Chocking on interpolation with interpolation

[akimd]

Hi Markus!

Weird things happen when there are interpolations within interpolations.

$ unparser -ve '"#{"#{true ? "1
" : "2"}
"}"'
(string)
Original-Source:
"#{"#{true ? "1
" : "2"}
"}"
Generated-Source:
"#{<<-HEREDOC}"
Original-Node:
(dstr
  (begin
    (dstr
      (begin
        (if
          (true)
          (str "1\n")
          (str "2")))
      (str "\n"))))
Generated-Node:
#<Parser::SyntaxError: unexpected token tLSHFT>
/opt/local/lib/ruby3.0/gems/3.0.0/gems/parser-3.0.1.0/lib/parser/diagnostic/engine.rb:72:in `process'
/opt/local/lib/ruby3.0/gems/3.0.0/gems/parser-3.0.1.0/lib/parser/base.rb:286:in `on_error'
(eval):3:in `_racc_do_parse_c'
(eval):3:in `do_parse'
/opt/local/lib/ruby3.0/gems/3.0.0/gems/parser-3.0.1.0/lib/parser/base.rb:190:in `parse'
/opt/local/lib/ruby3.0/gems/3.0.0/gems/unparser-0.6.0/lib/unparser.rb:116:in `block in parse_either'
/opt/local/lib/ruby3.0/gems/3.0.0/gems/unparser-0.6.0/lib/unparser/either.rb:34:in `wrap_error'
/opt/local/lib/ruby3.0/gems/3.0.0/gems/unparser-0.6.0/lib/unparser.rb:115:in `parse_either'
/opt/local/lib/ruby3.0/gems/3.0.0/gems/unparser-0.6.0/lib/unparser/either.rb:115:in `bind'
/opt/local/lib/ruby3.0/gems/3.0.0/gems/unparser-0.6.0/lib/unparser/validation.rb:63:in `from_string'
/opt/local/lib/ruby3.0/gems/3.0.0/gems/unparser-0.6.0/lib/unparser/cli.rb:40:in `validation'
/opt/local/lib/ruby3.0/gems/3.0.0/gems/unparser-0.6.0/lib/unparser/cli.rb:143:in `public_send'
/opt/local/lib/ruby3.0/gems/3.0.0/gems/unparser-0.6.0/lib/unparser/cli.rb:143:in `process_target'
/opt/local/lib/ruby3.0/gems/3.0.0/gems/unparser-0.6.0/lib/unparser/cli.rb:133:in `block in exit_status'
/opt/local/lib/ruby3.0/gems/3.0.0/gems/unparser-0.6.0/lib/unparser/cli.rb:132:in `each'
/opt/local/lib/ruby3.0/gems/3.0.0/gems/unparser-0.6.0/lib/unparser/cli.rb:132:in `exit_status'
/opt/local/lib/ruby3.0/gems/3.0.0/gems/unparser-0.6.0/lib/unparser/cli.rb:64:in `run'
/opt/local/lib/ruby3.0/gems/3.0.0/gems/unparser-0.6.0/bin/unparser:10:in `<top (required)>'
/opt/local/bin/unparser:23:in `load'
/opt/local/bin/unparser:23:in `<main>'
Error: (string)

I agree no human being would write such code. But our generator does :(

Cheers!

[mbj]

2 issues play in here:

First:

There are dynamic strings that produce AST that can only be the result of parsing heredocs. And the signature for these ASTs of "must have been a heredoc and can only be produced from a heredoc" are almost random.

Unparser has some heuristics to trigger heredoc emits, and sometimes (which happened here): We end up with triggering heredocs where we should not have. But somtimes emitting more heredocs makes the emitter code easier, while still producing the same AST on parsing.

Second:

Heredocs need 'parallel' dispatch, as they have a very dislocal effect on the token stream. Its likely that one of the parrallel all sides is missing hence not getting you the body of the heredoc, only the header.

Thanks for reporting: Right now, I'm super busy commercially, assume this one takes a very long time to fix.

[akimd]

Markus,
In trying to find something to reproduce our failure, I may well have written something actually more complex than our original use case. And so far, we can't reproduce the original issue, so we might just as well have used some older version of unparser somewhere.
In other words: we are not blocked by this issue, and AFAIC you are welcome to let it rust.
Cheers!

[mbj]

@akimd its my end goal that unparser can round trip ALL ast that parser emits. Which may mean I'll have to ask the parser team to simplfy the ast.

I'm close to ask them to give me a dedicated node for heredocs in the original source, this would make my part so much easier.

[mbj]

@akimd I recently spend a bit more time to think about the problem. Overall I've to conclude that the current parser AST, on any form of dynamic strings is currently lossy.

Not lossy for execution semantics, but lossy in terms of round trip semantics. Meaning you cloud certainly implement a ruby on top of that AST that executes correctly, but you cannot implement an unparser that produces source that produces the same AST, as critical information is lost/noisy.

I've got a local AST builder that subclasses the offiical parser builder producing an AST that does not suffer this compression losses.

Now the question is: Whats your use case? Do you round trip hand written ruby, (That may include heredocs / dynamic strings)? Or do you generate an AST from other sources to than unparse with unparser?

[akimd]

Hey Markus,
Oh, you have a new face :)

Unparser is the final part of a parse|transform|unparse pipe. We don't care about exact syntactic round tripping, we just care about the (execution) semantics.

Cheers!

[mbj]

Oh, you have a new face :)

ageless version ;)

Unparser is the final part of a parse|transform|unparse pipe. We don't care about exact syntactic round tripping, we just care about the (execution) semantic

You are using Unparser.parse, if not would you consider using that one instead of Parser::CurrencyRuby.parse ?

[akimd]

Markus,
I have been told that since we have something that works (and works around that particular issue), we won't change it again.
So I guess we'll stick to Parser's parser.

[mbj]

@akimd Unparser.parse "is" the parsers parser, but it uses the modernized AST format. Thats all I want to know.

[akimd]

Markus,
Yes, I know, that was very clear. Yet this stuff is in production, and the guy in charge of that does not want to change it now.
Maybe in the future this will be revisited, but not now. Sorry!

[mbj]

So are you using the modernized AST Format or not?

[akimd]

Hi Markus,
No, we are still using the "native" AST from parser itself.

[mbj]

kk, so as a recomemndation: The parsers default AST format is lossy on semantics, and unparser only supports the modern one for that reason.

Also all other major users of parser opt into the modern format for that reason, and it can very well be that you have semantic issues as you process the default format.

[mbj]

Also a fix for this issue will probably only ever land in the modern AST format.

[akimd]

I perfectly understand all these points, but the decision is not mine.  I'm just a proxy here.