dejafu

Systematic testing for Haskell concurrency.

https://github.com/barrucadu/dejafu

Version on this page:0.9.0.0
LTS Haskell 22.13:2.4.0.5
Stackage Nightly 2024-03-14:2.4.0.5
Latest on Hackage:2.4.0.5

See all snapshots dejafu appears in

MIT licensed by Michael Walker
Maintained by [email protected]
This version can be pinned in stack with:dejafu-0.9.0.0@sha256:b5c62549e0de6e39ab3b6d8ebf78641b0ea7e22e8b5e462be4b6aeb7dccc9994,3085

dejafu

[Déjà Fu is] A martial art in which the user’s limbs move in time as well as space, […] It is best described as “the feeling that you have been kicked in the head this way before”

– Terry Pratchett, Thief of Time

Concurrency is nice, deadlocks and race conditions not so much. The Par monad family, as defined in abstract-par provides deterministic parallelism, but sometimes we can tolerate a bit of nondeterminism.

This package builds on the concurrency package (also in this repository) by enabling you to systematically and deterministically test your concurrent programs.

The documentation of the latest developmental version is available online. Examples can be found in the test suite.

Note on the test suite: This is in a separate project (dejafu-tests) because Cabal-the-library is a bit naff. See this issue.

Déjà Fu and IO

The core assumption underlying Déjà Fu is that any apparent nondeterminism arises purely from the scheduling behaviour. To put it another way, a given computation, parametrised with a fixed set of scheduling decisions, is deterministic.

Whilst this assumption may not hold in general when IO is involved, you should strive to produce test cases where it does.

Memory Model

The testing functionality supports a few different memory models, for computations which use non-synchronised CRef operations. The supported models are:

  • Sequential Consistency: A program behaves as a simple interleaving of the actions in different threads. When a CRef is written to, that write is immediately visible to all threads.

  • Total Store Order (TSO): Each thread has a write buffer. A thread sees its writes immediately, but other threads will only see writes when they are committed, which may happen later. Writes are committed in the same order that they are created.

  • Partial Store Order (PSO): Each CRef has a write buffer. A thread sees its writes immediately, but other threads will only see writes when they are committed, which may happen later. Writes to different CRefs are not necessarily committed in the same order that they are created.

If a testing function does not take the memory model as a parameter, it uses TSO.

Contributing

Bug reports, pull requests, and comments are very welcome!

Feel free to contact me on GitHub, through IRC (#haskell on freenode), or email ([email protected]).

Changes

Release Notes

All notable changes to this project will be documented in this file.

This project is versioned according to the Package Versioning Policy, the de facto standard Haskell versioning scheme.

0.9.0.0

Test.DejaFu.Common

  • New isInternalError, isAbort, isDeadlock, isUncaughtException, and isIllegalSubconcurrency functions for matching failure types. Also exported from Test.DejaFu.

  • The UncaughtException Failure constructor now includes the exception.

    The Read, Enum, and Bounded instances are gone. The Eq, Ord, and NFData instances use the show of the exception. Pretty-printed failures include the exception text.

  • New ThreadDelay and WillThreadDelay constructors in ThreadAction and Lookahead. Uses of threadDelay are no longer reported as a use of yield.


0.8.0.0

Test.DejaFu.Common

  • Execution traces now only include a single item of lookahead (#120).
  • STM traces now include IDs of created TVars (#80).

Test.DejaFu.Schedule

  • Schedulers no longer take the execution trace so far (#106).
  • The Scheduler type is now a newtype (#122).

0.7.3.0

Test.DejaFu.Common

  • A new function threadNames, to get all named threads from a trace.

Miscellaneous

  • Escaping a mask by raising an exception now correctly restores the masking state (#118).
  • Named threads which are only started by a pre-emption now show up in the trace (#101).

0.7.2.0

Test.DejaFu.STM

  • The STM n r monad now has Alternative and MonadPlus instances, using orElse for the binary operation and retry for the unit.

Miscellaneous

  • The Eq instance for ThreadId, CRefId, MVarId, and TVarId now only compares the numbers, not the names.

    This makes it consistent with the Ord instances, and is also a small performance gain.

  • Now compatible with concurrency-1.2.0.0.


0.7.1.3

Miscellaneous

  • Aborted STM transactions are now rolled back correctly (issue #111).
  • Slightly improved run-time of systematic testing.

0.7.1.2

Miscellaneous

  • Errors thrown with Control.Monad.fail no longer terminate testing, and are now correctly treated as asynchronous exceptions.

0.7.1.1

Miscellaneous

  • Significantly reduced memory usage in systematic testing when discarding traces.

