02Credits Blog

Day5 - Pando Parser

Parsing the Pando todo file format

2019-02-11

Output

Lets get to coding. Today I'd like to get started with the todo format parser. In contrast to Okeydokey where the file format is simple enough to parse and manipulate, the format for Pando is information dense. Since we will be compiling it to the DOT format, it would be a good idea to use a more principled approach. To that end I have decided to use nom which is a parser combinator library.

I'm no expert in parsers, but my understanding is that parser combinators are a different branch in the taxonomy of parser libraries from Lex and Yacc. Where Lex and Yacc describe the desired format in a domain specific language for grammars, parser combinator libraries are built up with combinator functions which are composed to consume a data stream.

In the case of nom, a large library of macros are provided for producing new combinator function out old ones. So the complicated job of parsing an entire file format turns into a collection of the simpler tasks of parsing portions of the file format.

For example, in the Pando specification (see yesterday's post on Pando for more details) there are 3 parts of a todo item: the status bullet, the identifier, and the dependencies list. We will implement parsers for each. We will then combine the combinators in a future post.

Status

  - Command Line [ IP, DG ]
^

The status bullet determines whether the described task is Completed, In Progress, or Waiting.

  #[derive(Debug, PartialEq)]
enum TaskStatus {
    Completed,
    InProgress,
    Waiting
}
named!(parse_status<CompleteStr, TaskStatus>,
    alt!(map!(tag!("x"), |_| TaskStatus::Completed) |
         map!(tag!(">"), |_| TaskStatus::InProgress) |
         map!(tag!("-"), |_| TaskStatus::Waiting)));

This code reads pretty much as you might expect. Each tag! macro takes a string indicating what to match. The map! macro takes the result of a successful match and runs it through the past in lambda, in this case we use the lambda to return the enum version of the token. alt! tries each of it's arguments in sequence proceeding to the next one when a given argument fails to parse the input. Lastly, named! wraps the passed in combinators in a function threading the input type (in this case CompleteStr, more on this type later. For now, think of CompleteStr as a small wrapper over &str) through each combinator. The type parameters specify the expected input and output types.

Calling the combinator we just made is pretty self explanatory.

  #[test]
fn parse_status_works() {
    assert_eq!(parse_status(CompleteStr("x")), Ok((CompleteStr(""), TaskStatus::Completed)));
    assert_eq!(parse_status(CompleteStr(">")), Ok((CompleteStr(""), TaskStatus::InProgress)));
    assert_eq!(parse_status(CompleteStr("-")), Ok((CompleteStr(""), TaskStatus::Waiting)));
}

Every combinator returns a Result<(I, O), E> where I is the remaining input to be parsed, O is the output of the combinator, and E is the nom error type. So in the case of the test we expect an empty string remaining, and the enum value.

Identifier

  - Command Line [ IP, DG ]
  ^^^^^^^^^^^^

The identifier is used in the output format and as the basis for the dependency abbreviations

  named!(parse_identifier<CompleteStr, &str>, map!(is_not!("[\n"), |id| id.trim()));

The identifier parser is simpler. Moving from the inside out, is_not! matches the longest stream of characters which do not contain any of the characters in the input stream. In our case, since the dependencies of a task are optional, we want to match any character that is not the start of a dependency list or the end of a line. Moving outward, map! is used this time to trim the white-space around the string. Note: technically the string being parsed and produced is a CompleteStr however since CompleteStr impls Deref to &str we still have access to the trim function. Finally named! wraps the combinators in a function.

Dependencies

  - Command Line [ IP, DG ]
               ^^^^^^^^^^

The dependencies list is used to describe what tasks must finish before this one.

  named!(parse_dependencies<CompleteStr, Vec<&str> >,
    map!(opt!(complete!(delimited!(
        char!('['),
        separated_list!(char!(','), map!(is_not!(",]"), |dep| dep.trim())),
        char!(']')))),
    |deps| deps.unwrap_or(Vec::new())));

The dependency parser was the most difficult to get right. I will describe the operators and then give my theory for why they were required. I feel strongly that this section could be done simpler... but this is as good as I could figure out.

Moving from the inside out again, map!(is_not!(",]"), |dep| dep.trim()) parses any dependency abbreviation. We know that an abbreviation must either be followed by a , or a ] since the abbreviations are in the list structure. So we can play the same trick as we did with the identifier, and match everything including white space, but then trim the result.

  // fn dependency(input: CompleteStr) -> &str
map!(is_not!(",]"), |dep| dep.trim())

separated_list! takes a separator parser, and an item parser. In our case, the separator is the single character , so we use the char!(',') parser to match it. This returns a list of the results we produced in the dependency item parser.

  // fn dependencies(input: CompleteStr) -> Vec<&str>
separated_list!(char!(','), map!(is_not!(",]"), |dep| dep.trim()))

delimited! takes 3 arguments, a leading delimiter parser, a body parser, and a trailing delimiter parser. Conveniently the only result is the body so we don't have to do a further map here.

  // fn dependency_list(input: CompleteStr) -> Vec<&str>
delimited!(
    char!('['),
    separated_list!(char!(','), map!(is_not!(",]"), |dep| dep.trim())),
    char!(']')),

opt! modifies a parser to return an Option and succeed if the the parser fails, but return None instead. Here we need to play a trick however and use the complete! combinator to tell nom that the input passed into the dependency_list block is all of the values, so fail instead of returning the Incomplete error. Nom was written originally for parsing data streams, not necessarily text. In that domain frequently parsing an incomplete data source may result in incorrect data, so Nom has many guards in place to try to ensure failure on incomplete results whenever possible. For our task however this is only a distraction which we solve with the CompleteStr input type and complete! combinator.

  opt!(complete!(delimited!(
    char!('['),
    separated_list!(char!(','), map!(is_not!(",]"), |dep| dep.trim())),
    char!(']'))))

Finally since the resulting type is now incorrect, we must correct it to return an empty list of dependencies when no dependency list is specified. This is handled simply with map!. unwrap_or is used to provide a default value if the parser would have returned None.

  // fn optional_dependency_list(input: CompleteStr) -> Option<Vec<&str>>
map!(opt!(complete!(delimited!(
    char!('['),
    separated_list!(char!(','), map!(is_not!(",]"), |dep| dep.trim())),
    char!(']')))),
|deps| deps.unwrap_or(Vec::new())));

Today was a long one. And again, its 1:55am my time, so I will wrap it up here. Next time I will finish up the parser and possibly look into the DOT generator.

Till tomorrow,
Keith