[PATCH] [Support] fix the ULEB128 decoding bug.

[Duncan P. N. Exon Smith]

Ouch.  LGTM!

> On 2014-Aug-22, at 09:30, Alex Lorenz <arphaman at gmail.com> wrote:
> 
> Hi bob.wilson, bogner, dexonsmith,
> 
> This patch fixes the ULEB128 decoding which currently overflows when evaluating (*p & 0x7f) << Shift were *p = 0x10 and Shift is 28.
> 
> http://reviews.llvm.org/D5029
> 
> Files:
>  include/llvm/Support/LEB128.h
>  unittests/Support/LEB128Test.cpp
> 
> Index: include/llvm/Support/LEB128.h
> ===================================================================
> --- include/llvm/Support/LEB128.h
> +++ include/llvm/Support/LEB128.h
> @@ -82,7 +82,7 @@
>   uint64_t Value = 0;
>   unsigned Shift = 0;
>   do {
> -    Value += (*p & 0x7f) << Shift;
> +    Value += uint64_t(*p & 0x7f) << Shift;
>     Shift += 7;
>   } while (*p++ >= 128);
>   if (n)
> Index: unittests/Support/LEB128Test.cpp
> ===================================================================
> --- unittests/Support/LEB128Test.cpp
> +++ unittests/Support/LEB128Test.cpp
> @@ -106,6 +106,7 @@
>   EXPECT_DECODE_ULEB128_EQ(0xffu, "\xff\x01");
>   EXPECT_DECODE_ULEB128_EQ(0x100u, "\x80\x02");
>   EXPECT_DECODE_ULEB128_EQ(0x101u, "\x81\x02");
> +  EXPECT_DECODE_ULEB128_EQ(4294975616ULL, "\x80\xc1\x80\x80\x10");
> 
>   // Decode ULEB128 with extra padding bytes
>   EXPECT_DECODE_ULEB128_EQ(0u, "\x80\x00");
> <D5029.12845.patch>