    Previously this was O(max trace length * number of executions)

    Now it’s O(max trace length + total size of traces kept)


0.7.1.0

Test.DejaFu

  • Exposed the new SCT discard functions through dejafuDiscard and dejafuDiscardIO.

    There are no dejafusDiscard and dejafusDiscardIO functions because this would probably be confusing, as the traces are shared.

  • The Discard type and defaultDiscard function are also exposed.

Test.DejaFu.Defaults

  • Added a new defaultDiscarder function, which discards nothing.

Test.DejaFu.SCT

  • Added new SCT functions to selectively discard results or traces, which can be a significant memory saving if you know what sorts of results you are interested in:
    • New type: Discard.
    • New functions: runSCTDiscard, resultsSetDiscard, sctBoundDiscard, sctUniformRandomDiscard, and sctWeightedRandomDiscard.
    • resultsSet and resultsSet' now discard traces as they are produced, rather than all at the end, greatly improving performance when traces are large.

0.7.0.2

Test.DejaFu.Refinement

  • Removed unnecessary typeclass constraints from check, check', checkFor, and counterExamples.

Miscellaneous


0.7.0.1

Test.DejaFu.Refinement

  • check, check', and checkFor are now faster if there are multiple counterexamples.
  • The above and counterExamples are now faster even if there is only a single counterexample in some cases.

0.7.0.0

Test.DejaFu

  • The new Test.DejaFu.Defaults and Test.DejaFu.Refinement modules are re-exported.
  • The new smart constructors from Test.DejaFu.SCT are exported.

Test.DejaFu.Defaults

  • The default* values are now defined in the new Test.DejaFu.Defaults module. There is no breaking API change as they are re-exported from Test.DejaFu.

Test.DejaFu.Refinement

  • A new module for checking observational properties of the side-effects of stateful, concurrent functions.

    This is related to my work on CoCo, allowing dejafu to test what CoCo discovers.

Test.DejaFu.SCT

  • The Way type is now abstract and exposes smart constructor functions:
    • systematically, corresponding to the old Systematically.
    • randomly, corresponding to the old Randomly,
    • uniformly, a new uniform random (as opposed to weighted random) scheduler.
    • swarmy, corresponding to the old Randomly and specifying how many executions to use the same weights for.
  • A new sctUniformRandom function to do uniform (non-weighted) scheduling.
  • The sctRandom function is now called sctWeightedRandom and can now re-use the same weights for multiple executions.
  • The sctPreBound, sctFairBound, and sctLengthBound functions have been removed.

Fixed

  • An issue where subconcurrency would re-use MVar IDs, leading to false reports of deadlock on occasion (issue #81).

0.6.0.0

Test.DejaFu.Conc

  • The Conc n r a type is now ConcT r n a, and has been given a MonadTrans instance. Uses of lift appear in the execution trace in the same way as liftBase and liftIO.
    • The ConcIO and ConcST aliases have been updated, so this should be an invisible change to most users.

Test.DejaFu.SCT

  • Way is now a GADT, no longer taking a type parameter. This greatly improves type inference when the Systematically constructor is used.
    • The NFData instance for Way is now gone. The alternative was requiring that any RandomGen used also implement NFData, which is very restrictive

Miscellaneous

  • There is now a changelog.
  • Test.DejaFu.Common is now considered to form part of the public API of the library.
  • Every definition and instance now has a Haddock “@since” annotation.

0.5.1.3

Miscellaneous

  • The version range on the concurrency package has been changed to 1.1.*.

0.5.1.2

This version was misnumbered! It should have caused a minor version bump!

Test.DejaFu.Conc

  • New MonadRef and MonadAtomicRef instances for the Conc type using CRef.

Fixed

  • A long-standing bug where if the main thread is killed with a throwTo, the throwing neither appears in the trace nor correctly terminates the execution.

Miscellaneous

  • The maximum supported version of the concurrency package has been changed to 1.1.1.*.

0.5.1.1

Fixed

  • The correct scheduler state is now passed to the scheduler immediately after the termination of a subconcurrency action.
  • SCT of subconcurrency no longer loops infinitely.

0.5.1.0

Test.DejaFu

  • A new NFData instance for Result.

Test.DejaFu.Common

  • New instances:
    • NFData for ThreadId, CRefId, MVarId, TVarId, IdSource, ThreadAction, Lookahead, ActionType, TAction, Decision, Failure, and MemType.
    • Eq, Ord, and Show instances for IdSource.

