Use triq instead of proper #80
Conversation
|
I'll have to read more, but first, the whole rebar.config.script you got rid of wasn't for licensing issues. It was so that downstream users don't have to depend on PropEr. You'll most definitely have to undo that removal if triq is running in the many gigabyte range for tests. And as I said, I'll have to read more, but many gigabytes for a few tests seems odd to say the least. If we're going to change things what I'd rather switch to is just a simpler fuzzing library that we can use to generate random JSON. |
|
Ah, its just a reimplementation. In that case I bet triq is just doing something different for the JSON generator: https://github.com/davisp/jiffy/blob/master/test/jiffy_11_proper_tests.erl#L94-L129 I never fully groked how that recursive data generation is supposed to work so I just tweaked parameters long enough to generate "interesting" tests withouth blowing things up. I'm guessing its probably just a bug in the JSON generator definition. |
|
Just running 10 tests would mostly defeat the whole purpose of using Proper/Triq in the first place. |
|
@eproxus more tests will cause OOM even for 8GB host. Triq doesn't support |
|
That sounds extremely excessive. Some careful usage of the Just curious, what are the licensing issues with Proper? |
|
@eproxus had no luck with, but may be I should try another shot. Problem with licensing is that you cannot include GPL stuff into release tarball in order to run tests on end host while your product is not GPL. |
|
@kxepal Yeah, found the Proper issue: proper-testing/proper#29 |
|
Closing this one out as I've swapped PropEr for EQC and just detect when its present or not. |
|
Good choice as well, thanks! |
This solves license issues caused by proper while tests still does their job right.
Note that amount of tests is dropped from 1000 to 10. This is because triq generates much more heavy and comprehensive cases than proper on each cycle. As the result, it requires much more memory to run the same amount of tests as proper has. I wasn't able to handle 100 tests with 8GB RAM without falling into swap. Starting from 30 you'll need ~4GB and you may not fit in 300 seconds limit. As for 10 the peak memory usage is somewhere around 200MB.