[Feature] Implement error recovery for the Kipper compiler

[Luna-Klatzer]

Is there an existing proposal for this?

• I have searched the existing issues

This feature does not exist in the latest version

• I am using the latest version

Proposal

Implement error recovery for the Kipper compiler, which allows it to attempt recovery from an error and continue the semantic analysis if possible. This will make it be able to throw multiple compile errors and provide more insight into the errors of a program.

Example

• Code:

  var x: typ = 5 + "str";

• Errors:

  [ERROR]: Traceback:
    File 'anonymous-script', line 1, col 7:
      var x: typ = 5 + "5";
             ^^^ 
  TypeError: Unknown type 'typ'.
  [ERROR]: Traceback:
    File 'anonymous-script', line 1, col 13:
      var x: typ = 5 + "5";
                   ^^^^^^^ 
  InvalidArithmeticOperationError: Invalid arithmetic operation between types 'num' and 'str'.

Exact behaviour / changes you want

• Add compile config field recover, which is per default true.
• Add new CLI flags -r/--recover and --no-recover.
• Implement the recover algorithm, which follows the following procedure (Lexing and Parsing does not apply to these rules):
  • Run semantic analysis for an AST node.
  • If the semantic analysis fails due to an error, try to determine whether the semantic data is present:
    • If not, then recovering for this AST node and its primary parent (statement or definition) is impossible, and as such, the whole statement/definition is discarded and the analysis should continue with the next statement or definition.
    • If it is present, then recover from this error and continue the analysis for the statement/definition.
  • Afterwards, the error should be added to a list of errors and then thrown using a single error at the end of the semantic analysis. (KipperCompileResult should contain an errors field, which provides a list of all errors).
  • Continue with the next expression, statement or definition.

Note

The recover algorithm can result in semantic data being missing. For example, if a variable can not be created/registered due to an error, recovering from that error will mean that any reference to it afterwards will cause more errors as the compiler does not know about the variable. This is bad behaviour, as it results in errors that are wrong and provide invalid insight into what went wrong. Such behaviour should therefore be avoided at all costs and the semantic analysis should be structured as robust as possible to make sure semantic data that is valid is registered, and as such "invalid errors" are avoided.

[Luna-Klatzer]

Basic error recovery implementation implemented with 79f94a4. Development and tests will continue on the branch, and then later be merged into dev for the next release v0.10.0.