Use char_is_identifier_start() to check for valid method name and fix serialization

[maxprokopiev]

I've noticed an interesting case where snippets like foo.🌊 = 1 are not accepted and failing with:

[#<Prism::ParseError @message="unexpected write target" @location=#<Prism::Location @start_offset=10 @length=1 start_line=1>>]

The reason for that is because the current check is only for _ and any alphanumeric char, so it makes sense to change that to use char_is_identifier_start() which will check for any valid identifier (while also making the condition cleaner ✨).

After fixing that I've noticed that test cases started to fail with:

...
prism/test/prism/test_helper.rb:60:in `assert_equal_nodes'
prism/test/prism/parse_test.rb:230:in `block (3 levels) in <class:ParseTest>'
prism/test/prism/parse_test.rb:225:in `each'
prism/test/prism/parse_test.rb:225:in `block (2 levels) in <class:ParseTest>'
<:🌊=>(UTF-8) expected but was
<:"\xF0\x9F\x8C\x8A=">(ASCII-8BIT)

diff:
? :🌊                =
?  "\xF0\x9F\x8C\x8A "
?  ? +

? Encoding: UTF  -8
?           ASCII  BIT
?           ???  +++

That seems to be an issue with not forcing encoding during the deserialization, which is also fixed in 9323243

[kddnewton]

Yeah this looks good, thanks!

[kddnewton]

It looks like there's a failure on truffleruby because of Symbol#inspect, where on TruffleRuby they're making it :"🌊" whereas on CRuby it's :🌊.

To fix this, can you move this fixture into its own file and then skip it on TruffleRuby in parse_test.rb? We have something like this already there because of this exact problem (which I forgot about...).

[kddnewton]

And then we just need to regenerate the snapshots by rerunning the tests

[maxprokopiev]

was running the tests with truffleruby locally 😅 pushed the snapshots in 459a9f5

[eregon]

I filed oracle/truffleruby#3407 to fix this in TruffleRuby