README Contents --------
- DOCTEST - Run examples embedded in documentation
- Example output
- Defining your expectations
- Expecting an error
DOCTEST - Run examples embedded in documentation
With doctest, you can put an example of using your function, right in the m-file help. Then, that same example can be used like a unit test, to make sure the function still does what the docs say it does.
Here's a trivial function and its documentation:
function sum = add3(value) % adds 3 to a number % % add3(value) % returns (value + 3) % % Examples: % % >> add3(7) % % ans = % % 10 % % >> add3([2 4]) % % ans = % % 5 7 % % >> add3('hi') % ??? Error using ==> add3 *** % add3(value) requires value to be a number % % % TWO blank lines before the prose description of the function continues % if ~ isnumeric(value) error('add3(value) requires value to be a number'); end sum = value + 3;
Here's the output we get from running doctest on the add3 function above:
TAP version 13 1..3 ok 1 - "add3(7)" ok 2 - "add3([2 4])" ok 3 - "add3('hi')"
Here's an example of what happens when something changes and your test fails.
By the way, output is in the Test Anything Protocol format, which I guess is mostly used by Perl people, but it's good enough for now. See http://testanything.org/
Normally, the failure report would include a link to somewhere near the doctest that failed, but that doesn't format properly in published m-files.
type should_fail disp ------------- doctest('should_fail', 'CreateLinks', 0) % the links don't work in publish()
% Has a doctest that should fail. % % >> 3 + 3 % % ans = % % 5 % ------------- TAP version 13 1..1 not ok 1 - "3 + 3" expected: ans = 5 got : ans = 6
Defining your expectations
Each time doctest runs a test, it's running a line of code and checking that the output is what you say it should be. It knows something is an example because it's a line in help('your_function') that starts with '>>'. It knows what you think the output should be by starting on the line after >> and looking for the next >>, two blank lines, or the end of the documentation.
If the output of some function will change each time you call it, for instance if it includes a random number or a stack trace, you can put '***' (three asterisks) where the changing element should be. This acts as a wildcard, and will match anything. See the example below.
Here are some examples of formatting, both ones that work and ones that don't.
type formatting disp ------------- doctest('formatting', 'CreateLinks', 0)
% formatting examples % % >> 1 + 1 % should work fine % % ans = % % 2 % % >> 1 + 1 % comparisons collapse all whitespace, so this passes % ans = 2 % % >> 1 + 1; % expects no output, since >> is on the next line % >> for I = 1:3 % FAILS: code to run can only be one line long % disp(I) % end % 1 % % 2 % % 3 % % >> for I = 1:3; disp(I); end % but this works % 1 % % 2 % % 3 % % >> 1 + 4 % FAILS: there aren't 2 blank lines before the prose % % ans = % % 5 % % Blah blah blah oops! This prose started too soon! % % % Sometimes you have output that changes each time you run a function % >> dicomuid % FAILS: no wildcard on changing output % % ans = % % 188.8.131.52.4.1.95184.108.40.206.944807727511025110.343357080818013 % % % You can use *** as a wildcard to match this! % >> dicomuid % passes % % ans = % % 220.127.116.11.4.1.*** % % % I guess that's it! ------------- TAP version 13 1..8 ok 1 - "1 + 1 % should work fine" ok 2 - "1 + 1 % comparisons collapse all whitespace, so this passes" ok 3 - "1 + 1; % expects no output, since >> is on the next line" not ok 4 - "for I = 1:3 % FAILS: code to run can only be one line long" expected: disp(I) end 1 2 3 got : ??? Error: At least one END is missing: the statement may begin here. ok 5 - "for I = 1:3; disp(I); end % but this works" not ok 6 - "1 + 4 % FAILS: there aren't 2 blank lines before the prose" expected: ans = 5 Blah blah blah oops! This prose started too soon! got : ans = 5 not ok 7 - "dicomuid % FAILS: no wildcard on changing output" expected: ans = 18.104.22.168.4.1.9522.214.171.124.944807727511025110.343357080818013 got : ans = 126.96.36.199.4.1.95188.8.131.52.138027182622631308123079093120036407030 ok 8 - "dicomuid % passes"
Expecting an error
doctest can deal with errors, a little bit. You might want this to test that your function correctly detects that it is being given invalid parameters. But if your example will emit other output BEFORE the error message, the current version can't deal with that. For more info see Issue #4 on the bitbucket site (below). Warnings are different from errors, and they work fine.
type errors disp ------------- doctest('errors', 'CreateLinks', 0)
% Errors and doctest - demonstrates a current limitation of doctest % % This one works fine. % % >> not_a_real_function(42) % ??? Undefined function or method 'not_a_real_function' for input % arguments of type 'double'. % % % This one breaks. % % >> disp('if at first you don''t succeed...'); error('nevermind') % if at first you don't succeed... % ??? nevermind ------------- TAP version 13 1..2 ok 1 - "not_a_real_function(42)" not ok 2 - "disp('if at first you don''t succeed...'); error('nevermind')" expected: if at first you don't succeed... ??? nevermind got : ??? nevermind
All adjascent white space is collapsed into a single space before comparison, so right now doctest can't detect a failure that's purely a whitespace difference.
It can't run examples that are longer than one line of code (so, for example, no loops that take more than one line). This is difficult because I haven't found a good way to mark these subsequent lines as part-of-the-source-code rather than part-of-the-result. However, variables that you define in one line do carry over to the next.
I haven't found a good way of isolating the variables that you define in the tests from the variables used to run the test. So, don't run CLEAR in your doctest, and don't expect WHO/WHOS to work right, and don't mess with any variables that start with doctest_. :-/
When you're working on writing/debugging a Matlab class, you might need to run 'clear classes' to get correct results from doctests (this is a general problem with developing classes in Matlab).
The latest version from the original author, Thomas Smith, is available at http://bitbucket.org/tgs/doctest-for-matlab/src
The bugtracker is also there, let me know if you encounter any problems!
Published with MATLAB® 7.10