# tap-parser parse the [test anything protocol](http://testanything.org/) [![build status](https://secure.travis-ci.org/tapjs/tap-parser.png)](http://travis-ci.org/tapjs/tap-parser) [![browser support](http://ci.testling.com/substack/tap-parser.png)](http://ci.testling.com/substack/tap-parser) [![coverage status](https://coveralls.io/repos/tapjs/tap-parser/badge.svg?branch=master&service=github)](https://coveralls.io/github/tapjs/tap-parser?branch=master) # example ``` js var parser = require('tap-parser'); var p = parser(function (results) { console.dir(results); }); process.stdin.pipe(p); ``` given some [TAP](http://testanything.org/)-formatted input: ``` $ node test.js TAP version 13 # beep ok 1 should be equal ok 2 should be equivalent # boop ok 3 should be equal ok 4 (unnamed assert) 1..4 # tests 4 # pass 4 # ok ``` parse the output: ``` $ node test.js | node parse.js { ok: true, count: 4, pass: 4, plan: { start: 1, end: 4 } } ``` # usage This package also has a `tap-parser` command. ``` Usage: tap-parser Parses TAP data from stdin, and outputs the parsed result in the format specified by the options. Default output is uses node's `util.format()` method. Options: -j [] | --json[=indent] Output event data as JSON with the specified indentation (default=2) -t | --tap Output data as reconstituted TAP based on parsed results -l | --lines Output each parsed line as it is recognized by the parser -b | --bail Emit a `Bail out!` at the first failed test point encountered -w | --ignore-all-whitespace Skip over blank lines outside of YAML blocks -o | --omit-version Ignore the `TAP version 13` line at the start of tests ``` # methods ``` js var parser = require('tap-parser') ``` ## var p = parser(options, cb) Return a writable stream `p` that emits parse events. If `cb` is given it will listen for the `'complete'` event. If `options` is given, it may contain the following flags: - `preserveWhitespace` boolean which is `false` by default and will cause the parser to emit `line` events even for lines containing only whitespace. (Whitespace lines in yaml blocks are always emitted, because whitespace is semantically relevant for yaml.) - `strict` boolean which is `false` by default and causes the parser to treat non-TAP input as a failure. Strictness is heritable to child subtests. You can also turn strictness on or off by using the `pragma +strict` line in the TAP data to turn strictness on, or `pragma -strict` to turn strictness off. - `bail` boolean which is `false` by default and will cause the parser to bail out (including emitting a synthetic `Bail out!` line) whenever a failed test point is encountered. - `omitVersion` boolean which is `false` by default and will cause the parser to ignore `TAP version 13` lines. Version lines in subtests cause problems with some parsers, so they are always ignored. - `passes` boolean which is false by default and will add "passes" property for that contains the result of all passed tests The `parent`, `level` and `buffered` options are reserved for internal use. # events ## p.on('complete', function (results) {}) The `results` object contains a summary of the number of tests skipped, failed, passed, etc., as well as a boolean `ok` member which is true if and only if the planned test were all found, and either "ok" or marked as "TODO". ## p.on('line', function (line) {}) As each line of input is parsed, a `line` event is emitted. "Synthetic" line events will be emitted to support the `bail` behavior, and to inject `1..0` plan lines in subtests that have no test points. They can be used as a sort of "passthrough stream" to sanitize and filter a TAP stream, with the caveat that, while `line` events will be semantically equivalent to the TAP input, they will not be a perfect replica of the input. ## p.on('assert', function (assert) {}) Every `/^(not )?ok\b/` line will emit an `'assert'` event. Every `assert` object has these keys: * `assert.ok` - true if the assertion succeeded, false if failed * `assert.id` - the assertion number * `assert.name` - optional short description of the assertion and may also have * `assert.todo` - optional description of why the assertion failure is not a problem. (Boolean `true` if no explaination provided) * `assert.skip` - optional description of why this assertion was skipped (boolean `true` if no explanation provided) * `assert.diag` - a diagnostic object with additional information about the test point. ## p.on('comment', function (comment) {}) Every `/^# (.