History item is not kept while module compiler is running

[luos]

Hi,

I am trying to use num_calls with a mocked function, but some of the history gets dropped. I identified the problematic part, which is here:

https://github.com/eproxus/meck/blob/master/src/meck_proc.erl#L308-L315

If Reload is set then the history doesn't get saved. Unfortunately for me it's not clear why this is happening, or if it's the intended way of working.

Reproduction Steps

I put a test repo here:
https://github.com/luos/meck-test

The test can be found in here:
https://github.com/luos/meck-test/blob/master/test/mt_reload_test.erl

Basically when an expect is set while something is calling the function some calls will be dropped even though seems to me the expect is working. Seems like a race condition in this compiler handling part?

Expected behavior

I expect that the history contains all calls to the function

Observed behavior

Usually the first call immediately after setting the expect is dropped.

Versions

• Meck version: 0.8.6
• Erlang version: 17.5, 20

[eproxus]

Finally got around to looking at this today. I couldn't reproduce it with the simple test on my (faster?) machine, so I had to create this monster version:

-module(mt_reload_test).
-include_lib("eunit/include/eunit.hrl").

-compile(export_all).

proc_fn() ->
    receive
        {do_call, Arg} ->
            ?debugFmt("[~p] called: ~p~n", [self(), mt_app:test_fn(Arg)])
    end,
    proc_fn().

reload_test() ->
    Proc = spawn(fun proc_fn/0),
    meck:new(mt_app, [passtrough]),
    meck:expect(mt_app, test_fn, fun(Arg) -> {mocked, Arg} end),
    spawn(fun() ->
            lists:map(fun(N) ->
                        timer:sleep(1),
                        Proc ! {do_call, N}
                     end, lists:seq(0, 800))
          end),
    [meck:expect(mt_app, test_fn, fun(Arg) -> {mocked2, Arg} end) || _ <- lists:seq(1, 5)],

    timer:sleep(4500),
    ?debugFmt("History: ~p~n",[meck:history(mt_app)]),
    ?assertEqual(801, meck:num_calls(mt_app, test_fn, '_'))
    .

I'm also not sure why the reload case is there, and why we don't store history while reloading is active (history is historical, so it should be stored regardless I think). By removing it, this test passes and it doesn't seem to break any other tests.

[luos]

Hi,

Thanks for checking and for the fix!

[eproxus]

@luos Thanks for reporting it and providing a base test case, very much appreciated!