replaying a built-in returned a different result than expected

[NOMORECOFFEE]

I got this error message while running concurerror

veselov@poisk-001:/media/data/Concuerror$ ./concuerror -m send_order
Concuerror started at 08 Dec 2015 13:11:11
Writing results in concuerror_report.txt

Info: Instrumented io_lib
Info: Instrumented send_order
Info: Instrumented crypto
Info: Instrumented erlang
Info: You can see pairs of racing instructions (in the report and --graph) with '--show_races true'
Error: On step 1, replaying a built-in returned a different result than expected:
  original:
    {event,<0.53.0>,
       {builtin_event,<0.53.0>,undefined,false,
                      {erlang,spawn,
                              [erlang,apply,[#Fun<send_order.1.58635779>,[]]]},
                      <0.63.0>,ok,false},
       #Ref<0.0.0.133>,
       [2318,{file,"erlang.erl"}],
       [{new,<0.63.0>}]}
  new:
    {event,<0.53.0>,
       {builtin_event,<0.53.0>,undefined,false,
                      {erlang,spawn,
                              [erlang,apply,[#Fun<send_order.1.58635779>,[]]]},
                      <0.63.0>,ok,false},
       #Ref<0.0.0.133>,
       [2318,{file,"erlang.erl"}],
       [{new,<0.63.0>}]}

Please notify the developers, as this is a bug of Concuerror.

Done! (Exit status: error)
  Summary: 1 errors, 2/128 interleavings explored

-module(send_order).
-export([test/0]).

test() ->
    Orig = [I || <<I>> <= crypto:rand_bytes(128)],
    Master = self(),
    [spawn(fun() -> Master ! I end) || I <- Orig],
    Orig = [receive I -> I end || _ <- Orig].

[aronisstav]

Thanks for the bug report!

The problem here is the call to crypto:rand_bytes/1, which ultimately results to a call to a NIF, whose result is impossible to control just by instrumenting the Erlang source code. Concuerror requires this level of control, in order to be able to ensure that it can reach the 'same state' as in a previous scheduling and only afterwards have different things happening.

Concuerror should of course report this limitation and not crash. I will write a patch for this, but it will not really make the particular test work, just fail in a better way.

[NOMORECOFFEE]

Thanks @aronisstav !

this is test with crypto:rand_bytes/1 work

-module(sleep_sort).
-export([test/0]).

test() ->
    Orig = [I || <<I>> <= crypto:rand_bytes(128)],
    Master = self(),
    Sorted = lists:sort(Orig),
    [spawn(fun() -> timer:sleep(I), Master ! I end) || I <- Orig],
    Sorted = [receive I -> I end || _ <- Orig].

[aronisstav]

Indeed, but it shouldn't. It should have the same problem.