Tests a command line program by executing it in a temporary sandbox directory and inspecting its result.

Or tests properties of existing files, directories etc.

Assertions are expressed in a declarative style, using a specialized type system.

Syntax is inspired by shell script, but thoroughly specialized.

A test may have setup and cleanup actions.

Supports individual test cases and test suites.

Supports easy referencing of predefined files and files created in the temporary sandbox.

Supports execution of arbitrary programs, as well as checking their result.

Exactly has a built in help system, which can, among other things, generate this Reference Manual.

TEST CASES

A test case is written as a plain text file.

Testing stdin, stdout, stderr, exit code

The following checks that your new my-contacts-program reads a contact list from stdin, and is able to find the email of a person:

[setup]

stdin = -contents-of some-test-contacts.txt

[act]

my-contacts-program get-email --name 'Pablo Gauss'

[assert]

exit-code == 0

stdout equals <<EOF
pablo@gauss.org
EOF

stderr is-empty

If the file 'contacts.case' contains this test case, then Exactly can execute it:

 > exactly contacts.case
 PASS

"PASS" means that all assertions were satisfied.

This test assumes that

• the system under test - my-contacts-program - is is found in the same directory as the test case file
• the file "some-test-contacts.txt" (that is referenced from the test case) is found in the same directory as the test case file

If the actual email address of "Pablo Gauss" is not the expected one, then Exactly will report failure. For example:

 > exactly contacts.case
 FAIL
 In [assert]
 contacts.case, line 13

   stdout equals <<EOF
   pablo@gauss.org
   EOF


 Unexpected contents of stdout from the "action to check"

   @[EXACTLY_RESULT]@/stdout


 (F) equals
       Expected
         STRING
           'pablo@gauss.org\n'
       Diff
 --- Expected
 +++ Actual
 @@ -1 +1 @@
 -pablo@gauss.org
 +pablo.gauss@masters.org

Testing side effects on files and directories

When the execution of a test case starts, the current directory is set to a temporary directory. This gives the test case a sandbox where it can create and manipulate files.

The sandbox - and all files within it - are removed when the execution ends.

The following tests a program that classifies files as either good or bad, by moving them to the appropriate directory:

[setup]

dir output/good
dir output/bad

dir input =
{
    file a.txt = 'GOOD contents'
    file b.txt = 'bad contents'
    dir  sub   = { file c.txt = 'more bad contents' }
}

[act]

classify-files-by-moving-to-appropriate-dir GOOD input/ output/

[assert]

dir-contents input       : is-empty

dir-contents output/good : matches -full { a.txt : type file }

dir-contents output/bad  : matches -full
    {
        b.txt : type file
        sub   : type dir &&
                dir-contents matches -full
                {
                    c.txt : type file
                }
    }

dir creates a directory in the current directory (by default).

Testing and transforming the contents of files

Use contents to test the contents of a file, or a transformed version of it, by applying a "text transformer".

Such a "text transformer" may be given a name using the def instruction to make the test easier to read.

The following case tests that "timing lines" are output as part of a log file "log.txt".

The challenge is that the (fictive) log file contains non-timing lines that the test is not interested in, and that timing lines contains a time stamp of the form "NN:NN", who's exact value also is not interesting.

A "text transformer" is used to extract all timing lines and to replace "NN:NN" time stamps with the constant string TIMESTAMP:

[setup]

def line-matcher     IS_TIMING_LINE     = contents matches ^timing

def text-transformer REPLACE_TIMESTAMPS = replace [0-9]{2}:[0-9]{2} TIMESTAMP

def text-transformer GET_TIMING_LINES   = filter IS_TIMING_LINE | REPLACE_TIMESTAMPS

[act]

program-that-writes-log-file

[assert]

contents log.txt :
         -transformed-by GET_TIMING_LINES
         equals <<EOF
timing TIMESTAMP begin
timing TIMESTAMP preprocessing
timing TIMESTAMP validation
timing TIMESTAMP execution
timing TIMESTAMP end
EOF

The -transformed-by option does not modify the tested file, it just applies the assertion to a transformed version of it.

Using external programs

External programs can help with setup, assertions etc.

Exactly can run executable files, shell commands and programs in the OS PATH, using run, $, %.

