Frequently Asked Questions
- How do I report a bug?
- How can I tell Concuerror where to find the code used by my test?
- Will the exploration ever finish?
- Can Concuerror do ‘random testing’?
- How does Concuerror work? (extended)
- How does Concuerror control the scheduling of processes?
- Does Concuerror support all of Erlang’s operations/libraries?
- How can I get rid of ‘…’ in output files?
- Is there a mailing list?
- Is there a changelog?
How do I report a bug?
How can I tell Concuerror where to find the code used by my test?
Concuerror automatically instruments and reloads any code used by your test transparently relying on Erlang’s code path.
It also supports a number of familiar options to specify input files.
concuerror -h input for more info.
Will the exploration ever finish?
Complex concurrent programs can have a LOT of schedulings! Concuerror uses a technique called Optimal Dynamic Partial Order Reduction to filter through them and has special knowledge about the interferences of Erlang/OTP built-ins (see the relevant Publications). Still, all this reduction power may sometimes not be enough.
If Concuerror keeps running for a while, you may want to limit the exploration
(using e.g. the
and visualize the explored schedulings
--graph option (the dot tool
is needed to produce an image).
You can then see which operations the tool considers as racing
--show_races option and possibly simplify the test.
Concuerror may also print tips during its execution,
suggesting ways to improve the effectiveness of testing.
A large number of options are available to fine tune the tool.
You can find out more about them by using
Can Concuerror do ‘random testing’?
A ‘random testing’ mode has not yet been implemented in Concuerror (add your voice to the related Issue page).
If your goal is not verification, Concuerror has its own way to sample
the interleavings in a less systematic way: schedule bounding. You can
read more about it by
How does Concuerror work? (extended)
Concuerror runs the program under test in a controlled way so that only one process runs at a time.
- It begins with an arbitrary initial run, during which it logs any operations that affect shared state, such as sending and delivery of messages, operations on ETS tables etc. The result is a sequence of such events, called an interleaving.
- Afterwards it analyses this interleaving, finding pairs of events that could have different results if they were to happen in the reverse order (e.g. a message being delivered before the receiving process could have executed an after clause instead).
- Then it “plans” new interleavings of the program that will force this reverse order for each such pair. Any such interleavings “replay” some events from the one currently analyzed up to some instruction and then diverge, in order to execute other events before those involved in a race.
- Finally it checks for the “latest” place where it has made such a “plan”, actually replays all events up to that point and continues as described in the plan. At some point in this new interleaving new behaviours may emerge.
- It repeats this approach from step 2, until no other such plans remain.
This is a technique known as stateless model checking with dynamic partial order reduction (read more).
How does Concuerror control the scheduling of processes?
Concuerror automatically adds instrumentation code and reloads any module that is used by the test. The instrumentation forces any process involved in the test to stop and report any operation that affects shared state.
Does Concuerror support all of Erlang’s operations/libraries?
Concuerror supports the complete Erlang language and can instrument programs of any size. It can be the case that support for some built-in operations is not complete but this is a matter of prioritization of tasks. There are however certain limitations regarding e.g. timeouts and non-deterministic functions.
How can I get rid of ‘…’ in output files?
Use a higher
Is there a mailing list?
Is there a changelog?
How does Concuerror handle timeouts and other time-related functions?
Timeouts may appear as part of an Erlang
statement or calls to
to the fact that Concuerror’s instrumentation has an overhead on the execution
time of the program, Concuerror normally disregards the actual timeout values
afterclause is always assumed to be possible to reach. Concuerror will explore interleavings that trigger the
afterclause, unless it is impossible for a matching message to not have arrived by the time the
receivestatement is reached.
send_after-like timeouts: The timeout message may arrive at any time until canceled.
You can use
-- after-timeout N to make Concuerror regard timeouts higher than
N as infinity.
Time-related functions (E.g.
Concuerror handles such functions together with other non-deterministic functions.
How does Concuerror handle non-deterministic functions?
The first time Concuerror encounters a non-deterministic function it records the returned result. In every subsequent interleaving that is supposed to ‘branch-out’ after the point where the non-deterministic function was called, the recorded result will be supplied, to ensure consistent exploration of the interleavings.
This may result in unexpected effects, as for example measuring elapsed time by
erlang:time/0 twice may use a recorded result for the first call and
an updated result for the second call (if it after the branching-out point),
with the difference between the two increasing in every interleaving.
How does Concuerror handle NIFs?
Concuerror cannot intercept, instrument or perform race analysis in modules using NIFs and is very likely to crash if such modules are used. Better support may be included in the future.