Review of chapter 3 complete about liquidhaskell-tutorial

[nikivazou]

@christetreault you can take a look here for 3.2:

http://goto.ucsd.edu:8090/index.html#?demo=permalink%2F1423863481.hs

[christetreault]

Doesn't work for me (liquid haskell 0.3.0.0, liquid-fixpoint 0.2.2.0, ghc 7.8.3, z3 4.4.0).

To be fair, if I copy my code and paste it into your demo page, it type checks.

[nikivazou]

This is weird, it is safe for me with liquidHaskell-3....
In the Context the specified refinement of vec is lost and I do not know why....

[christetreault]

I am using the binaries on hackage, not the current repo. Perhaps something you did in the last few days fixed this?

[nikivazou]

Maybe. Can you please do a make test to see what else is broken in your current setup?

[christetreault]

To be clear, you want me to clone the current liquidhaskell repo and build it with make test?

I'm having some trouble getting the dependencies resolved. It seems some module tasty has a dependency on optparse-applicative == 0.11.*, and cabal install installs version 0.8.1

[ranjitjhala]

OK, I think I know the problem (@gridaphobe : I fear its DIFFCHECK)

@christetreault can you do this:

1. Use exactly the same signature as you have in the screenshot above,
2. Change the body of the function a little bit by adding whitespace e.g. to

unsafeLookup index vec     =     vec     !     index

Can you let me know if that works?

[ranjitjhala]

@christetreault do it on the version you have -- no need to rebuild.

[ranjitjhala]

Also, very sorry you ended up wasting so much time on this! Going forward, if you get stuck for even more than 20mins, shoot an email/issue because to me thats a "bug" in the tutorial...

[christetreault]

Well, I did that, and it worked. Then I undid it, and now it still works. Is it caching results in that .liquid folder that's in the source dir?

[ranjitjhala]

Sigh. Sorry again!

Yes, its caching the results and it only "rechecks" a function
if you change the implementation for the function (even
slightly, e.g add a blank space.) This makes things very fast,
BUT has these weird effects (and of course, should be something
that is explained in the tutorial...!)

Most certainly my bad...

[gridaphobe]

@ranjitjhala given that we also support checking specific binders with -b, we could extend the emacs mode with a command to check the function the point is in.

[christetreault]

So if I write a function, then write a messed up specification for said function, then fix the spec, it will never recheck it unless I change the function or clear the cache?

[ranjitjhala]

Good idea, except that would require writing emacs-lisp
to determine the binder right?

May be easier to tweak LH with an option:

"--check-line N"

which would force checking of all binders around N-4 ... N+4
which would be:

(a) trivial to do in the "Diff" module
(b) trivial to add in the emacs mode (easy to get N = cursor's position)

On Fri, Feb 13, 2015 at 3:52 PM, Eric Seidel [email protected]
wrote:

@ranjitjhala https://github.com/ranjitjhala given that we also support
checking specific binders with -b, we could extend the emacs mode with a
command to check the function the point is in.

Reply to this email directly or view it on GitHub
#3 (comment)
.

Ranjit.

[ranjitjhala]

Yes, it will not recheck UNTIL you "jostle" the function a bit,
e.g. go inside the body and add a " " somewhere ...

(or you go and DELETE the "--diff" flag at the top of the file...

On Fri, Feb 13, 2015 at 3:54 PM, Chris Tetreault [email protected]
wrote:

So if I write a function, then write a messed up specification for said
function, it will never recheck it unless I change the function or clear
the cache?

Reply to this email directly or view it on GitHub
#3 (comment)
.

Ranjit.

[gridaphobe]

@ranjitjhala most major-modes (including haskell-mode) provide functions to determine the function-at-point :)

[gridaphobe]

This wouldn't directly solve the issue of only editing a spec, but we could extend hsakell-mode's function to notice when it's inside a liquid-spec. I'll look into it next week.

[gridaphobe]

That being said, it might be nicer to just fix the issue at the level of diffcheck. E.g. do an actual diff of the new and saved files and check everything that depends on changed lines.

[ranjitjhala]

Good point (Re: major modes) but in general, given
the choice between hacking emacs-lisp and hacking
haskell, I think the right answer is pretty clear to me :)

On Fri, Feb 13, 2015 at 3:58 PM, Eric Seidel [email protected]
wrote:

This wouldn't directly solve the issue of only editing a spec, but we
could extend hsakell-mode's function to notice when it's inside a
liquid-spec. I'll look into it next week.

Reply to this email directly or view it on GitHub
#3 (comment)
.

Ranjit.

[christetreault]

This issue should be closed after PR #27 is merged

[christetreault]

So it seems that this particular case is fixed. Using the following command line, it works:

cabal exec liquid ../liquidhaskell-tutorial/src/04-poly.lhs -- --diffcheck