Even though a Symbol is not technically an honest-to-goodness String, from the standpoint of simple practicality it would help to have Symbol#to_str defined.
There are times when we want an argument to accept a String or a Symbol, but don't really want it to accept any type of object under the sun that responds to #to_s --which is just about anything. This is especially the case when writing DSLs. Having Symbol#to_str is the nice solution to this.
Defining Symbol#to_str may be an exception to the rule, but it's one worth making.
#1 [ruby-core:52237] Updated by Nathan Zook over 3 years ago
Bad idea. to_str should only be defined on things that really are Strings, and Symbol are most definitely not Strings.
I agree that Symbol is unusually close to String. If, for your needs, you were to define to_st on String & on Symbol, you could have the utility you desire.
#2 [ruby-core:52241] Updated by Thomas Sawyer over 3 years ago
If, for your needs, you were to define to_st on String & on Symbol, you could have the utility you desire.
Yes, I thought about that. But concluded it was most likely unnecessary complexity when #to_str would work fine.
You say "Bad idea". But show me why it is bad idea other then "them's the rules". I tried to think of a problem case, and the only one I can think of is using
foo.respond_to?(:to_str) to identify Stringy things and very specifically not meaning to include Symbols. It's possible, but it's a fairly narrow proposition. Not the least reason being that one should never use
respond_to? if one does not need to b/c it is a fragile approach. But more significantly, what is more likely to be used? This narrow usecase or Symbol#to_str? Clearly the later by far. And the former is easily solved with
&& !Symbol === foo.
#5 [ruby-core:52286] Updated by Thomas Sawyer over 3 years ago
You cannot gsub, enumerate characters in or alter encoding of a Symbol, so it is not a string representation.
That the official spec on the definition of a Stringy-thing? That's the "problem" with #to_str, #to_ary, etc. isn't it? There is no absolute interface that dictates their proper use. As long the method returns the expected type then its purely a question of practicality. And I submit that Symobol#to_str is about a practical as it gets.
And let me put it another way. If you inherited some code that relied on an object responding to #to_str to ensure it also responded to #gsub, #map and #force_encoding (which is the crux of your "definition"), what would you think? Yes, you'd have seriously fragile code on your hands and you'd be a'fixing it.
I think you rejected this issue far too prematurely. Do you guys even know the purpose of dialog?
#6 [ruby-core:52287] Updated by Thomas Sawyer over 3 years ago
@charliesome Actually, that's exactly what my proposal attempts to address. You don't always have a choice in what type of object you receive, but you want it to become a string. Consider a DSL like Rake's. One could use:
task :foo do ...
task 'foo' do ...
Either one is acceptable, and I think it would be overreaching to make people not be able to use a symbol here.
On the other hand do we want any object to be acceptable? B/c just about every object responds to #to_s. To avoid this, you would end up with something like: (WARNING! Fugly code ahead.)
name = (Symbol === name ? name.to_s : name.to_str)
There has to be a clearer solution than that.
P.S. Just for fun of it I tried this on rake and discovered the Jim decided not to care what get's passed to task. Try this in your Rakefile:
desc "OMG!" task Object.new do puts "OMG is right!" end
A Duck-typing true beleiver!!! Yea, looks like a bug to me. If the user really needs it they can call #to_s.
#7 [ruby-core:52294] Updated by Eric Hodel over 3 years ago
The purpose of to_str, to_int, to_ary and to_sym are to convert string, integer, array and symbol representations to objects of that class.
The rope data structure (which supports insertion, deletion and random access) can be used to implement a representation of a ruby string so it would be a good candidate for to_str.
A linked-list implementation could be a good candidate for to_ary
A roman numeral implementation that does not descend from Numeric represents an integer and would be a good candidate for to_int
A string can be used as an identifier (as in rake) so it has to_sym.
A symbol, being an identifier alone is not anything like a String.