Var init about refscript HOT 14 CLOSED

Btw, the error message I get is:

/Users/rjhala/research/liquid/RefScript/tests/pos/misc/init-00.ts:4:9-4:17: Error: TC-ERROR
Variable definition of 'i_SSA_0' with neither type annotation nor initialization is not supported.

from refscript.

Agreed. This has been around for a while, but now that we seem to have a clear story for 'undefined' we can just implicitly use that in cases where we don't have an initializer.

For the moment, writing down the 'undefined' should do it.

from refscript.

I see. Adding undefined doesn't quite cut it -- or rather it works for the one case, but not for the second second test I added, where TC complains about joining undefined and number. In general this is sketchy, because indeed, due to the while loop, maybe the i is indeed not defined but since the i is dead -- i.e. not used afterwards it shouldn't matter...

from refscript.

Also, I thought I would hack around it by using the crash from the prelude to just make up a fictitious value of any-old type, but it get some strange error, see:

https://github.com/UCSD-PL/RefScript/blob/varinit/tests/pos/misc/init-02.ts

from refscript.

See the new tests I have added -- bottom-line is that the simple undefined
trick is not working. These seem like pretty simple things, and should be
fixed soon IMO.

On Mon, Sep 15, 2014 at 7:28 PM, Panagiotis Vekris <[email protected]

wrote:

Agreed. This has been around for a while, but now that we seem to have a
clear story for 'undefined' we can just implicitly use that in cases where
we don't have an initializer.

For the moment, writing down the 'undefined' should do it.

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

Ranjit.

from refscript.

OK I made the fix that addresses:
https://github.com/UCSD-PL/RefScript/blob/varinit/tests/pos/misc/init-00.ts

This one:
https://github.com/UCSD-PL/RefScript/blob/varinit/tests/pos/misc/init-01.ts
will require some more thought - e.g. introduce a union at the exiting (from the loop) phi var.

The last one you posted:
https://github.com/UCSD-PL/RefScript/blob/varinit/tests/pos/misc/init-02.ts
is a TypeScript error. It has to do with the way TS sometimes treats open (polymorphic - as in forall A . A) types as {} -- the empty object type.

from refscript.

I suggest just merging (joining) T + undefined == undefined (keep it simple) ?

from refscript.

Yikes. Is there any way around the TS error? crash or diverge is a pretty handy thing to have...
but anyway, if you fix the union issue -- by just special casing undefined -- thats fine too IMO, because
this is a common imperative idiom.

from refscript.

You can always 'any' it up.
I.e.

/*@ crash :: forall A . () => A */
declare function crash(): any;

from refscript.

OK I'll do the "simple - undefined" solution for the moment.

If we end up having situations like:

var i;
if (a > 0) i = a;
if (i) assert(i > 0);

we might have to revise

from refscript.

Indeed -- good plan.

On Sep 15, 2014, at 8:15 PM, Panagiotis Vekris [email protected] wrote:

OK I'll do the "simple - undefined" solution for the moment.

If we end up having situations like:

var i;
if (a > 0) i = a;
if (i) assert(i > 0);
we might have to revise

—
Reply to this email directly or view it on GitHub.

from refscript.

Btw wouldn't be explicit undefined plus number annotation on i work in this case?

On Sep 15, 2014, at 8:15 PM, Panagiotis Vekris [email protected] wrote:

OK I'll do the "simple - undefined" solution for the moment.

If we end up having situations like:

var i;
if (a > 0) i = a;
if (i) assert(i > 0);
we might have to revise

—
Reply to this email directly or view it on GitHub.

from refscript.

Yes, in this specific one the number annotation can be inferred.

But note that adding the annotation, means no SSA. So with multiple assignments you'd lose flow sensitivity.

from refscript.

Addressed by #26

(all tests pass.)

from refscript.