README for story testing
========================

Introduction
------------

`yarn` is a story testing tool: you write a story describing how a
user uses your software and what should happen, and express, using
very lightweight syntax, the story in such a way that it can be tested
automatically. The story has a simple, but strict structure:

    GIVEN some setup for the test
    WHEN thing that is to be tested happens
    THEN the post-conditions must be true

As an example, consider a very short test story for verifying that
a backup program works, at least for one simple case.

    GIVEN some live data in a directory
    AND an empty backup repository
    WHEN a backup is made
    THEN the data case be restored

(Note the addition of AND: you can have multiple GIVEN, WHEN, and
THEN statements. The AND keyword makes the text be more readable.)

Stories are meant to be written in somewhat human readable language.
However, they are not free form text. In addition to the GIVEN/WHEN/THEN
structure, the text for each of the steps needs a computer-executable
implementation. This is done by using IMPLEMENTS. The backup story
from above might be implemented as follows:

    IMPLEMENTS GIVEN some live data in a directory
    rm -rf "$TESTDIR/data"
    mkdir "$TESTDIR/data"
    echo foo > "$TESTDIR/data/foo"

    IMPLEMENTS GIVEN an empty backup repository
    rm -rf "$TESTDIR/repo"
    mkdir "$TESTDIR/repo"

    IMPLEMENTS WHEN a backup is made
    backup-program -r "$TESTDIR/repo" "$TESTDIR/data"

    IMPLEMENTS THEN the data can be restored
    mkdir "$TESTDIR/restored"
    restore-program -r "$TESTDIR/repo" "$TESTDIR/restored"
    diff -rq "$TESTDIR/data" "$TESTDIR/restored"

Each "IMPLEMENT GIVEN" (or WHEN, THEN) is followed by a regular
expression on the same line, and then a shell script that gets executed
to implement any step that matches the regular expression.  The
implementation can extract data from the match as well: for example,
the regular expression might allow a file size to be specified.

The above example is a bit silly, of course: why go to the effort
to obfuscate the various steps? The answer is that the various
steps, implemented using IMPLEMENTS, can be combined in many
ways, to test different aspects of the program being tested.

Moreover, by making the step descriptions be human language
text, matched by regular expressions, most of the test can
hopefully be written, and understood, by non-programmers. Someone
who understands what a program should do, could write tests
to verify its behaviour. The implementations of the various
steps need to be implemented by a programmer, but given a
well-designed set of steps, with enough flexibility in their
implementation, that quite a good test suite can be written.

Test language specification
---------------------------

A test document is written in [Markdown][markdown], with block
quoted code blocks being interpreted specially. Each block
must follow the syntax defined here.

* Every step in a story is one line, and starts with a keyword.
* Each implementation (IMPLEMENTS) starts as a new block, and
  continues until there is a block that starts with another
  keyword.

The following keywords are defined.

* **STORY** starts a new story. The rest of the line is the name of
  the story. The name is used for documentation and reporting
  purposes only and has no semantic meaning. STORY MUST be the
  first keyword in a story, with the exception of IMPLEMENTS.
  The set of documents passed in a test run may define any number of
  stories between them, but there must be at least one or it is a
  test failure. The IMPLEMENTS sections are shared between the
  documents and stories.

* **ASSUMING** defines a condition for the story. The rest of the
  line is "matched text", which gets implemented by an
  IMPLEMENTS section. If the code executed by the implementation
  fails, the story is skipped.

* **GIVEN** prepares the world for the test to run. If
  the implementation fails, the story fails.

* **WHEN** makes the change to the world that is to be tested.
  If the code fails, the story fails.

* **THEN** verifies that the changes made by the GIVEN steps
  did the right thing. If the code fails, the story fails.

* **FINALLY** specifies how to clean up after a story. If the
  code fails, the story fails. All FINALLY blocks get run either when
  encountered in the story flow, or at the end of the story, regardless
  of whether the story is failing or not.

* **AND** acts as ASSUMING, GIVEN, WHEN, THEN, or FINALLY: whichever
  was used last. It must not be used unless the previous step was
  one of those, or another AND.

* **IMPLEMENTS** is followed by one of ASSUMING, GIVEN, WHEN, or
  THEN, and a PCRE regular expression, and then shell commands until
  the end of the block quoted code block. Markdown is unclear whether
  an empty line (no characters, not even whitespace) between two
  block quoted code blocks starts a new one or not, so we resolve the
  ambiguity by specifiying that a code block directly following a code
  block is a continuation unless it starts with one of the story
  testing keywords.

  The shell commands get parenthesised parts of the match of the
  regular expression as environment variables (`$MATCH_1` etc). For
  example, if the regexp is "a (\d+) byte file", then `$MATCH_1` gets
  set to the number matched by `\d+`.

  The test runner creates a temporary directory, whose name is
  given to the shell code in the `TESTDIR` environment variable.

  The shell commands get invoked with `/bin/sh -eu`, and need to
  be written accordingly. Be careful about commands that return
  a non-zero exit code. There will be a library of shell functions
  supplied which allow handling the testing of non-zero exit codes
  cleanly. In addition functions for handling stdout and stderr will
  be provided.

  The code block of an IMPLEMENTS block fails if the shell
  invocation exits with a non-zero exit code. Output to stderr is
  not an indication of failure. Any output to stdout or stderr may
  or may not be shown to the user.

Semantics:

* The name of each story (given with STORY) must be unique.
* All names will be normalised before use (whitespace collapse, leading
  and trailing whitespace
* Every ASSUMING, GIVEN, WHEN, THEN, FINALLY must be matched by
  exactly one IMPLEMENTS. The test runner checks this before running
  any code.
* Every IMPLEMENTS may match any number of ASSUMING, GIVEN, WHEN,
  THEN, or FINALLY. The test runner can warn if an IMPLEMENTS is unused.
* If ASSUMING fails, that story is skipped, and any FINALLY steps
  are not run.

See also
--------

Wikipedia has an article on [Behaviour Driven Development][BDD],
which can provide background and further explanation to what this
tools tries to do.

[BDD]: https://en.wikipedia.org/wiki/Behavior-driven_development
[Markdown]: http://daringfireball.net/projects/markdown/

TODO
----

* Add DEFINING, PRODUCING, if they turn out to be useful.
* Need something like ASSUMING, except fail the story if the
  pre-condition is not true. Useful for testing that you can ssh
  to localhost when flinging, for example.
  **DJAS**: We think this might be 'REQUIRING' and it still does
  not run the FINALLY group.
* Consider the ordering some more.  IMPLEMENTS can come anywhere
  but otherwise stories are defined as:
    * STORY
    * 0+: ASSUMING
    * 1+:
        * 1+: GIVEN
        * 1+:
            * 1+: WHEN
            * 1+: THEN
        * 0+: FINALLY