Empty storable vectors in GHCi error with divide by zero.

[mihaimaruseac]

In GHCi:

=> import Data.Vector.Storable
=> empty
*** Exception: divide by zero

Enabling -XNoMonomorphismRestriction, giving it an explicit type or having it compiled by ghc from a file makes the error go away.

This doesn't happen for either boxed (Data.Vector) or unboxed (Data.Vector.Unboxed) vectors.

I consider it a vector bug since it's an inconsistency between the unboxed and the storable case. But, if it is not, feel free to close this/raise it to the proper bug tracker.

[RyanGlScott]

Well, this issue didn't exist in GHC 7.10.3 or earlier by sheer coincidence. The reason it changed in 8.0 is that a new Storable () instance was added:

instance Storable () where
  sizeOf _ = 0
  alignment _ = 1
  peek _ = return ()
  poke _ _ = return ()

And luck just has it that if you type empty at the REPL with no type annotations, its type will default to Vector (). It seems that having sizeOf _ = 0 causes the divide by zero error. I'm not familiar enough with the codebase to know where the faulty division is happening, though.

[RyanGlScott]

A quick search reveals this as a likely culprit:

instance Storable a => G.MVector MVector a where
  ...
  basicUnsafeNew n
    | n < 0 = error $ "Storable.basicUnsafeNew: negative length: " ++ show n
    | n > mx = error $ "Storable.basicUnsafeNew: length too large: " ++ show n
    | otherwise = unsafePrimToPrim $ do
        fp <- mallocVector n
        return $ MVector n fp
    where
      size = sizeOf (undefined :: a)
      mx = maxBound `quot` size :: Int

[mihaimaruseac]

Makes sense and indeed that seems to be the culprit.

Now, for solution, the simplest way would be to disallow making a storable vector of (). At least this will give educative error messages. I made #168 for this, as a starting point.

[cartazio]

I think there's a preexisting ticket about handling zero width storable values. Making them erroneous probably isn't the right approach ...

…

On Sun, Apr 30, 2017 at 9:57 PM Mihai Maruseac ***@***.***> wrote: Makes sense and indeed that seems to be the culprit. Now, for solution, the simplest way would be to disallow making a storable vector of (). At least this will give educative error messages. I made #168 <#168> for this, as a starting point. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#167 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAAQwklKu6VUkS2OGNyQrIeIY20s24nQks5r1TvqgaJpZM4NMuYD> .

[mihaimaruseac]

Yeah, it's issue #105 (I only searched for it now). I see that no progress has been made on it :(

I changed my PR so now it divides by 1 if the size of the elements is 0.

[mihaimaruseac]

Compiled the following with profiling

import qualified Data.Vector.Storable as VS

main = print (VS.empty :: VS.Vector ())

and indeed got the expected error

*** Exception (reporting due to +RTS -xc): (THUNK_STATIC), stack trace:
  GHC.Real.CAF
  --> evaluated by: Data.Vector.Storable.Mutable.basicUnsafeNew,
  called from Main.main,
  called from Main.CAF:main3
  --> evaluated by: Main.main,
  called from Main.CAF:main2
  --> evaluated by: Main.main
mm: divide by zero

The error goes away with fix I proposed in #168

[dolio]

Fix merged.