testing macros and fixtures

testing macros and fixtures

[Phillip Lord]

I have been thinking about testing since it was raised in one of the
last gargantuan threads.

ERT is quite nice, but one of the things that I have found lacking is a
nice set of predicates, for use within should.

So, when I wrote test for my "lentic" package I needed some functions
like, so that I could do things like:

(should
  (test-eq-after-this
     "blah-before.txt"
     "blah-after.txt"
     (insert "hello")))

which opens "blah-before.txt" runs (insert "hello") then compares the
result with "blah-after.txt". My version of this also does a diff of the
results if the two are not equal.

I've noticed that "puppet-mode" has some thing similar. For instance:

(puppet-test-with-temp-buffer "# class
bar"
    (should (eq (puppet-test-face-at 1) 'font-lock-comment-delimiter-face))
    (should (eq (puppet-test-face-at 3) 'font-lock-comment-face))
    (should (eq (puppet-test-face-at 7) 'font-lock-comment-face))
    (should (eq (puppet-test-face-at 8) 'font-lock-comment-face))
    (should-not (puppet-test-face-at 9))))

And julia-mode has indentation checking tests like so:

   (julia--should-indent
     "
if foo
    bar
else
baz
end"
     "
if foo
    bar
else
    baz
end"))


My own experience is that these are actually quite hard to right. The
ones in lentic have never worked quite right -- that is, when it all
works they are fine, but restoring state after a crash doesn't always
work. Similarly, checking that, for example, test files are not already
open before a test is run interactively.

So, the point of my question is this; are there any good libraries
providing this kind of fixture logic? A lot of this should surely be
re-usable between different packages.

Phil

Re: testing macros and fixtures

[Kaushal Modi]

[-- Attachment #1: Type: text/plain, Size: 2361 bytes --]

verilog-mode runs similar tests from outside emacs.

From the terminal, the verilog-mode functions are run in emacs batch mode
on files in the tests/ dir and and the results are compared against the
files in tests_ok/ dir (if I understand correctly).

Repo on github which is then commited to emacs core:
https://github.com/veripool/verilog-mode

Perl script to run batch tests:
https://github.com/veripool/verilog-mode/blob/master/batch_test.pl


--
Kaushal Modi

On Mon, Oct 12, 2015 at 5:13 PM, Phillip Lord <phillip.lord@russet.org.uk>
wrote:

>
> I have been thinking about testing since it was raised in one of the
> last gargantuan threads.
>
> ERT is quite nice, but one of the things that I have found lacking is a
> nice set of predicates, for use within should.
>
> So, when I wrote test for my "lentic" package I needed some functions
> like, so that I could do things like:
>
> (should
>   (test-eq-after-this
>      "blah-before.txt"
>      "blah-after.txt"
>      (insert "hello")))
>
> which opens "blah-before.txt" runs (insert "hello") then compares the
> result with "blah-after.txt". My version of this also does a diff of the
> results if the two are not equal.
>
> I've noticed that "puppet-mode" has some thing similar. For instance:
>
> (puppet-test-with-temp-buffer "# class
> bar"
>     (should (eq (puppet-test-face-at 1) 'font-lock-comment-delimiter-face))
>     (should (eq (puppet-test-face-at 3) 'font-lock-comment-face))
>     (should (eq (puppet-test-face-at 7) 'font-lock-comment-face))
>     (should (eq (puppet-test-face-at 8) 'font-lock-comment-face))
>     (should-not (puppet-test-face-at 9))))
>
> And julia-mode has indentation checking tests like so:
>
>    (julia--should-indent
>      "
> if foo
>     bar
> else
> baz
> end"
>      "
> if foo
>     bar
> else
>     baz
> end"))
>
>
> My own experience is that these are actually quite hard to right. The
> ones in lentic have never worked quite right -- that is, when it all
> works they are fine, but restoring state after a crash doesn't always
> work. Similarly, checking that, for example, test files are not already
> open before a test is run interactively.
>
> So, the point of my question is this; are there any good libraries
> providing this kind of fixture logic? A lot of this should surely be
> re-usable between different packages.
>
> Phil
>
>
>
>
>