The following case shows some examples, but doesn't make sense as a realistic test case, tough:

[setup]

run my-setup-helper-program first "second arg"

run my-setup-helper-program arg
    -stdin 'the stdin of the program'

run -ignore-exit-code my-setup-helper-program


def list DB_ARGS = -uu -pp -hlocalhost -Dd

run % mysql @[DB_ARGS]@ --batch --execute "create table my_table(id int)"

run % mysql @[DB_ARGS]@ --batch --execute :> create table my_table(id int)

def list MYSQL_BATCH = @[DB_ARGS]@ --batch --execute

file interesting-records.txt =
     -stdout-from
       % mysql @[MYSQL_BATCH]@ :> select * from a_table where name = "interesting"


% touch file

$ ls *.txt

file root-files.txt =
     -stdout-from % ls /
       -transformed-by
         run my-text-transformer-program

file interesting-pgm-output.txt =
     -stdout-from
       -python @[EXACTLY_HOME]@/my-text-generating-program.py
       -transformed-by
         strip -trailing-new-lines

[act]

$ echo ${PATH} > output.txt

[assert]

run my-assert-helper-program

% stat root-files.txt

$ test -f root-files.txt

stdout -from $ echo 'Interesting output'
       equals "Interesting output@[NEW_LINE]@"

exit-code -from my-assert-helper-program
          ( <= 2 || > 10 )

exists output.txt : (
       type file
       &&
       run -python @[EXACTLY_HOME]@/my-file-matcher.py arg1
       &&
       contents run -python @[EXACTLY_HOME]@/my-text-matcher.py arg1 "arg 2"
       )

[cleanup]

% mysql @[MYSQL_BATCH]@ :> drop table my_table

A program executed in [assert] becomes an assertion that depends on the exit code.

Program values can be defined for reuse using def, and referenced using @:

[setup]

def program RUN_MYSQL   = % mysql -uu -pp -hlocalhost -Dd
def program EXECUTE_SQL = @ RUN_MYSQL --skip-column-names --batch --execute


run @ EXECUTE_SQL "create table my_table(id int)"

[act]

@ EXECUTE_SQL :> CALL MyStoredProcedure()

[assert]

stdout -from
       @ EXECUTE_SQL "select * from my_table"
       ! is-empty

[cleanup]

run @ EXECUTE_SQL :> drop table my_table

:> treats the rest of the line as a single string.

Thus :> a b c becomes the string a b c.

Testing existing OS environment - tests without [act]

A test case does not need to have an [act] phase. This way, Exactly can be used to check existing files and directories, for example.

The following case checks your hierarchy of software projects.

The projects are rooted at the directory 'my-projects'. Each 'project' sub directory contains a project, and must contain a 'Makefile' with a target 'all':

[assert]

exists @[MY_PROJECTS_ROOT_DIR]@ : type dir && ALL_PROJECT_DIRS_ARE_VALID

[setup]

def path   MY_PROJECTS_ROOT_DIR = -rel-act-home my-projects
def string MY_PROJECT_DIR_NAME  = project

def file-matcher IS_VALID_MAKEFILE =

    type file &&
    contents
      -transformed-by
        filter contents matches '^all:'
        num-lines == 1


def file-matcher IS_VALID_PROJECT_DIR =

    type dir &&
    dir-contents
       matches { Makefile : IS_VALID_MAKEFILE }


def file-matcher ALL_PROJECT_DIRS_ARE_VALID =

    dir-contents -recursive
      -selection name MY_PROJECT_DIR_NAME
        every file : IS_VALID_PROJECT_DIR

The @[symbol_name]@ syntax is a reference to the "symbol" symbol_name. This syntax must be used wherever a string value is accepted, as in @[MY_PROJECTS_ROOT_DIR]@, where a file name string is accepted.

Just symbol_name will do in all other contexts.

The @[symbol_name]@ syntax can be used in any context.

Testing a git commit hook

The following tests a git commit hook (prepare-commit-msg).

The hook should add the issue id in the branch name, to commit messages:

[setup]


def string ISSUE_ID            = ABC-123
def string MESSAGE_WO_ISSUE_ID = "commit message without issue id"

def program GET_LOG_MESSAGE_OF_LAST_COMMIT = % git log -1 --format=%s


#### Setup a git repo with the commit hook to test

% git init

copy prepare-commit-msg .git/hooks


#### Setup a branch, with issue number in its name,
# and a file to commit

% git checkout -b @[ISSUE_ID]@-branch-with-issue-id

file file-on-branch.txt

% git add file-on-branch.txt


[act]


% git commit -m @[MESSAGE_WO_ISSUE_ID]@


[assert]


exit-code == 0

stdout -from
       @ GET_LOG_MESSAGE_OF_LAST_COMMIT
       equals
<<-
@[ISSUE_ID]@ : @[MESSAGE_WO_ISSUE_ID]@
-

% ... runs a program in the OS PATH.

