mirror of
https://github.com/barrucadu/dejafu.git
synced 2024-11-23 06:02:52 +03:00
cb118e4f41
It's wrong to use initialDepState when there is a setup action, as the
action could end with a DepState which is not the same as the initial
one. Here's an example of it going wrong:
> :{
resultsSet defaultWay defaultMemType $ do
v <- newMVar ()
fork (takeMVar v)
readMVar v
:}
fromList [Left Deadlock,Right ()]
> :{
resultsSet defaultWay defaultMemType $
withSetup (newMVar ()) $ \v -> do
fork (takeMVar v)
readMVar v
:}
fromList [Right ()]
This PR pushes responsibility for the DepState into the Context, and
the DepState is passed to all schedulers. That means it's been
promoted from an internal type to a user-facing one, so I gave it the
more generic name "ConcurrencyState". Furthermore, the snapshotted
DepState is passed to all the DPOR functions, and the trace
simplification functions.
initialDepState is now only used:
- in Conc.Internal to initialise a new context
- in SCT.Internal when there is no snapshot
|
||
|---|---|---|
| .. | ||
| Examples | ||
| Integration | ||
| Test/Tasty | ||
| Unit | ||
| Common.hs | ||
| Examples.hs | ||
| Integration.hs | ||
| QSemN.hs | ||
| Unit.hs | ||