path: root/fable-arch.md

---
documentclass: report
title: Acceptance testing using Fable
author:
- The Fable project
- Lars Wirzenius
- Daniel Silverstone
date: WORK IN PROGRESS
keywords:
- automated acceptance testing
- scenario testing
- gherkin
- cucumber
abstract: |
  Fable is a tool that supports acceptance testing in two ways: it is a
  way to write and run automated acceptance tests, and also
  presents the acceptance test suite to non-expert readers as a
  human-readable text document.

  This document explains Fable, its architecture, its input
  language, and specifies the acceptance criteria for Fable.
...

# Beware

This document describes a future that will be, not current status quo.
Fable is its initial, active development period, and we don't want to
update this document every time we make a change. Thus, be aware that
parts of this document is speculative and may describe a version of
Fable that does not yet exist.

# Introduction

Fable is a tool for acceptance testing of software. It helps describe
acceptance criteria for all stakeholders in a software project, and
also for computers. Such criteria may come from paying clients, users,
the developers, their employers, or elsewhere.

Fable is specifically meant for describing automated acceptance tests.
It takes a two-pronged approach, where it lets developers write
automated tests for all the acceptance criteria they have, and runs
the tests. On the other hand, Fable also produces a standalone
document, which describes the automated tests all stakeholders,
including non-technical ones, such as project managers and clients.

More concretely, Fable helps developers specify the automated
acceptance tests so that the tests can be executed, but also
understood without requiring programming knowledge to understand.

Fable is meant to be used to produce a document to facilitate
communication between various shareholders of the software being
developed.

Fable's overall working principle is that the tests are implemented and
documented in a number of source files, which by two Fable tools. One
tool produces a PDF or HTML document, for humans to read. The other
produces a test program, which executes the acceptaance tests.

# Requirements

This chapter lists requirements for Fable. These requirements are not
meant to be testable as such. For more specific, testable acceptance
criteria, see the later [chapter with acceptance tests for
Fable](#acceptance).

Each requirement here is given a unique mnemnoic id for easier
reference in discussions.

* **UnderstadableTests:** Acceptance tests should be possible to express in
  a way that's easily understood by non-programmers.

* **EasyToWriteDocs:** The markup language for writing documentation
  should be easy to write.

* **AidsComprehension:** The formatted human-readable documentation
  should use good layout and typography to enhance comprension.

* **CodeSeparately:** The code to implement the acceptance tests
  should not be embedded in the documentation source, but be in
  separate files. This makes it easier to edit without specialised
  tooling.

* **AnyProgammingLanguage:** The developers implementing the
  acceptance tests should be free to use a language they're familiar
  and comfortable with. Fable should not require them to use a
  specific language.

* **FastTestExecution:** Executing the acceptance tests should be
  fast.

* **NoDeployment:** The acceptance test tooling should assume the
  system under test is already deployed and available. Deploying is
  too big of a problem space to bring into the scope of acceptance
  testing, and there are already good tools for deployment.

# Fable architecture

```dot
md [label="document source\n(Markdown)"];
md [shape=box];

bindings [label="bindings file\n(YAML)"];
bindings [shape=box];

impl [label="step implementations\n(Python, Rust, ...)"]
impl [shape=box];

fable [label="Fable"];
fable [shape=ellipse];

pdf [label="PDF document"]
pdf [shape=box];

testprog [label="Test program\n(generated)"]
testprog [shape=box];

report [label="Test report"]
report [shape=box];

md -> fable;
bindings -> fable;
impl -> fable;
fable -> pdf;
fable -> testprog;
testprog -> report;
```

Fable reads input files, and produces two outputs.
On the one hand, it outputs a program that executes the tests
specified in the input files. The person running Fable then runs the
test program to get a test report, with results of each of the test
scenarios. On the other hand it outputs a human-readable document (as
PDF or HTML), for communicating what is being tested.

