Give more graceful error message when compiling Array 0 to C99?

[RyanGlScott]

It is possible to define a Copilot Array of length 0. For example:

zero :: Stream (Array 0 Word32)
zero = constant (array [])

This is a bit weird, as you can't really do much with the resulting array (for instance, you can't index into it), but this is nevertheless a legal stream definition in Copilot. In fact, you can compile this to Bluespec and simulate it:

-- Zero.hs
{-# LANGUAGE DataKinds #-}
module Main (main) where

import Copilot.Compile.Bluespec
-- import Copilot.Compile.C99
import Language.Copilot

zero :: Stream (Array 0 Word32)
zero = constant (array [])

spec :: Spec
spec = do
  trigger "trig" true [arg zero]

main :: IO ()
main = do
  spec' <- reify spec
  compile "Zero" spec'

Interestingly, if you replace the import Copilot.Compile.Bluespec line with import Copilot.Compile.C99, then Copilot crashes outright if you try to compile it:

$ runghc Zero.hs
Zero.hs: NonEmpty.fromList: empty list
CallStack (from HasCallStack):
  error, called at libraries/base/Data/List/NonEmpty.hs:200:15 in base:Data.List.NonEmpty
  fromList, called at src/Copilot/Compile/C99/Expr.hs:381:3 in copilot-c99-4.5-inplace:Copilot.Compile.C99.Expr

I don't believe that C99 permits empty arrays, so it is somewhat understandable that Copilot would reject this program. Still, it would be nicer if Copilot threw a proper error message here rather than crashing with a generic NonEmpty.fromList: empty list error, which leaks copilot-c99's internals.

[ivanperez-keera]

What module would that be in? Should we make it part of:

https://hackage-content.haskell.org/package/copilot-language-4.5/docs/src/Copilot.Language.Reify.html#reify

[RyanGlScott]

This limitation is specific to C99, so I don't think copilot-language would be a good place for a validity check that arrays are non-empty. (After all, you can use empty arrays in Bluespec without issue.) I think somewhere in copilot-c99 would be a better fit. Perhaps Copilot.Compile.C99.Type, when translating Arrays?

[ivanperez-keera]

After all, you can use empty arrays in Bluespec without issue

What for? What can you really do with them?

Could they ever be useful for anything at all?

On a similar note, should we allow empty structs?

[RyanGlScott]

Ah, to be clear, I wasn't trying to impose a judgment value on whether empty arrays are useful or not. I was only remarking that Bluespec does allow you to define them, print them, and check them for equality, if nothing else. In that sense, Array 0 is kind of like ().

I wouldn't be opposed to just having Array 0 be forbidden altogether in Copilot, but then again, I don't think it's hurting anyone to allow them either.

On a similar note, should we allow empty structs?

There is a similar situation with empty structs in that Bluespec allows them, but I believe that they are undefined behavior in C99. If we disallow empty arrays, then we should probably disallow empty structs as well for consistency.

[ivanperez-keera]

Going back on what I said.

I wonder if a struct with no fields that is also an extern could be used to pass around pointers using type synonyms (e.g., intptr_t). It could be useful if a handler needs a pointer to be passed as argument.