bjmc / pest-test Goto Github PK

****# pest-test

Testing framework for pest parser (similar to tree-sitter test).

A test case is a text file with three sections:

• The test name must be on the first line of the file. It may contain any characters except newline.
• The source code block is delimited by any sequence of three or more non-whitespace characters. The same sequence of characters must preceed and follow the code block, and that sequence of characters may not appear anywhere within the code block.
• The code block must be both preceeded and followed by a line separator. The outer-most line separators are trimmed off. This means that if your parser is sensitive to leading/trailing whitespace, you must make sure to put the correct number of empty lines before/after the code block.
• The expected output syntax tree written as an S-expression. Optionally, a terminal node may be followed by its expected string value.

Here is an example test. Note that the code block delimiter is exactly 7 '=' characters. In this case, the parser ignores implicit whitespace, so the numbers of blank lines before/after the code are arbitrary.

My Test

=======

fn x() int {
  return 1;
}

=======

(source_file
  (function_definition
    (identifier: "x")
    (parameter_list)
    (primitive_type: "int")
    (block
      (return_statement 
        (number: "1")
      )
    )
  )
)

Nodes in the expected output S-expression may be annotated with attributes of the form #[name(args)]. The currently recognized attributes are:

• skip: For some grammars, there are multiple levels of nesting that are only necessary to work around the limitations of PEG parsers, e.g. mathematical expressions with left-recursion. To simplify test cases involving such grammars, the skip attribute can be used to ignore a specified number of levels of nesting in the actual parse tree.

  (expression
    #[skip(depth = 3)]
    (sum
      (number: 1)
      (number: 2)
    )
  )
  

The main interface to the test framework is pest_test::PestTester. By default, tests are assumed to be in the tests/pest directory of your crate and have a .txt file extension. The example below shows using the lazy_static macro to create a single PestTester instance and then using it to evaluate any number of tests.

#[cfg(test)]
mod tests {
  use mycrate::parser::{MyParser, Rule};
  use lazy_static::lazy_static;
  use pest_test::{Error, PestTester};

  lazy_static! {
    static ref TESTER: PestTester<Rule, MyParser> = 
      PestTester::from_defaults(Rule::root_rule);
  }

  // Loads test from `tests/pest/mytest.txt` and evaluates it. Returns an `Err<pest_test::Error>`
  // if there was an error evaluating the test, or if the expected and actual values do not match.
  fn test_my_parser -> Result<(), Error> {
    (*TESTER).evaluate_strict("mytest")
  }
}

If you add pest-test-gen as a dev dependency, then you can use the pest_tests attribute macro to generate tests for all your test cases:

// Generate tests for all test cases in tests/pest/foo/ and all subdirectories. Since
// `lazy_static = true`, a single `PestTester` is created and used by all tests; otherwise a new
// `PestTester` would be created for each test.
#[pest_tests(
  mycrate::parser::MyParser,
  mycrate::parser::Rule,
  "root_rule",
  subdir = "foo",
  recursive = true,
  lazy_static = true,
)]
#[cfg(test)]
mod foo_tests {}

To disable colorization of the diff output, run cargo with CARGO_TERM_COLOR=never.

Note that a test module is only recompiled when its code changes. Thus, if you add or rename a test case in tests/pest without changing the test module, the test module might not get updated to include the new/renamed tests, so you may need to delete the target folder to force your tests to be recompiled.

Test files are parsed using pest. The source code is parsed using your pest grammar, and the resulting tree is iterated exhaustively to build up a nested data structure, which is then matched to the same structure built from the expected output. If they don't match, the tree is printed with the differences in-line.