Fable is able to produce the test program in various languages, using
a templating system to make it simple to add support for more
languages. Fable comes with support for Python and Rust. It's the
user's choice which language they're most comfortable with.

Acceptance tests are expressed to Fable in the form of test scenarios,
in which a sequence of actions are taken, and then the results are
checked. If the checks fail, the scenario fails.

Fable runs the scenarios concurrently (but see the USING keyword),
within the constrains of hardware resources. If Fable determines it
doesn't have all the resources to run all scenarios at once, it will
run fewer, but randomly chosen scenarios concurrently, to more likely
to detect unintentional dependencies between scenarios.

Note that Fable or the scenarios it runs aren't meant deploy the
software, or start services to run in the background.

# The Fable input language

Fable input consists of three types of files:

* one or more markdown files which document the acceptance tests
* a binding file (in YAML) which binds scenario steps to their
  implementations
* scenario step implementations, which are implemented in some
  programming language (e.g., Python or Rust); Fable will combine this
  code with some scaffolding provided by Fable itself

The input files for a simple acceptance test suite for Fable would be
divided into three files: `foo.md`, `foo.yaml`, and `foo.py` (assuming
step implementation in Python).

## Markdown files

[fenced code blocks]: https://pandoc.org/MANUAL.html#fenced-code-blocks
[Gherkin]: https://en.wikipedia.org/wiki/Cucumber_(software)#Gherkin_language
[Cucumber]: https://en.wikipedia.org/wiki/Cucumber_(software)

The Fable input language is markdown files using [fenced code blocks][]
with backticks. The code blocks must indicate that they contain Fable
language:

    ```fable
    given a service
    and I am Tomjon
    when I access the service
    then it's OK
    ```

Code blocks for the PlantUML or GraphViz dot markup languages are
also supported. These can be used for generating embedded images in
the documented produced by Fable.

Any unfenced code blocks (indented code blocks) are ignored by Fable,
as are fenced code blocks using unknown content types.

The Fable code generator understands the full Markdown language, by
ignoring everything except headings and its own code blocks. The Fable
document generator uses [Pandoc][] to produce standalone files, and
anything that Pandoc supports is OK to use.

[Pandoc]: https://pandoc.org/

Fable treats multiple Markdown files as one, as if they had been
concatenated with the **cat**(1) utility. Within the logical file,
normal Markdown and Pandoc markup can be used to structure the
document in any way that aids human understanding of the acceptance
test suite, which the caveat that chapter or section headings are used
by Fable to group code blocks into scenarios.

All code blocks for the same scenario must be grouped under a single
heading. Sub-headings are permitted within a scenario, but the next
heading at the same or a higher level will end the scenario. This
allows for scenarios to begin at any level, but not to leak into a
wider scope within the acceptance document. For example, a scenario
which starts after a level 2 heading may have subdivisions marked with
level 3 or below headings, but will end at the next level 2 or level 1
heading.

Within the Fable code blocks, Fable understands a special language,
derived from [Gherkin][], as defined by the [Cucumber][] testing tool.
The language understood by Fable has the following general structure:

* each logical line starts with a keyword at the beginning of the line
* logical lines may be broken into physical lines, by starting the
  continuation lines with one or more space or TAB characters; the
  physical line break and white space characters are preserved
* logical lines define steps in a test scenario
* the meaning and implementation of the steps are defined by other
  Fable input files
* the keywords are: ASSUMING, USING, GIVEN, WHEN, THEN, and AND, with
  meanings defined below; keywords can be written in upper or lower
  case, or mixes, Fable doesn't care

The keywords have the following meanings:

* ASSUMING—a condition for the scenario. If an ASSUMING step
  fails, the scenario is skipped.

  This is used for skipping test scenarios that require specific
  software to be installed in the test environment, or access to
  external services, but which can't be required for all runs of the
  acceptance tests.

* USING—indicate that the scenario uses a resource such as a
  database, that's constrained and can't be used by all scenarios if
  they run concurrently. When scenarios declare the resource, Fable can
  limit which scenarios run concurrently.

