ownSizeOf macro gives me negative sizes

[janflyborg]

Have been wrapping libfuse succesfully with Futhark and I find it excellent. To make it work for 32-bit targets, I had to override two types before importing the generated code:

# These overrides are needed for this to work on 32-bit platforms
struct_stat* = posix.Stat
off_t* = posix.Off
include "libfuse_generated.nim"

Everything works as it should, however, it gave me a warning I couldn't get rid of:

libfuse_generated.nim(908, 16) Warning: Declaration of off_t exists but with different size [User]

When further investigating this, it is the following guard that gives me this:

when ownSizeof(off_t) != ownSizeof(off_t_536871605):
  static :
    warning("Declaration of " & "off_t" & " exists but with different size")

But the strange thing is that ownSizeof(off_t_536871605) gives me the value -3 (when I print it out with a hint), so this comparison will always fail, so something seems to be wrong with ownSizeof for some arguments. I have now changed it into the following definition and the warning is gone:

macro ownSizeof(x: typed): int =
  newLit(typeof(x).sizeof)

I can of course override it before including the generated code, but is this maybe something that should be changed in the code generator?

[PMunch]

Hmm, I'm not sure how your version differs from what Futhark already generates. The warning is just to let you know that you might have a size conflict. If you know you don't it's safe to ignore the warning.

[janflyborg]

Sorry, my bad. I wrote the macro that Futhark generates, and not the one that I replaced it with to silence the warning. Will update now.

[janflyborg]

Updated. Thanks for your assistance so far! Please take another look. The issue is quite clear: without this override, I receive a warning about a size differences for off_t, but with the override, the warning disappear.

[PMunch]

Ah, the problem with sizeof is that it fails for certain types in an unrecoverable way. That's why I switched to ownSizeof because it returns negative numbers instead of crashing.

1985fed

It's strange though that ownSizeof returns a negative number while sizeof works, haven't seen that happen before.

[janflyborg]

In my case it probably has something to do with my redefinition of off_t:

off_t* = posix.Off

I found out that this was necessary for 32-bit targets, so it would be good if this was supported out-of-the-box by Futhark. Taking the size of the type, seems to solve it.