Testing source code files

The actor instruction can specify an interpreter to test a source code file:

[conf]

actor = file % python3

[act]

my-python-program.py 'an argument' second third

[assert]

stdout equals
<<EOF
Argument: an argument
Argument: second
Argument: third
EOF

Testing source code

The actor instruction can specify an interpreter to test source code in [act]:

[conf]

actor = source % python3

[act]

import sys
sys.stdout.write('Hello\n')
sys.stdout.write('world!\n')

[assert]

stdout equals
<<-
Hello
world!
-

Print output from the tested program

If --act is used, the output of the "act" phase (the "action to check") will become the output of exactly - stdout, stderr and exit code:

[setup]

dir  a-dir
file a-file

[act]

$ ls

[assert]

stdout num-lines == 314

 > exactly --act my-test.case
 a-dir
 a-file

The test case is executed in a temporary sandbox, as usual, but assertions are ignored.

Referencing files

The home directory structure is directories containing predefined files involved in a test case:

act-home

Location of the program file being tested

home

Location of arbitrary test resources

Both of them defaults to the directory that contains the test case file, but can be changed via [conf].

Exactly does its best to prevent files in these directories from being modified.

The sandbox directory structure is temporary directories for files involved in a single execution of a test case:

act

The current directory, when execution begins

result

Stores the output from the tested program

tmp

A directory for arbitrary temporary files

There are options for making paths relative to all of these.

-rel-home refers to the home directory, and -rel-act to the act directory, for example:

[conf]

act-home = ../bin/

home     = data/

[setup]

copy  -rel-home input.txt  -rel-act actual.txt

[act]

my-grep-tool "text to find" actual.txt

[assert]

contents -rel-act actual.txt :
         equals
         -contents-of -rel-home expected.txt

-rel-home input.txt becomes a single path argument.

These "relativity" options have defaults designed to minimize the need for them. The following case does the same thing as the one above:

[conf]

act-home = ../bin/

home     = data/

[setup]

copy input.txt actual.txt

[act]

my-grep-tool "text to find" actual.txt

[assert]

contents actual.txt :
         equals
         -contents-of expected.txt

ORGANIZING TESTS

File inclusion

Test case contents can be included from external files:

[setup]

including my-dir-symbols.def

including my-common-setup-and-cleanup.xly

Test suites

Tests can be grouped in suites:

first.case
second.case

or:

[cases]

helloworld.case
*.case
**/*.case


[suites]

sub-suite.suite
*.suite
pkg/suite.suite
**/*.suite

If the file my-suite.suite contains this text, then Exactly can run it:

 > exactly suite my-suite.suite
 ...
 OK

The result of a suite can be reported as simple progress information, or JUnit XML.

Suites can contain test case functionality that is common to all cases in the suite. For example:

[cases]

*.case

[conf]

act-home = ../bin/

[setup]

def string CONF_FILE = my.conf

file @[CONF_FILE]@ =
<<EOF
common = configuration
EOF

The common functionality is included in each test case.

MORE EXAMPLES

The examples/ directory of the source distribution contains more examples.

INSTALLING

Exactly is written in Python and does not require any external libraries.

Exactly requires Python >= 3.6.

Use pip or pip3 to install:

 > pip3 install exactly

The program can also be run from a source distribution:

 > python3 src/default-main-program-runner.py

DEVELOPMENT STATUS

Current version is fully functional, but some syntax and semantics is inconsistent:

• Some instructions allow arguments to span multiple lines, some do not.
• Support for character escaping in strings is missing.
• Support for comments inside instructions is missing.

Incompatible changes to syntax and semantics may occur in every 0.x release.

Comments are welcome!

Future development

More functionality is needed, smaller and larger. Including (but by no means limited to):

• Improved string character escaping
• Improved syntax using parentheses
• Concurrent execution of processes
• Support for non-terminating processes
• Windows port (most features work, but have not been thoroughly tested)
• Symbol substitution in files
• More matchers, text transformers, etc
• Long term goals
  • Dynamic symbol values (contents of dir, current date, e.g.)
  • Macros and functions
  • Embedding of Python code in test cases
  • Python library for running cases and suites from within Python as a DSEL

AUTHOR

Emil Karlén

emil@member.fsf.org

THANKS

The Python IDE PyCharm from JetBrains has greatly helped the development of this software.

Thanks for the great

• Python language
• GNU/Linux
• GNU Emacs
• git
• Docker
• Rembrandt Harmenszoon van Rijn's "De Staalmeesters"

DEDICATION

Aron Karlén

Tommy Karlsson

Götabergsgatan 10, lägenhet 4