Michael Walker
3e3385d0be
Improve POR for relaxed memory
...
- Refine notion of dependent actions tot ake the memory model into
account.
- Don't consider context switches for commits as viable candidates
for conservative backtracking points.
2015-10-05 16:17:03 +01:00
Michael Walker
4cc5f5c921
Tidy up stepWrite/stepCommit
2015-10-05 16:16:43 +01:00
Michael Walker
0c94c16d16
Move dejafus{,IO}' into different documentation section
2015-10-02 11:27:15 +01:00
Michael Walker
9ce7f51b92
Document default memory model
2015-10-01 16:40:31 +01:00
Michael Walker
a272a0199b
Update documentation
2015-10-01 15:46:50 +01:00
Michael Walker
eb7c663993
Bit of refactoring in Test.DejaFu.Deterministic
2015-10-01 15:11:49 +01:00
Michael Walker
415e238f20
Nicely display commits in traces to users
2015-10-01 15:10:48 +01:00
Michael Walker
615535b49b
Remove _concNoTest
...
It's not clear how it can be safely implemented with the possibility
for relaxed memory effects to occur.
2015-10-01 14:32:36 +01:00
Michael Walker
10a8c5f559
Remove old TODO
2015-10-01 14:23:11 +01:00
Michael Walker
842460daf8
Fix phantom thread numbering at 0
2015-10-01 12:11:33 +01:00
Michael Walker
2529a9e95c
Expose memory model choice in testing functions
2015-10-01 01:23:51 +01:00
Michael Walker
de75c17c0f
Implement phantom threads.
2015-09-30 21:48:18 +01:00
Michael Walker
39ff237422
Expose non-synchronised write primitive
2015-09-30 20:58:52 +01:00
Michael Walker
c51ed65f69
Implement separate write and commit primitives
2015-09-30 20:52:15 +01:00
Michael Walker
24b7cd00c1
Separate writing and committing CRef modifications.
...
This is NOT yet exposed in any interesting way to the user. A separate
commit primitive is needed for that.
2015-09-30 18:09:23 +01:00
Michael Walker
20a54e95da
Suppress some more hlint warnings
2015-09-30 16:55:34 +01:00
Michael Walker
dbc1126466
Parameterise conc runner with memory model to use.
...
Three memory models will eventually be supported:
- SequentialConsistency, which is the current behaviour.
- TotalStoreOrder, where each thread has a write buffer.
- PartialStoreOrder, where each CRef has a write buffer.
The parameter is currently ignored.
2015-09-30 16:51:36 +01:00
Michael Walker
c107f2e6b1
Redefine writeCRef to suffer from relaxed memory issues, add atomicWriteCRef
2015-09-28 15:39:18 +01:00
Michael Walker
c55e0249b3
Release 0.1.0.0
2015-08-27 13:37:20 +01:00
Michael Walker
819cbaebf6
Add Thief of Time quote
2015-08-27 13:33:48 +01:00
Michael Walker
e767326660
Clean up warnings in 7.10
2015-08-27 01:05:59 +01:00
Michael Walker
4779ac07e5
Improve documentation and instances
2015-08-27 00:41:53 +01:00
Michael Walker
f8056f13b9
Add efficient Foldable.toList for NonEmpty when GHC >= 7.10
2015-08-27 00:18:48 +01:00
Michael Walker
5ff263d96f
Rename ThreadAction' to Lookahead
2015-08-26 22:48:38 +01:00
Michael Walker
a1144610b2
General tidying ahead of release
2015-08-26 11:56:26 +01:00
Michael Walker
55a6239503
Update build status link in README
2015-08-16 00:19:05 +01:00
Michael Walker
046847c714
Upgrade to Stackage LTS 3.0
2015-08-15 23:45:22 +01:00
Michael Walker
3544ad08c3
Fix compiler warnings
2015-08-14 15:50:11 +01:00
Michael Walker
15fb0c391b
Drop the default implementations of putCVar and takeCVar
2015-08-14 15:27:50 +01:00
Michael Walker
faa0febdba
Fix some inconsistencies in class docs
2015-08-14 15:26:43 +01:00
Michael Walker
4a76ecedcc
Add Applicative constraint to classes for pre-AMP GHC
2015-08-14 15:25:06 +01:00
Michael Walker
0e558b271f
Handle exceptions in CVar functions as in MVar functions
2015-08-14 15:06:03 +01:00
Michael Walker
dea286a582
Add Functor context to sctBoundedM for pre-AMP GHC
2015-08-10 10:37:42 +01:00
Michael Walker
d32ee150bd
Reduce code duplication with Identity monad
2015-08-09 21:00:11 +01:00
Michael Walker
a9f25fbb71
Update README
2015-08-01 15:20:20 +01:00
Michael Walker
79ed016dda
Build with LTS by default
2015-08-01 15:20:20 +01:00
Michael Walker
ac12036bad
Use Sequence for backtracking points, rather than ++ a lot
2015-07-24 19:57:46 +01:00
Michael Walker
37d25048d3
Improve memory usage of pbBacktrack
2015-07-24 17:44:48 +01:00
Michael Walker
7aa7f5cc80
Use 'maximumBy' rather than 'sortBy' in 'next'.
...
Also swap the order of the append in 'prefixes'. For some reason this makes it
work much better in the test cases. No idea why.
2015-07-24 17:25:28 +01:00
Michael Walker
929eceab95
Inline ThreadId information into BacktrackStep
2015-07-24 16:35:23 +01:00
Michael Walker
a6055a30a0
Use strict IntMaps where possible.
2015-07-24 15:47:24 +01:00
Michael Walker
9d2c3cd42e
Comment the tricky bits in Test.DejaFu.SCT more
2015-07-21 17:28:28 +01:00
Michael Walker
5649c4d2a7
Use Set instead of [] where it makes sense.
...
A lot of the places where lists were used, there was an assumed invariant that
there would be no duplicate entries, for some criteria of equality. In some
cases this was enforced by `nub`, in others not. Sets are a much better choice
here, as they actually enforce the invariant, with better complexity for some
operations.
2015-07-21 16:09:47 +01:00
Michael Walker
112a7cd138
Add some instances for NonEmpty
2015-07-21 16:09:11 +01:00
Michael Walker
c786a30448
Fix accidentally recursive binding
2015-07-21 15:30:29 +01:00
Michael Walker
3c1dadffa6
Take advantage of the fact that tagged is sorted
2015-07-21 14:38:00 +01:00
Michael Walker
d4b1ea8bc5
Slightly simplify dependent/dependent' defns
2015-07-21 14:37:36 +01:00
Michael Walker
15f79088c9
Correctly sort prefixes in next
2015-07-21 14:17:16 +01:00
Michael Walker
148cd0a351
Gradually accumulate allThreads in findBacktrack.
...
Recomputing it every single time is a waste of effort.
2015-07-21 14:16:34 +01:00
Michael Walker
d498ebd355
Extend blocking lookahead.
...
Adds more information to traces with a new Trace' type, which includes the
*sequence* of actions a thread will perform next, and use that for blocking
lookahead. This allows skipping over things like `_concKnowsAbout`, and so
brings the analysis behaviour of `spawn` in-line with `fork`.
For the test cases, this further reduces the average number of runs to 23.
2015-07-20 18:43:37 +01:00
