“parser did not give circumfix an EXPR” after “Re-implement colon list processing” commit

[AlexDaniel]

• MessagePack – Fail, Bisected: 9135914

Error message during tests:

===SORRY!===
internal problem: parser did not give circumfix an EXPR
t/00-unpack.t .. 

[AlexDaniel]

Ping @skids

[0racle]

The problem seems to be occurring in the return statement of method !unpack inside lib/MessagePack/Unpacker.pm6.

There is a given/when wrapped in parens and used as a return value. Inside the given when is a check for hash key :exists. Something in there is conufising the parser.

Abbreviated version of what it looks like

method !unpack() {
    my $type = self!unpack-uint8;

    return (given $type {
        ...
        when %unpack-for-type{$type} :exists {
            %unpack-for-type{$type}(self);
        }   
        default {
            fail sprintf("Unknown type 0x%02x", $type);
        }   
    }); 
} 

As given returns a value, and it's the last thing in this method, the return is a little unidiomatic. In fact $type is never referred to implicitly inside the given block, so the given is redundant.

In any case, I change return (given $type { ... }); to just given $type { ... }, (or remove the given all togther and just end the method with a string of when clauses and a default clause) the issue goes away and all MessagePack tests pass.

Also, if I leave the return statement as is, but change the condition from when %unpack-for-type{$type}:exists to when (%unpack-for-type{$type}:exists), the issue also goes away and all MessagePack tests pass.

At the very least, a pull-request can be made to fix MessagePack, but hopefully this info might help identify what's confusing the parser.

[0racle]

Golfed...

(when %()<>:exists { ... })

These are ok

when %()<>:exists { ... }
(when (%()<>:exists) { ... })

Also, this issue is not specific to when... if, for or given in place of when still trigger.

[AlexDaniel]

I tried to debug this but I'm a bit confused. So $<semilist><statement>[$semi] does exist, but <EXPR> is not quite there. If I .dump $<semilist><statement>[$semi], I get this:

- statement_control: when %()<>:exists { ... }
  - sym: when
  - xblock: %()<>:exists { ... }
    - pblock: { ... }
      - blockoid: { ... }
        - statementlist:  ... 
          - statement: 1 matches
            - EXPR: ... 
              - sym: ...
              - args:  
                - arglist: 
    - EXPR: :exists 
      - 0: <>
        - 0: %()
          - variable: %()
            - contextualizer: %()
              - coercee: 
                - statement:  isa NQPArray
              - sigil: %
              - sequence: 
                - statement:  isa NQPArray
        - OPER: <>
          - nibble: 
          - O: 
        - postfix_prefix_meta_operator:  isa NQPArray
        - postcircumfix: <>
          - nibble: 
          - O: 
      - OPER: 
        - O: 
      - colonpair: :exists
        - identifier: exists
      - fake_infix: 
        - O: 

So the expected EXPR is a bit deeper, but why is it looking for it at all? Could it be that $*FAKE_INFIX_FOUND is not reset correctly?

[AlexDaniel]

Ping @skids

[lizmat]

Fixed for now with c227693 . It spectests clean and makes MessagePack install again with zef. There is probably a better fix possible, but this should unblock the 2019.06 release for now.

[AlexDaniel]

Seems like it's not that simple… see https://colabti.org/irclogger/irclogger_log/perl6-dev?date=2019-06-29#l168

The code to trigger the issue gets weirder, but it's still somewhat realistic.

[lizmat]

More complete solution pushed with a26e95b