The current design of the parser produces unusable error messages and is awkward to extend.
A redesign is called for. It needs to support (or be easy to add) the following features:
Continue parsing a file after the first error
When multiple parse branches were tried, give context as to why they all failed, and not just why the most recent one failed.
Move more errors from runtime to parsing
The unit type of scalar constants should immediately produce an error, rather than waiting until they are executed.
New design overview
The following layers should be used to process input files.
Source String
Raw input from a file
Verify UTF8 encoding
Tokenizer
Convert to tokens
Comments
Literals
Operators
Statement keywords
Abstract Syntax Tree
Convert tokens into context sensitive objects
Struct definitions
Function definitions
Statements
This design should simplify the creation of higher level structures and the reporting of error messages related to them.
It should also make it much easier to write a formatter later.
Goals and reasoning
The current design of the parser produces unusable error messages and is awkward to extend. A redesign is called for. It needs to support (or be easy to add) the following features:
New design overview
The following layers should be used to process input files.
This design should simplify the creation of higher level structures and the reporting of error messages related to them. It should also make it much easier to write a formatter later.