  For example, if several scenarios require uncontested use of the
  GPU, of which there is typically only one per machine, they can all
  declare "using the graphical processing unit", and Fable will run
  them one at a time.

  (This is an intentionally simplistic way of controlling concurrency.
  The goal is to be simple and correct rather then achievee maximal
  concurrency.)

  The actual management of resources belongs to the generated test
  program at runtime, not the Fable compiler.

* GIVEN—set up the test environment for the action (WHEN). This
  might create files, start a background process, or something like
  that. This also sets up the reversal of the setup, so that any
  background processes are stopped automatically after the scenario
  has ended. The setup and cleanup must succeed, or the scenario will
  fail.

  The cleanups are executed in the reverse order of the GIVENs, and
  they're done always, whether the scenario succeeds or not.

* WHEN—perform the action that is being tested. This must
  succeed. This might, for example, execute a command line program,
  and capture its output and exit code.

* THEN—test the results of the action. This would examine the
  output and exit code of the program run in a WHEN step, or examine
  current content of the database, or whatever is needed.

* AND—this keyword exists to make scenarios "read" better in
  English. The keyword indicates that this step should use the same
  keyword as the previous step, whatever that keyword is. For example,
  a step "THEN output is empty" might be followed by "AND the exit
  code is 0" rather than "THEN the exit code is 0".

The order of steps is somewhat constrainted: first any ASSUMING steps,
then any USING steps, at least one WHEN must come before a THEN.

## Bindings

FIXME: The binding specification needs thought. This is just a sketch.

Binding files match scenario steps to functions that implement them,
using regular expressions. The bindings may also extract parts of the
steps, and pass them onto the functions as parameters.

Binding files are YAML files, with lists of bindings, each binding
being a dict. For example:

```yaml
- define:
    name: string
    exit_code: int

- pattern: given a service
  function: start_service
  cleanup: stop_service

- pattern: given I am (?P<name\S+)
  function: set_name
  produces: [name]

- pattern: when I access the service
  function: access_service
  requires: [name]
  produces: [exit_code]

- pattern: then it's OK
  function: check_access_was_ok
  requires: [exit_code]
```

In the example above, the "I am" step extracts the name of the user
from the step. It's type is declared, and the value is saved for use
by a later step.

The "I access" step expects the name to have been set by a previous
step. Fable will check that the name is set, and give an error if it
isn't, before any scenario runs. If name is set, it is given to the
function to be called as a function argument.

The "I access" step further sets the variable "exit_code", and the
"it's OK" step expects it to be set.

## Step implementations

Continuing the example from the previous section, the following Python
code might implement the functions:

```python
def start_service():
    ...

def set_name(**matches):
    ...

def access_service(name):
    ...
    return {
        'exit_code': 0,
    }

def check_access_was_ok(exit_code):
    assert exit_code == 0
```

With these bindings, Fable produces a Python program, which calls
these functions in order, and passes values between them via function
arguments and return values. The generated program will handle running
scenarios concurrently, and taking care of USING constraints, and
other resource constraints.

# Acceptance tests for Fable {#acceptance}

FIXME: This needs to be written eventually, when Fable is implemented.

## Simple scenario

We test that Fable can execute a very simple scenario. The test is
based on generating the test program from an input file, running the
test program, and examining the log file. This will require that the
generated test program logs the things it does in a known format.
Effectively, the log messages become a part of the contract.

````fable
given file "simple.md" containing
   # Simple
   ```fable
   given precondition foo
   when I do bar
   then bar was done
   ```
and file "simple.yaml" containing
    - given: precondition foo
      function: nop
    - when: I do bar
      function: nop
    - then: bar was done
      function: nop
and file "simple.py" containing
    def foo(ctx): pass
when I run ftt-codegen --run simple.md
then log file says scenario "Simple" was run
and log file says step "given precondition foo" was run
and log file says step "when I do bar" was run
and log file says step "then bar was don" was run
and program finished successfully
````