Implement file parsing error handling

[eileencodes]

Implement file parsing error handling

This PR implements proper file parsing error handling. Previously
file_options would call pm_string_mapped_init which would print an
error from perror. However this wouldn't raise a proper Ruby error so
it was just a string output. I've done the following:

• Raise an error from rb_syserr_fail with the filepath in
  file_options.
• No longer return Qnil if file_options returns false (because now
  it will raise)
• Update file_options to return static void instead of static bool.
• Update file_options and profile_file to check the type so when
  passing nil we see a TypeError.
• Delete perror from pm_string_mapped_init
• Update FFI backend to raise appropriate errors when calling
  pm_string_mapped_init.
• Add tests for dump_file, lex_file, parse_file,
  parse_file_comments, parse_lex_file, and parse_file_success?,
  when a file doesn't exist and for nil.
• Updates the bin/parse script to no longer raise it's own
  ArgumentError now that we raise a proper error.

Fixes: #2207

[eileencodes]

Oops, forgot to run all the tests. Will update.

[kddnewton]

Great work!

[kddnewton]

Looks like the windows tests are failing because it's a different errno error (I think that's expected). Can you check for a superclass?

[kddnewton]

Ahh it looks like FFI is failing because it needs to be updated. You can check that with:

PRISM_FFI_BACKEND=1 bundle exec rake

[eileencodes]

I'm a little bit stuck. FFI tests are failing because we check LibRubyParser.pm_string_mapped_init and raise if it's false.

prism/lib/prism/ffi.rb

Line 163 in 837af48

raise unless LibRubyParser.pm_string_mapped_init(pointer, filepath)

I can't figure out how (or even if) I can get the errno here in order to raise a SystemCallError.new(filepath, ????) with the right error code. The errno is different for nil or doesn't exist, so I can't pass in the Errno::ENOENT constant.

[eileencodes]

After pairing with Kevin (thank you!) we've changed pm_string_mapped_init to return the error code instead of a boolean so we can raise an appropriate SystemCallError from FFI.

[eregon]

FWIW there is FFI.errno to get the errno.
That's fully reliable if the previous call was a call done with FFI (i.e. it's properly saved&restored around FFI calls and not just accessing the native errno).

[eregon]

It feels a bit strange that nil would raise EFAULT.
I would expect a coercion error, just like File.read does:

$ ruby -e 'File.read nil'
-e:1:in `read': no implicit conversion of nil into String (TypeError)

[eileencodes]

It's now raising a TypeError.

[kddnewton]

This one you should remove, because it's not implemented in the FFI. We'll trust that it'll work since it's not supposed to be exposed anyway. But thank you for being thorough!

[eileencodes]

Ok I think this is ready now! 😄

[kddnewton]

Nice job!