another benefit to calling the method map
instead of yield_self
or then
is most Rubyists already understand map
and how it's applied to collections, so there probably wouldn't be a large learning curve for it. It can be summarized like this: "Object#map
is similar to applying map
to an array containing 1 element."
Though I'm not familiar with the details, Enumerable#map
and friends' implementation of map
could then be an override of Object#map
, yielding the block to each element in the collection, which would maintain how map
currently works on collections.
Can anyone think of any issues with using map
as a different/better name for yield_self
?
Updated by zverok (Victor Shepelev) over 6 years ago
Why not use
map
paragraphs
.map { something }
.reject { something }
.yield_self { |ps| external_post_processor(ps, **options) }
Updated by sawa (Tsuyoshi Sawada) over 6 years ago
map
is one of the worst names for this feature in my opinion. Generalizaing from cmoel's proposal to use map
for this feature, we expect:
[1].map{|e| e * 2} #=> [1, 1]
but that is not what we currently have, nor does cmoel seem to want that. Instead, we have, and cmoel seems to want to handle as an exceptional case, the following:
[1].map{|e| e * 2} #=> [2]
That is confusing. Why do you want to make something that has worked without problems to this day to become an exception?
Updated by sawa (Tsuyoshi Sawada) over 6 years ago
I disagree with the names proposed, but I agree that the current name is too long.
Here are yet other name candidates from me:
- deform
- expand
- reform
Edit. I just realized that reform has already been mentioned. Sorry.
Updated by bughit (bug hit) over 6 years ago
then
does seem better but you have to keep in mind you'd be barring this name from all present and future domain specific uses in libraries and apps
What about tap!
- since tap is already in use, this won't be stealing another word, and it seems to make sense, it's a "dangerous" tap that alters the flow.
Updated by jeromedalbert (Jerome Dalbert) over 6 years ago
I agree that yield_self
is not a great name. But then
does not "say what it does" enough in my opinion. I prefer itself
: this name has problems too, but it is more descriptive at least.
Updated by zverok (Victor Shepelev) over 6 years ago
I prefer
itself
itself
is already taken in block-less form.
It was my initial idea too (see links in the post for names discussion), but I am happy it was not accepted: iterator-returning block-less yield_self
turns out to be useful too.
Updated by cmoel (Christopher Moeller) over 6 years ago
bughit (bug hit) wrote:
What about
tap!
- since tap is already in use, this won't be stealing another word, and it seems to make sense, it's a "dangerous" tap that alters the flow.
I think tap!
is the best option currently being discussed:
- There's no need to create another word for this kind of generic functionality,
- Rubyists already know and use
tap
, and - Rubyists also know the
!
suffix modifies the callee.
Updated by americodls (Americo Duarte) over 6 years ago
My two cents:
This method acts like a pipe operator, it passes itself through the block and returns whatever the block returns.
The block execution maybe will change the value itself, maybe not. But yield_self
proposes chaining changes without mutation (tap
is recommended to make mutations since tap
returns the receiver itself instead of the block result).
So I think good candidates to rename or alias yield_self
:
pipe
pipe_to
chain
-
chain_with
.
Updated by BatuhanW (Batuhan Wilhelm) over 6 years ago
I think expand is cool, it is expanding method to another one, so it can be chained.
My alternative suggestion would be "pass" since it is passing itself. What do you think about it?
construct_url
.pass(&Faraday.method(:get)).body
.pass(&JSON.method(:parse))
.dig('object', 'id')
.pass { |id| id || '<undefined>' }
.pass { |id| "server:#{id}" }
Updated by sowieso (So Wieso) over 6 years ago
I just like to stress that having an easy and short name is more important than having a name that correctly describes the behaviour. tap
is also a horrible name that every ruby newbie has to look up first, still it is loved in the community. And that's especially due to the fact that it is short and feels lightweight. As for yield_self
, no discussion is needed whether this method is needed, but the naming is indeed the difficult part. So I support this discussion (as annoying as it is, how many years is it now?). Let's find a better name (probably an alias)!
I would go with (in this order):
-
tip
(the closest relative to tap, absolutely short, tip over suggests some movement/transfer, probably never used anywhere, tip-tap sounds like a cat walking, downside: not very descriptive at all) -
and
(like then, even shorter, focus on continuing, probably completely unused, downside: people might think of boolean comparisons, not very descriptive) -
then
(short, downside: people might think of promises, not very descriptive) -
pass
(short and somewhat descriptive, might be in use already somewhere) -
pipe
(short and more descriptive, but used in other contexts (threads, shell))
Some other ideas:
-
hand
(as in hand over, short, probably never used elsewhere, descriptive? (with some imagination)) -
chain
(probably in use too often) -
use
(use_here) -
pass_me
(somewhat longer, pretty clear, reads much nicer than yield_self, but I don't like the me either) -
yield
(already in use, so a no-go, without the self it's short, similar to pass, but too confusing I guess) -
relay
(probably in use too often)
I don't like tap!
