- update copyright date
- point to new Haskell.org downloads rather than HP
- new WiX version
- smoke test subsumes `:prove True`
- repo layout simpler (no `notebook`, `sbv` subdirs)
- no notebook documentation
Closes#112.
This adds a check at REPL startup that `cvc4` is an executable on the
path. It doesn't check for versions, as we're mostly stuck with
"prerelease" versions.
I made it easy to add more smoke tests at startup; we should add these
as we think of them.
We previously read a batch script from `$HOME/.cryptol` on REPL startup,
but this turns out to conflict with the directory returned by
`System.Directory.getAppUserDataDirectory`.
This commit changes the name to `.cryptolrc` and updates the interpreter
flags and such accordingly.
This involved:
- Moving a couple REPL modules into the Cryptol library hierarchy (those
that don't depend on console libraries)
- Splitting up the Makefile, which unfortunately resulted in a lot of
not-quite-duplication between the two Makefiles. Let's look into
better abstraction...
- Introduce monad type SpecT m a = StateT SpecCache (ModuleT m) a
(easier to call from code written in ModuleT monad)
- Fix bug where specializing expressions under ETAbs could cause
type variables to `escape` from their scopes.
- Function `withDeclGroups` lets you run arbitrary specializer monad
action within the context of a set of DeclGroups. This is used to
implement specialization of where-expressions, and also for the new
function `specializeDeclGroups`, which specializes a set of DeclGroups
using the monomorphic bindings as the 'root set' to determine what
other specialized bindings to include.
Previously we were just using Prelude's `readFile`, which uses the
system default locale. This meant that people writing Cryptol in other
locales might produce source files that work fine for them but not
others. Now the interpreter sets the default locale to utf8 at startup.
Additionally, the code to catch exceptions from loading modules was too
lazy, allowing exceptions to bring down the whole process when the
module contents were forced outside of the `try`.
We also assumed that any IO exception was from files not being found;
there's now an "Other IO exception" possiblity. Incorrect locales will
trigger this alternative because the actual IOException raised isn't
specific to locale errors.
There is still an issue with the encoding of finitness.
For example, if we know:
a = 1 + min b a
And we know that `b`z is finite, we SHOULD know that `a` is finite too,
but apparently we don't.
1. Preserve order of TODO list, os that we prefer unification variables
2. Stop early when looking for `x = y`.
So if we have 'a = b = c` now we'll get:
a = b
b = c
Before we would get:
a = b
a = c
Revised how we do output for `:sat` and `:prove` without arguments,
making it more clear what properties are being checked in each case.
Also reworded the output of `:check` slightly in the case where the
property has no inputs. It would be nice to make `:check` output more
consistent with the others.
