Feature #16670
closedReverse order of `expression` in `pattern` for 1-line pattern matching while it's still experimental
Description
Currently the 1-line syntax for pattern matching is:
# Usage: <expression> in <pattern>
expression = {
pattern: "Example"
}
expression in {pattern: something}
# something => "Example"
Is it technically possible, and desirable to switch the order of this syntax to:
# Usage: <pattern> in <expression>
expression = {
pattern: "Example"
}
{pattern: something} in expression
# something => "Example"
?
Here are my reasons:
- It is more intuitive in English -- we are "finding a pattern in something". Finding "something in a pattern" doesn't seem to make sense.
- Assignment is happening, and this keeps assignment on the left side of the operator which feels more natural.
- It matches existing behavior with the workings of the case statement:
Understanding that a case
block evaluates each when
expression using when_expression === case_expression
makes more consistency with when_pattern in case_pattern
using the new operator.
case something
when /pattern/
end
# is equivalent to
/pattern/ === something
# This creates more parity with
case something
in {pattern: x}
# would be equivalent to
{pattern: x} in something
Please see the following discussion on Reddit: https://www.reddit.com/r/ruby/comments/favshb/27s_pattern_matching_official_docs_recently_merged/fj2c7ng/
Updated by shevegen (Robert A. Heiler) over 4 years ago
I do not think reddit discussions are that useful, largely because it is difficult to find
what is really going on (content-to-noise ratio), and the article title was actually
"2.7's pattern matching official docs (recently merged)". zverok is active here on the
bugtracker, though, so I suppose he can comment on this issue if he has time and is
motivated. :)
If I understood your comment, then you refer mostly to the " in " part, and not regular
pattern-matching via case/in, yes?
If this is the case, and if I do not remember incorrectly, then I think the one-line
" in " was suggested by mame, not by the original author who suggested pattern matching.
(I mention this just so there is no confusion, because we should be clear with this -
ruby users may be confused about which syntax is valid, and which one is not, even
more so when it would suddenly change.)
Personally I have no specific opinion either way, largely because I am sticking to
what is very simple for me to understand; and pattern matching, while interesting,
is way over my head. :D But specifically, it may be best if mame and zverok could
comment on the proposal, if possible, to compare trade-offs, in particular upon
this:
expression in {pattern: something}
versus
{pattern: something} in expression.
Updated by zverok (Victor Shepelev) over 4 years ago
In favor of the proposal, standalone match looks this way in other languages:
# Rust:
let (x, y, z) = (1, 2, 3);
# Erlang:
{X, Y} = {1, 2}.
# Elixir:
{a, b, c} = {:hello, "world", 42}
# F#:
let {First=first} = alice
# Swift (?):
case let Media.movie(title, _, _) = m
(Probably there could be more examples, though I am not super-familiar with many languages, and for what I can judge from online docs, lot of them -- like Haskell or OCaml -- seem not to have standalone let
expression, just match
with several alternative patterns)
Updated by decuplet (Nikita Shilnikov) over 4 years ago
For the record, both Haskell and OCaml have let expressions. And yes, both have patterns on the left side, just like in F# or Rust.
Updated by zverok (Victor Shepelev) over 4 years ago
....and also, one more example, now just in Ruby:
# that's what we always had:
a, b = [1, 2, 3]
# could be pretty complicated:
a, (b, c), *d = [1, [2, 3], 4, 5]
# ...but now let's imagine we need to unpack a simple Hash...
# ...that's how it will look:
{a: 1, b: 2, c: 3} in {a:, **rest}
# ...while "intutively" I always want it look as close array examples above as possible:
{a:, **rest} in {a: 1, b: 2, c: 3}
I believe it should've been discussed on "standalone in
" introduction ticket (and probably it was found that the "natural" order will be impossible to parse?), but I don't see detailed discussion here: #15865
Updated by nobu (Nobuyoshi Nakada) over 4 years ago
The pattern syntax looks very close to hash literals, but its semantics is quite different.
So I think it is very hard to distinguish them without a preceding mark (in
keyword for now).
Updated by decuplet (Nikita Shilnikov) over 4 years ago
I think it's important to realize that it's the goal of pattern matching to be similar to data constructors. It proved itself to be a good solution in the long run, this is why languages mentioned above chose this approach. Of course, I'm not taking into consideration implementation difficulties here.
Updated by Dan0042 (Daniel DeLorme) over 4 years ago
I think the in
syntax and order feels natural when you're actually matching a pattern. It's not AI-like pattern recognition like "find a pattern in this" but rather pattern correspondence like "is this in the range expressed by this pattern".
But when using it only as destructuring assigment it does feel that something is off. Maybe it's because expr in var
has the opposite order of for var in expr
.
So rather than reversing the order, I'd like to tentatively propose ~|>
as a more natural-feeling alias for rightward destructuring assignment. Full proposal at #16794
Updated by ttilberg (Tim Tilberg) about 4 years ago
There were notes in the recent dev meeting that affect this issue: https://github.com/ruby/dev-meeting-log/blob/master/DevelopersMeeting20201026Japan.md#feature-17260-promote-pattern-matching-to-official-feature-ktsj
=>
is a new feature to support right hand assignment to allow things like do_stuff.then{|things| make_things things } => hotdogs
.
1-line pattern matching is currently "expression" in pattern
. Being very similar to right-hand assignment, it seems that =>
will now act as 1-line pattern matching, removing expr in pattern
. I find this attractive, as it reduces the number of concepts being added, and makes the 1-line matching more intuitive. It removes the interpretation of the word in
, and shares the existing notions of rescue HotDogError => e and {this => that}.
Given this new direction, this issue can be closed.
Discussion:
Updated by jeremyevans0 (Jeremy Evans) about 4 years ago
- Status changed from Open to Closed