, I think it's misleading, as I would guess the method receiver would get mutated. map
is way to confusing. Including self inside a method name is always somewhat confusing (self in which context?), so I would prefer not to do it.
So many ideas, there are definitely a lot I would be content with, we just need to decide.
Updated by rosenfeld (Rodrigo Rosenfeld Rosas) over 6 years ago
I actually liked the tip suggestion.
Updated by matz (Yukihiro Matsumoto) over 6 years ago
- Assignee set to matz (Yukihiro Matsumoto)
After a long consideration, I decided to pick then
as an alias to yield_self
.
As @zverok (Victor Shepelev) stated it describes intention, not behavior.
Because then
is a reserved word, it has some restriction, but I think it is acceptable.
Matz.
Updated by matz (Yukihiro Matsumoto) about 6 years ago
- Status changed from Open to Closed
Applied in changeset trunk|r63525.
object.c: Add a new alias then
to Kernel#yield_self
; [Feature #14594]
Updated by jrochkind (jonathan rochkind) about 6 years ago
I'm concerned that then
is used in Promises, and is already in use in many promise-implementing libraries, including concurrent-ruby.
The Promise libraries could change the name of their method to not conflict, but then it's not only backwards compatibility but a mis-match with Promises in other platforms, and the original JS spec that they are based on.
The Promise libraries can keep doing this, effectively overriding the Kernel method. Somewhat confusingly then
will mean something different when used on a Promise object, and Kernel#then won't be available on them. I guess since it's just an alias, this isn't a disaster, yield_self is still available, but if you're using Promises you've got to think about what kind of object you have and which alias you want, and that then
means something different on Promises.
Updated by matz (Yukihiro Matsumoto) about 6 years ago
@jrochkind (jonathan rochkind) It is introduced that a normal object can behave like promises.
So the name conflict is intentional.
If you really wanted a non-unwrapping method for promises, use yield_self
.
Matz.
Updated by avit (Andrew Vit) about 6 years ago
matz (Yukihiro Matsumoto) wrote:
If you really wanted a non-unwrapping method for promises, use
yield_self
.
If I understand what you mean by "unwrapping" here, the new method still doesn't call
yielded procs to make them composable: it's only an alias for yield_self
, right?
Is this still a possible consideration?
Thanks for this by the way, I very much prefer the new name!
Updated by matz (Yukihiro Matsumoto) about 6 years ago
Updated by avastor.developer (Avastor Avastor) about 6 years ago
a = 5
y = if a.then then 1 else 2 end
Hmm. What are you trying do with Ruby?
Updated by RichOrElse (Ritchie Buitre) almost 6 years ago
I suggest aliasing #yield_self
with #to_be. The to_
prefix suggest returning a value while the word "be" connotes both identity (#itself
) and it's new condition (&block
).
Not only is it shorter, when reading it repeatedly, I also don't find it annoying. It reads just like the lyrics to the song "Welcome to My Life" by Simple Plan.
construct_url.
to_be(&Faraday.method(:get)).body.
to_be(&JSON.method(:parse)).dig('object', 'id').
to_be { |id| id || '<undefined>' }.
to_be { |id| "server:#{id}" }
Updated by sowieso (So Wieso) almost 6 years ago
This looks to_be perfect, I really like this idea!
Updated by rosenfeld (Rodrigo Rosenfeld Rosas) almost 6 years ago
I liked to_be as well.
Updated by rosenfeld (Rodrigo Rosenfeld Rosas) almost 6 years ago
And I like pipe too, maybe even more.
Updated by theorygeek (Ryan Foster) almost 6 years ago
I think then
is going to be a bad name, because it has a special meaning in A+ Promises: https://promisesaplus.com/
This makes every object in Ruby a "thenable" object, and I think it will be bad for promise libraries.
Updated by sowieso (So Wieso) over 5 years ago
So we will have the questionable (as confusing for promise library users) then
in 2.6? There were a lot of concerns mentioned above. to_be
(or other suggestions) wouldn't have these problems. I would advice to re-think this one final time, as we can't change it anymore later…
Updated by mame (Yusuke Endoh) over 5 years ago
FYI: mruby 2.0 has been already released with Kernel#then
. I think we've gone beyond the point of no return.
Updated by WaKeMaTTa (Mohamed Ziata) over 5 years ago
I think the word should be hit
.
Example:
construct_url.
hit(&Faraday.method(:get)).body.
hit(&JSON.method(:parse)).dig('object', 'id').
hit { |id| id || '<undefined>' }.
hit { |id| "server:#{id}" }