Deckwreck is a library that helps with the difficult parts of writing a recursive descent parser. At present it provides
- fast maps from keyword strings to integers; and
- a precedence-based expression parser with O(n) performance.
The keyword map code is stable and well tested. The expression parsing code is more experimental at present.
Many parsers need to check identifiers against a small list of reserved keywords. The Go parser, for example, uses a map to perform this check.
Deckwreck provides fast trie-based maps from strings to integers. These are around 1.5-2 times faster than a hash map for typical use cases.
Parsing expressions with operators of different precedence levels is one of the more challenging aspects of writing a recursive descent parser. Deckwreck's expression parser uses a non-recursive iterative algorithm. The expression parser constructs a shadow parse tree from a reusable pool of nodes. The shadow parse tree is then walked to construct the real parse tree. Parse tree construction is controlled by user-supplied interface implementations.
Notable features of Deckwreck's expression parser:
- O(n) in the number of expressions and operators, as confirmed by benchmarks.
- Fast for small expressions (the most common case in typical source code).
- Works with existing parse tree data structures via a generic interface.
- Handles optionally binary/unary operators (e.g.
-in C and Javascript). - Handles operators that can either be prefix or postfix (e.g.
++and--in C and Javascript). - Handles expression combination via juxtaposition (as in e.g. ML, Haskell, Elm). Juxtaposition can be given a precedence level just like any other operator.
- Handles parenthetical operators (e.g. the
[...]indexation operator in C and Javascript). - Handles matching of parentheses of different types (e.g.
[]and()). - Operators can be defined as either left or right associative.
- Generates a partial parse tree for erroneous inputs.