+)/` line will emit the string contents of `comment`. ## p.on('plan', function (plan) {}) Every `/^\d+\.\.\d+/` line emits a `'plan'` event for the test numbers `plan.start` through `plan.end`, inclusive. If the test is [completely skipped](http://podwiki.hexten.net/TAP/TAP.html?page=TAP#Skippingeverything) the result will look like ``` { ok: true, count: 0, pass: 0, plan: { start: 1, end: 0, skipAll: true, skipReason: 'This code has no seat belt' } } ``` ## p.on('version', function (version) {}) A `/^TAP version (\d+)/` line emits a `'version'` event with a version number or string. ## p.on('bailout', function (reason) {}) A `bail out!` line will cause the parser to completely stop doing anything. Child parser bailouts will bail out their parents as well. ## p.on('child', function (childParser) {}) If a child test set is embedded in the stream like this: ``` TAP Version 13 1..2 # nesting 1..2 ok 1 - true is ok ok 2 - doag is also okay ok 1 - nesting ok 2 - second ``` then the child stream will be parsed and events will be raised on the `childParser` object. Since TAP streams with child tests *must* follow child test sets with a pass or fail assert based on the child test's results, failing to handle child tests should always result in the same end result. However, additional information from those child tests will obviously be lost. See `Subtests` below for more information on which sorts of subtest formats are supported by this parser. ## p.on('extra', function (extra) {}) All other lines will trigger an `'extra'` event with the line text. # install With [npm](https://npmjs.org) do: ``` npm install tap-parser ``` You can use [browserify](http://browserify.org) to `require('tap-parser')` in the browser. # license MIT # subtests 5 flavors of Subtests are suppored by this parser. 1. Unadorned. Indented TAP data with no comment, followed by a test point at the parent level. ``` ok 1 1..1 ok 1 - child test 1..1 ``` 2. Indented comment. An indented `# Subtest: ` comment, followed by indented TAP data, and then a not-indented test point with a matching name. The summary test point may have yaml diagnostics. ``` # Subtest: child test ok 1 1..1 ok 1 - child test 1..1 ``` 3. Unindented comment. A not-indented `# Subtest: ` comment, followed by indented TAP content, followed by a test point with a matching name. The summary test point may have yaml diagnostics. ``` # Subtest: child test ok 1 1..1 ok 1 - child test 1..1 ``` 4. Buffered, without diagnostics. A test point line ending in {, followed by indented TAP content, ended with a } to close the block. todo/skip directives may come *either* before or after the `{` character. Yaml diagnostics are not allowed. ``` ok 1 - child test { ok 1 1..1 } 1..1 ``` 5. Buffered, with diagnostics. A test point line with yaml diagnostics, followed by `{` alone on a line, indented TAP data, and then a `}`. ``` ok 1 - child test --- some: diagnostic data: true ... { ok 1 1..1 } ``` In all cases, the parsed behavior is identical: 1. The parent emits a `child` event with the `childParser` as an argument. 2. The `childParser` emits a `comment` with `# Subtest: ` (or `(anonymous)` for Unadorned subtests.) 3. When the test is over, the closing test point is emitted on parent test. That is, buffered and nonindented/indented comment subtests are parsed as if they are identical input, since their semantics are the same. This simplifies implementation of test harness and reporter modules. Since unadorned subtests have no introduction, a child event is not emitted until the first "relevant tap" line is encountered. This can cause confusion if the test output contains a spurious " 1..2" line or something, but such cases are rare. Similarly, this means that a test point ending in `{` needs to wait to emit *either* the 'assert' or 'child' events until an indented line is encountered. *Any* test point with yaml diagnostics needs to wait to see if it will be followed by a `{` indicating a subtest.