bugpoint reduced test case doesn't reproduce bug

[sunfishcode]

Bugzilla Link	1021
Version	trunk
OS	Linux
CC	@nlewycky

Extended Description

While reducing a bad-generated-code bug (which I'll submit a separate report
for), bugpoint produced a reduced test case which is supposed to reproduce the
problem, but it didn't. The bug ended up being with the folding of static
initializer data into instructions that use it. The reduced test case bugpoint
produced had the static initializers removed, so it didn't exhibit the problem.

Here's the reduced test case that reproduces the orignial bug, and which
bugpoint reduces into a test case that doesn't reproduce it.

target datalayout = "e-p:32:32"
target endian = little
target pointersize = 32
target triple = "i686-pc-linux-gnu"
%fmt = constant [4 x sbyte] c"%x\0A\00"
%bytes = constant [4 x sbyte] c"\AA\BB\CC\DD"

implementation

int %main() {
%y = alloca uint
%c = cast uint* %y to sbyte*
%z = getelementptr [4 x sbyte]* %bytes, int 0, int 0
call void %llvm.memcpy.i32( sbyte* %c, sbyte* %z, uint 4, uint 1 )
%r = load uint* %y
%t = cast [4 x sbyte]* %fmt to sbyte*
%tmp = call int (sbyte*, ...)* %printf( sbyte* %t, uint %r )
ret int 0
}

declare void %llvm.memcpy.i32(sbyte*, sbyte*, uint, uint)

declare int %printf(sbyte*, ...)

[lattner]

Unfortunately, this is a known problem with bugpoint. When it switches from "reduce the pass" mode to
"reduce the function list" mode, it implicitly puts all global initializers in the "safe" bc file. Instead, it
should try with and without it in the safe bc file. If they can't be in the safe bc file it should leave it in the
test bc file.

This would be really nice to have, it would have let bugpoint correctly grind the recent globalopt bug as
well.

-Chris

[lattner]

Is it worth keeping this bug open?

[nlewycky]

That or a list of improvements like lib/Target/README but for bugpoint.

[arsenm]

Don't think bugpoint still does this but llvm-reduce definitely doesn't have this issue