[-- Attachment #2: Type: text/html, Size: 4019 bytes --]

Re: testing macros and fixtures

[Artur Malabarba]

[-- Attachment #1: Type: text/plain, Size: 2286 bytes --]

2015-10-12 22:13 GMT+01:00 Phillip Lord

> ERT is quite nice, but one of the things that I have found lacking is a
> nice set of predicates, for use within should.
>
> So, when I wrote test for my "lentic" package I needed some functions
> like, so that I could do things like:
>
> (should
>   (test-eq-after-this
>      "blah-before.txt"
>      "blah-after.txt"
>      (insert "hello")))
>
> which opens "blah-before.txt" runs (insert "hello") then compares the
> result with "blah-after.txt".

I think that's a pretty specific use-case. So it's best to let people write
their own macros for this.

(defmacro should-eq-after-body (file-before file-after &rest body)
  `(let ((bef
          (with-temp-buffer
            (insert-file-contents file-before)
            ,@body
            (buffer-string))))
     (should (string= bef (with-temp-buffer
                            (insert-file-contents file-after)
                            (buffer-string))))))

> My version of this also does a diff of the
> results if the two are not equal.

This certainly seems very useful. Maybe ert should do this on *all*
multi-line string comparisons instead of doing its default
“different-strings” report (which, most of the time, just says “strings are
of different length”).

> I've noticed that "puppet-mode" has some thing similar. For instance:
> ...
>
> And julia-mode has indentation checking tests like so:
>
> ...
>
> My own experience is that these are actually quite hard to right. The
> ones in lentic have never worked quite right -- that is, when it all
> works they are fine, but restoring state after a crash doesn't always
> work. Similarly, checking that, for example, test files are not already
> open before a test is run interactively.

Yes, there are many packages that use custom-deisgned temp-buffers for
testing. In fact, most non-trivial packages do. I'd really like to see ert
offer a common interface for this. The difficulty for that is that each
package has very different needs when it comes to testing in buffers, so I
have no idea what this common interface could be.

> So, the point of my question is this; are there any good libraries
> providing this kind of fixture logic?

Don't know.

[-- Attachment #2: Type: text/html, Size: 2823 bytes --]

Re: testing macros and fixtures

[Phillip Lord]

Okay. This is definately something worth looking at, then. I think
writing tests is probably too hard at the moment!

Phil

Kaushal Modi <kaushal.modi@gmail.com> writes:

> verilog-mode runs similar tests from outside emacs.
>
>>From the terminal, the verilog-mode functions are run in emacs batch mode
> on files in the tests/ dir and and the results are compared against the
> files in tests_ok/ dir (if I understand correctly).
>
> Repo on github which is then commited to emacs core:
> https://github.com/veripool/verilog-mode
>
> Perl script to run batch tests:
> https://github.com/veripool/verilog-mode/blob/master/batch_test.pl
>
>
> --
> Kaushal Modi
>
> On Mon, Oct 12, 2015 at 5:13 PM, Phillip Lord <phillip.lord@russet.org.uk>
> wrote:
>
>>
>> I have been thinking about testing since it was raised in one of the
>> last gargantuan threads.
>>
>> ERT is quite nice, but one of the things that I have found lacking is a
>> nice set of predicates, for use within should.
>>
>> So, when I wrote test for my "lentic" package I needed some functions
>> like, so that I could do things like:
>>
>> (should
>>   (test-eq-after-this
>>      "blah-before.txt"
>>      "blah-after.txt"
>>      (insert "hello")))
>>
>> which opens "blah-before.txt" runs (insert "hello") then compares the
>> result with "blah-after.txt". My version of this also does a diff of the
>> results if the two are not equal.
>>
>> I've noticed that "puppet-mode" has some thing similar. For instance:
>>
>> (puppet-test-with-temp-buffer "# class
>> bar"
>>     (should (eq (puppet-test-face-at 1) 'font-lock-comment-delimiter-face))
>>     (should (eq (puppet-test-face-at 3) 'font-lock-comment-face))
>>     (should (eq (puppet-test-face-at 7) 'font-lock-comment-face))
>>     (should (eq (puppet-test-face-at 8) 'font-lock-comment-face))
>>     (should-not (puppet-test-face-at 9))))
>>
>> And julia-mode has indentation checking tests like so:
>>
>>    (julia--should-indent
>>      "
>> if foo
>>     bar
>> else
>> baz
>> end"
>>      "
>> if foo
>>     bar
>> else
>>     baz
>> end"))
>>
>>
>> My own experience is that these are actually quite hard to right. The
>> ones in lentic have never worked quite right -- that is, when it all
>> works they are fine, but restoring state after a crash doesn't always
>> work. Similarly, checking that, for example, test files are not already
>> open before a test is run interactively.
>>
>> So, the point of my question is this; are there any good libraries
>> providing this kind of fixture logic? A lot of this should surely be
>> re-usable between different packages.
>>
>> Phil
>>
>>
>>
>>
>>

