UTF-8 encode and decode (Discussion)

Can you create a PR to add them to the tests in src/Tests/library/? The more tests the better!

1 Like

QuickCheck may be useful here. Using Logtalk 3.35.0 with SWI-Prolog 8.1.22 to test utf8_codes//1 on macOS 10.14.6:

?- {lgtunit(loader)}.
...
% (0 warnings)

true.

?- [user].
|: property(Atom) :-
|:     atom_codes(Atom, Codes),
|:     phrase(utf8_codes(Codes), UTF8),
|:     phrase(utf8_codes(Codes2), UTF8),
|:     Codes == Codes2.
|: ^D% user://1 compiled 0.01 sec, 1 clauses
true.

?- lgtunit::quick_check(property(+non_empty_atom(unicode_bmp)), [n(1000)]).
% 1000 random tests passed
true.

?- lgtunit::quick_check(property(+non_empty_atom(unicode_full)), [n(1000)]).
% 1000 random tests passed
true.

According to Jan observations, the second test is expected to fail when running on Windows, correct? Anyone on Windows that can try the queries above? @EricGT maybe you can try something similar with plunit + quickcheck pack?

Not sure exactly which test you mean?

If you are asking if any Unicode code point greater than U+FFFF will cause an error on Windows with SWI-Prolog when trying to be created as a character or in a string, then that is what I am finding. (ref)

But as I noted in this reply, using SWI-Prolog on Ubuntu on WSL on Windows is working for Unicode code points greater than U+FFFF as far as I can tell, but I literally just started looking into this.

OK, I’ll take a look.

EDIT

I tried to install the package quickcheck but it doesn’t find MinGW for some reason, and I don’t want to spend the day chasing this problem down.

Running this by hand did work.

?- mingw_root(X).
Correct to: "prolog_pack:mingw_root(X)"? yes
X = 'C:/MinGW'.

:slightly_frowning_face:

The one with the unicode_full argument.

P.S. The main reason I suggested QuickCheck was that I noticed your hard-coded test cases (defined using the unicode_string_utf8_test_case_generator/2 and unicode_utf8_test_case_generator/2 predicates).

Glad you noticed.

My view on them at present is they are better than what we had which was nothing, and that they are just a start for checking on Unicode with SWI-Prolog. If your goal is to add more test cases in this area to the master source code then we are on the same path. :slightly_smiling_face:

I’m planning to add some QuickCheck-based tests to the Logtalk distribution that verify Unicode support in backend Prolog systems. Other Prolog systems already run the Logtalk tests suites routinely. SWI-Prolog can do it too as a second line of testing. I do it myself everytime I commit and push to GitHub using the SWI-Prolog stable Docker image. Adding the specific tests I posted to the master source would require, however, solving the odd issue with the quickcheck pack you mentioned and defining the types used (which are parametrized by the char set).

1 Like

That makes sense. Last time I looked at it (some years ago), my complaint was that running the test suite is pretty verbose and it fails on several ISO tests where I deliberately decided not to follow the ISO standard. Going through that once is fine, but for routine testing I just want an all passed or test XYZ failed and otherwise pretty quiet operation. Is that (now) possible?

CMake’s ctest has that pretty neatly solved, were ctest only prints some summary and progress reports for normal operation and the full output for a failing test. It also runs tests concurrently, so I can run the whole suite in 10 seconds elapsed time on a quad core i7 machine. Part of the speed is due to ctest starting multiple Prolog processes where each process runs a fair number of individual tests controlled by plunit.

I’m glad you quickly run the tests on suspicious updates from my side. It did good work for the rational syntax. Otherwise the coverage of the bundled tests seems pretty much ok as your testing rarely finds something.

1 Like

The default output (from the logtalk_tester automation script) is verbose (but made of 100% recycled bits). But you also have TAP and xUnit output options. The xUnit output in particular can be easily converted into HTML (e.g. https://logtalk.org/files/blog/2019_11_06/xunit_report.html which provide a nice way to dig into the test results). Note that all output from lgtunit uses the message printing mechanism and thus easy to customize (that’s how TAP and xUnit outputs are implemented).

You don’t need to run all test suites. E.g. when using TravisCI via GitHub to check Logtalk commits using SWI-Prolog stable Docker image, the Prolog compliance suite is not run (see the https://github.com/LogtalkDotOrg/logtalk3/blob/3b77b374128ee69a4ede0faccf857a490fbb2cb8/.travis.yml file).

One option would be to divert the output of the logtalk_tester automation script (to a file or dev/null) and use its exit status (see https://logtalk.org/man/logtalk_tester.html for the different values and their meanings) to tell if all tests pass or what kind of failure was found and, in the later case, just browse the generated HTML report(s). This is easy to script.

Running Logtalk test suites concurrently is supported. See the https://github.com/LogtalkDotOrg/logtalk3/blob/3b77b374128ee69a4ede0faccf857a490fbb2cb8/library/parallel_logtalk_processes_setup.pl file for the required configuration.

1 Like

In the just released Logtalk 3.36.0 version, you can do:

$ logtalk_tester -o minimal

To run test suites concurrently:

$ logtalk_tester -i '-f /path/to/library/parallel_logtalk_processes_setup.pl'
2 Likes