[bindings/go] Fix building on 32-bit systems (ARM etc.)

[aykevl]

Summary

The patch in https://reviews.llvm.org/D53883 (by me) fails to build on 32-bit systems like ARM. Fix the array size to be less ridiculously large. 2<<20 should still be enough for all practical purposes.

Bug: https://bugs.llvm.org/show_bug.cgi?id=40426

[whitequark]

Seems fine, with or without the length change. Hopefully this can get into 8.0.

bindings/go/llvm/ir.go

1265

Looks like the Go documentation link above recommends using 1 << 28. Seems like a really awkward pattern, but I'm not a Go expert...

[aykevl]

Can you (or someone else) merge this change? I do not have commit access. That would allow a request for inclusion in LLVM 8 (I hope).

bindings/go/llvm/ir.go

1265

That's a modification done by me, because the original code with 1 << 30 led to this bug. The exact number doesn't really matter, as long as it is "big enough".

Yeah this is rather dirty Go, but it's as elegant as things get in CGo land. There may be better ways, and if someone knows one they should probably change this. For now, this fixes a rather important bug (FTBFS).

bindings/go/llvm/ir.go

func InlineAsm(t Type, asmString, constraints string, hasSideEffects, isAlignStack bool, dialect InlineAsmDialect) (rv Value) {
   return
 }
 
 // Operations on aggregates
 func (v Value) Indices() []uint32 {
   num := C.LLVMGetNumIndices(v.C)
   indicesPtr := C.LLVMGetIndices(v.C)
   // https://github.com/golang/go/wiki/cgo#turning-c-arrays-into-go-slices
-  rawIndices := (*[1 << 30]C.uint)(unsafe.Pointer(indicesPtr))[:num:num]
+  rawIndices := (*[1 << 20]C.uint)(unsafe.Pointer(indicesPtr))[:num:num]
   indices := make([]uint32, num)
   for i := range indices {
     indices[i] = uint32(rawIndices[i])
   }
   return indices
 }
 
 //-------------------------------------------------------------------------