(times 0) semantics about ivory HOT 6 CLOSED

Thanks. Yes, please do; I'll merge.

from ivory.

OK, I have two solutions. One is a quick fix that does an explicit zero check for the for and times functions. This has the downside that it adds a conditional every time these functions are used. I'm also not sure if you allow zero length arrays, but if you do then it could be a problem there as well.

The other solution that I've implemented is a more thorough solution that adds parameters for whether or not to be inclusive with the end value to the LoopIncr constructors. This change required a fair amount of modification, and broke a couple of tests that I haven't had a chance to look into further.

Let me know which you'd prefer.

from ivory.

Can you post the more thorough solution? It is likely close to what we've been considering--- @jameysharp discovered an issue a while ago we still haven't fixed in which (0 :: Ix 1) passed to for results in a bug. We might as well fix this once and for all. Don't worry about breaking old tests.

from ivory.

OK, I've created a pull request.

I believe the issue existed for for and times, and should be fixed for both.

Also, thanks for the talk on Wednesday, it was swell.

from ivory.

In case anyone wonders, discussion of this issue continued over on pull request #30.

from ivory.

OK, I've implemented a few changes as of 9d2d186 to deal with this:

• The loop function in Ivory.Language.Loop now takes IxRep arguments for the bounds of the loop instead of Ix n, allowing for the indexes to be computed before the call to loop
• for and times now use loop directly, instead of via upTo or downTo
• for and times compute the loop bounds at IxRep instead of Ix n, which avoids the underflow in the case that n is 0

Additionally, I've made loop no longer exported from the Ivory.Language.Loop module, and similarly cleaned up the export list for Ivory.Language.Ix.

The downside to this fix is that if you do give an argument of type Ix 0 to times, you'll likely introduce either garbage data, or divide-by-zero. The fix for this is to always use the div-zero pass when generating code, which should catch cases like that.

from ivory.