Re: testing macros and fixtures

[Phillip Lord]

Artur Malabarba <bruce.connor.am@gmail.com> writes:

> 2015-10-12 22:13 GMT+01:00 Phillip Lord
>
>> ERT is quite nice, but one of the things that I have found lacking is a
>> nice set of predicates, for use within should.
>>
>> So, when I wrote test for my "lentic" package I needed some functions
>> like, so that I could do things like:
>>
>> (should
>>   (test-eq-after-this
>>      "blah-before.txt"
>>      "blah-after.txt"
>>      (insert "hello")))
>>
>> which opens "blah-before.txt" runs (insert "hello") then compares the
>> result with "blah-after.txt".
>
> I think that's a pretty specific use-case. So it's best to let people write
> their own macros for this.
>
> (defmacro should-eq-after-body (file-before file-after &rest body)
>   `(let ((bef
>           (with-temp-buffer
>             (insert-file-contents file-before)
>             ,@body
>             (buffer-string))))
>      (should (string= bef (with-temp-buffer
>                             (insert-file-contents file-after)
>                             (buffer-string))))))

I'd imagine it's pretty common -- assuming that (insert "hello") is
anything at all of course.

>> My version of this also does a diff of the
>> results if the two are not equal.
>
> This certainly seems very useful. Maybe ert should do this on *all*
> multi-line string comparisons instead of doing its default
> “different-strings” report (which, most of the time, just says “strings are
> of different length”).

Yep. I use a function to replace "string=" that looks like this.

(defun lentic-test-equal-loudly (a b)
  "Actually, this just tests equality and shouts if not."
  ;; change this to t to disable noisy printout
  (if lentic-test-quiet
      (string= a b)
    (if (string= a b)
        t
      (message "Results:\n%s\n:Complete\nShouldbe:\n%s\nComplete:" cloned-results cloned-file)
      (let* ((a-buffer
              (generate-new-buffer "a"))
             (b-buffer
              (generate-new-buffer "b"))
             (a-file
              (make-temp-file
               (buffer-name a-buffer)))
             (b-file
              (make-temp-file
               (buffer-name b-buffer))))
        (with-current-buffer
            a-buffer
          (insert a)
          (write-file a-file))
        (with-current-buffer
            b-buffer
          (insert b)
          (write-file b-file))
        (message "diff:%senddiff:"
                 (with-temp-buffer
                   (call-process
                    "diff"
                    nil
                    (current-buffer)
                    nil
                    "-c"
                    a-file
                    b-file)
                   (buffer-string))))
      nil)))


An egregious hack, of course, but it was quick and does what I need.


> Yes, there are many packages that use custom-deisgned temp-buffers for
> testing. In fact, most non-trivial packages do. I'd really like to see ert
> offer a common interface for this. The difficulty for that is that each
> package has very different needs when it comes to testing in buffers, so I
> have no idea what this common interface could be.

I think I'll put this next into my todo list. Maintaining the current
(dreadful) infrastructure that I have for lentic is painful to say the
least, and I think a lot of this should be reusuable.

As you say, there are lots of examples.

Phil