Feature #11999: MatchData#to_h to get a Hash from named captures

• Status: Closed
• Assignee: sorah (Sorah Fukumori)
• Target version: -

[sorah (Sorah Fukumori)]

class MatchData
  def to_h
    self.names.map { |n| [n, self[n]] }.to_h
  end
end

p '12'.match(/(?<a>.)(?<b>.)(?<c>.)?/).to_h #=> {"a"=>"1", "b"=>"2", "c"=>nil}

Sometimes I want to get a Hash from named capture, but currently I have to use #names + #captures. How about adding MatchData#to_h for convenience way?

[sorah (Sorah Fukumori)]

• File 11999.diff 11999.diff added

[sorah (Sorah Fukumori)]

Consideration is behavior for multiple captures with same group name:

/(?<a>.)(?<a>.)/

MatchData#[] returns the last one and my attached patch follows that behavior.

[danielpclark (Daniel P. Clark)]

I agree. Please add this feature. I have also looked to do the same thing.

[matz (Yukihiro Matsumoto)]

I don't think to_h is appropriate, because MatchData is not always able to convert to Hash/Map.
Is there any name candidate?

Matz.

[sorah (Sorah Fukumori)]

is not always able to convert to Hash/Map.

Ah -- agreed. How about MatchData#named_captures?

I can't think this name is the best, suggestions welcome.

[phluid61 (Matthew Kerwin)]

Shota Fukumori wrote:

is not always able to convert to Hash/Map.

Ah -- agreed. How about MatchData#named_captures?

I can't think this name is the best, suggestions welcome.

I think #named_captures is the best name, since that is precisely what it returns (i.e. it never includes numbered captures.)

[duerst (Martin Dürst)]

Matthew Kerwin wrote:

I think #named_captures is the best name, since that is precisely what it returns (i.e. it never includes numbered captures.)

Could it make sense to include numbered captures in the hash, too? Just thinking aloud.

[phluid61 (Matthew Kerwin)]

Martin Dürst wrote:

Matthew Kerwin wrote:

I think #named_captures is the best name, since that is precisely what it returns (i.e. it never includes numbered captures.)

Could it make sense to include numbered captures in the hash, too? Just thinking aloud.

I thought so myself, but the regular expression engine currently does numbered captures only if there are no named captures.

Note: A regexp can't use named backreferences and numbered backreferences simultaneously.

-- http://ruby-doc.org/core-2.1.1/Regexp.html#class-Regexp-label-Capturing

I guess this is spec.

[Hanmac (Hans Mackowiak)]

also interesting if you have a with | combined regexp where both of them does have a named capture:

reg = /(?<a>b)|(?<a>x)/ # => /(?<a>b)|(?<a>x)/ 
reg.match("abc") # => #<MatchData "b" a:"b" a:nil>
reg.match("abc").captures #=> ["b", nil]
reg.match("abc")[:a] # => "b" 
reg.match("xyz") # => #<MatchData "x" a:nil a:"x"> 
reg.match("xyz").captures #=> [nil, "x"]
reg.match("xyz")[:a] # => "x"

(also notice that in the inspect of MatchData the capture :a is shown twice.)
such things does need to be remembered when creating a new function for MatchData

[sorah (Sorah Fukumori)]

Looks like it's the same problem as I noted here https://bugs.ruby-lang.org/issues/11999#note-2

[Hanmac (Hans Mackowiak)]

@Shota: i do need to test your patch, but my case is a little bit different than yours.
because it can be nil, it seems to pick the first non-nil value in my case. (or is it the last non-nil?)

specially with your patch:

reg = /(?<a>b)|(?<a>x)/ # => /(?<a>b)|(?<a>x)/ 
reg.match("abc") # => #<MatchData "b" a:"b" a:nil>
reg.match("abc").to_h #=> {"a" => "b"} or {"a" => nil}

[sorah (Sorah Fukumori)]

that makes sense for me.

[naruse (Yui NARUSE)]

Yukihiro Matsumoto wrote:

I don't think to_h is appropriate, because MatchData is not always able to convert to Hash/Map.
Is there any name candidate?

I feel it can always convert to Hash because even if it doesn't use named captures, the numbering is 1-origin.

irb(main):001:0> /(a)(b)(c)/.match("abc")
=> #<MatchData "abc" 1:"a" 2:"b" 3:"c">
irb(main):002:0> /(a)(b)(c)/.match("abc").to_h
=> {1=>"a", 2=>"b", 3=>"c"}

[phluid61 (Matthew Kerwin)]

Yui NARUSE wrote:

Yukihiro Matsumoto wrote:

I don't think to_h is appropriate, because MatchData is not always able to convert to Hash/Map.
Is there any name candidate?

I feel it can always convert to Hash because even if it doesn't use named captures, the numbering is 1-origin.

irb(main):001:0> /(a)(b)(c)/.match("abc")
=> #<MatchData "abc" 1:"a" 2:"b" 3:"c">
irb(main):002:0> /(a)(b)(c)/.match("abc").to_h
=> {1=>"a", 2=>"b", 3=>"c"}

I did some experimenting of my own to this end, and came up with this: https://github.com/phluid61/mug/blob/master/lib/mug/matchdata/hash.rb

The only real weirdness arises from the fact that positional captures don't happen at all if there's a named capture group in the Regexp; but given the resulting mutual exclusivity, the code itself becomes pretty straight-forward.

[sorah (Sorah Fukumori)]

• Assignee set to sorah (Sorah Fukumori)

Discussed at https://bugs.ruby-lang.org/projects/ruby/wiki/DevelopersMeeting20160216Japan with Matz and several committers:

Log: https://docs.google.com/document/d/1rj7ODOCSfcsQeBd6-p-NiVwqxDUg05G66LwDOkKOGTw/pub

• #to_h is inappropriate name while non-named capture exists:

  1. Return Hash with integer keys? ( {0 => "a", 1 => "b"} )
     • there's no use case for this behavior.
     • rejected
  2. #named_captures (accepted)
     • matz said acceptable

• Behavior when there are multiple named captures with same name

  • Return last matched value

    /(?<a>b)|(?<a>x)/.match("abc").to_h #=> {"a" => "b"}
    

• Behavior when named captures didn’t match anything

  • Return nil as value

• Behavior when no named captures

  • #captures returns [] when a regexp has no capture, so #named_captures returns {} when a regexp has no named capture

---

@matz (Yukihiro Matsumoto) Could you confirm this ↑ and say accept here please?

[sorah (Sorah Fukumori)]

• File 11999-2.diff 11999-2.diff added

Updated patch (11999-2.diff).

[matz (Yukihiro Matsumoto)]

Accepted.

Matz.

[sorah (Sorah Fukumori)]

• Status changed from Open to Closed

Applied in changeset r53863.

---

• re.c: Add MatchData#named_captures
  [Feature #11999] [ruby-core:72897]

• test/ruby/test_regexp.rb(test_match_data_named_captures): Test for above.

• NEWS: News about MatchData#named_captures.

[headius (Charles Nutter)]

Shouldn't this produce Symbol keys?