Test.DejaFu.SCT

  • New NFData instances for Way, Bounds, PreemptionBound, FairBound, and LengthBound.
  • New strict variants of runSCT and resultsSet: runSCT' and resultsSet'.

Test.DejaFu.STM

  • A new NFData instance for Result.

0.5.0.2

This version was misnumbered! It should have caused a major version bump!

Test.DejaFu.Common

  • A new StopSubconcurrency constructor of ThreadAction.

Changed

  • A StopConcurrency action appears in the execution trace immediately after the end of a subconcurrency action (much like the PopCatching and ResetMasking actions which appear after a catch and mask).
  • A subconcurrency action now inherits the number of capabilities from the outer computation, rather than being reset to 2 as before.

Miscellaneous

  • Test.DejaFu.SCT now compiles with MonoLocalBinds enabled (implied by GADTs and TypeFamilies), which may be relevant to hackers.

0.5.0.1

  • Date 2017-02-21
  • Git tag dejafu-0.5.0.1
  • This version was never pushed to hackage, whoops!

Fixed

  • readMVar is once again considered a “release action” for the purposes of fair-bounding.

0.5.0.0

Test.DejaFu

  • All the functions which did take a Bounds now take a Way instead and support random scheduling as well.

Test.DejaFu.Common

  • New Eq instances for ThreadAction and Lookahead.
  • A TryReadMVar constructor for ThreadAction and a corresponding WillTryReadMVar constructor for Lookahead.

Test.DejaFu.Conc

  • A new testing-only subconcurrency function, to run a concurrent action and do something with its result in the same concurrent context, even if it fails.

Test.DejaFu.SCT

  • An sctRandom function to run a fixed number of randomly-scheduled executions of a program.
  • The Way type, to abstract over how to run a concurrent program, used by new functions runSCT and resultsSet.

Fixed

  • Some previously-missed CRef action dependencies are no longer missed.

Miscellaneous

  • The supported version of the concurrency package was bumped to 1.1.0.0, introducing tryReadMVar.
  • A bunch of things were called “Var” or “Ref”, these are now consistently “MVar” and “CRef”.
  • Significant performance improvements in both time and space.
  • The dpor package has been merged back into this, as it turned out not to be very generally useful. There is no direct replacement, but I have no intent to update it, so the dpor package is now deprecated.

0.4.0.0

Test.DejaFu

  • The autocheck' function now takes the schedule bounds as a parameter.
  • New runTestM and runTestM' functions, monad-polymorphic variants of the now-removed runTestIO and runTestIO' functions.

Test.DejaFu.Conc

  • The Conc type no longer has the STM type as a parameter.
  • A new runConcurrent function, a monad-polymorphic version of the now-removed runConcST and runConcIO functions.

Test.DejaFu.SCT

  • The ST-specific functions are now monad-polymorphic.
  • The IO function variants have been removed.

Test.DejaFu.STM

  • A new runTransaction function, a monad-polymorphic version of the now-removed runTransactionST and runTransactionIO functions.

Changed

  • The termination of the main thread in execution traces now appears as a single Stop, rather than the sequence Lift, Stop.
  • Execution traces printed by the helpful functions in Test.DejaFu now include a key of thread names.

Miscellaneous

  • Remodularisation:
    • The Control.* modules have all been split out into a separate “concurrency” package.
    • Many definitions from other modules have been moved to the new Test.DejaFu.Common module.
    • The Test.DejaFu.Deterministic module has been renamed to Test.DejaFu.Conc

0.3.2.1

Fixed

  • The implementation of the STM orElse for STMLike incorrectly handled some state non-associatively, leading to false deadlocks being reported in some cases.

0.3.2.0

Builds with both dpor-0.1 and dpor-0.2, however some improvements require dpor-0.2.

Fixed

  • (faster with dpor-0.2) Executions missed due to daemon threads with uninteresting first actions are no longer missed.

Changed

  • (requires dpor-0.2) Significantly improved dependency inference of exceptions, greatly improving performance of testcases using exceptions.
  • Significantly improved dependency inference of STM transactions, greatly improving performance of testcases using STM.

0.3.1.1

Miscellaneous

  • Now supports GHC 8.

0.3.1.0

Fixed

  • Context switches around relaxed memory commit actions could cause the number of pre-emptions in an execution to be miscounted, leading to the pre-emption bounding being too lenient.

0.3.0.0

The minimum supported version of GHC is now 7.10.

I didn’t write proper release notes, and this is so far back I don’t really care to dig through the logs.


0.2.0.0

I didn’t write proper release notes, and this is so far back I don’t really care to dig through the logs.


0.1.0.0

Initial release. Go read the API docs.