Elementary support for application_controller

[aronisstav]

No description provided.

[aronisstav]

One possible workaround: instead of starting sth as an application, call its top-level supervisor (or even better a top level gen_server). It focuses the test on the application itself and not on the boilerplate that is invoked by application_controller...

[essen]

Trying to use Concuerror on a real world project and running into issues with calls like application:get_env. Is there a way to make it work? It will be difficult to isolate the code entirely from application logic.

[aronisstav]

Hi Loïc! I can have another go on this issue. It would be great if you could send me some sample code showing what it is that you need supported.

[essen]

I will get back to you on that particular issue. Basically our system is large, we use sys.config etc. so we need applications to be at the very least loaded, if not started. I will try a custom concuerror script that does all the initialization beforehand.

But then I will run into this separate issue (I can open a separate ticket if you want):

Error: Concuerror does not support calls to built-in os:timestamp/0 (found in wpool_pool.erl line 7).

I'm not sure what actually calls it, the stacktrace it gives me is not really correct, that module doesn't even have os:timestamp on the version we use (it does in the current version; we use this one https://github.com/inaka/worker_pool/blob/dc59ce01207582046c70cb3b7f4bede93bf3472e/src/wpool_pool.erl - 1.0.4).

The problem is that we have functions -> worker pool -> mnesia calls -> inbox table, with a race condition when two workers save at the same place in mnesia at the same time (they need to first retrieve the data, and then save it, so there's a case where they both read the same and then one overwrites the other).

It's a simplistic view, but ultimately that's what we want to test, because we really want our data to be there. :-)

[aronisstav]

I just pushed #103 which has patches for both application_controller and os:timestamp/0. It will be merged to master after Travis gives the ok. 😄

For os:timestamp, keep in mind that Concuerror handles "time" in a very limited way: calls to time-related functions are not racing with each other (although that can be added) and they are "reusing" results from previous interleavings, if needed, to ensure that control flow remains the same.

I hope this helps!

[essen]

Sounds good, hopefully I can finish what I'm doing soon and try again. :-)

[aronisstav]

I am planning to close this one, with no fix on Concuerror's side. I am in the process of developing support for mocking (#197) and this will hopefully help in such cases.

[essen]

No worries.