Feature #20350: Return chilled string from Symbol#to_s

• Status: Open
• Assignee: -
• Target version: -

[Dan0042 (Daniel DeLorme)]

During Ruby 2.7 development there was an attempt to return a frozen string from Symbol#to_s (#16150#note-22)

This had to be rolled back due to incompatibility, but now we have chilled strings (#20205)

Symbol#to_s can safely return a chilled string, giving developers time to fix warnings before switching to a frozen string.

[byroot (Jean Boussier)]

To provide some data-point. Our monolith runs with https://github.com/Shopify/symbol-fstring, which makes the return value of Symbol#to_s frozen without any problem. I think we had to fix a single gem to make it work a few years back, that's it.

That said, while we run a lot of code (700+ transitive gems), that's not necessarily fully representative of the code out there.

Also one major difference with chilled string literals is that you can't just set RUBYOPT="--disable-frozen-string-literal" to continue running older code. I guess you could monkey patch String#to_s though, and provide that as a gem to make it easy to provide backward compatibility.

All this to say that as a Ruby user, I'd like this change to happen, because Symbol#to_s is a major source of allocation, and Ruby performance is very GC dependent.

But as a Ruby committer I think we need to be careful about the rate of deprecation and backward compatibility breaks. On one hand it would sound logical to ship such change at the same time as the change for frozen string literals, but on the other hand it makes the jump bigger and may alienate some users. So it's a fine line to walk.

So I'm positive, but with some reservations.

[Eregon (Benoit Daloze)]

+1, I was thinking the same yesterday, that we could revisit Symbol#to_s and have a proper deprecation before changing it due to chilled strings landing.

[Eregon (Benoit Daloze)]

I added this to the dev meeting: https://bugs.ruby-lang.org/issues/20336#note-8

[ko1 (Koichi Sasada)]

Symbol#name returns frozen string.
It is not enough?

[byroot (Jean Boussier)]

Symbol#name returns frozen string. It is not enough?

It doesn't work well with duck typing. It's common to have a method that call to_s on the argument, and the argument can be a Symbol or another type:

def some_method(arg)
  arg = arg.to_s
  # do things
end

We can debate the merits of such design of course, but from my experience it's extremely common. Years after Symbol#name introduction, Symbol#to_s remains visible on production profiles, and running https://github.com/Shopify/symbol-fstring/ still provide a small, but noticeable improvement.

[matz (Yukihiro Matsumoto)]

I am in favor of experimenting. I am in favor of moving to immutable strings in the future if there are no significant incompatibility issues.

Matz.