[llvm] r174543 - Add a test for checking the current .debug_frame dumping capability.

[Eli Bendersky]

On Fri, Feb 8, 2013 at 11:21 PM, David Tweed <David.Tweed at arm.com> wrote:
> Hi,
>
> After Takumi's patch the test values print correctly. I was more just checking that the casting solution was OK, as someone had asked me if I knew why the values in the 64-bit variables were actually forced to only have 32 set bits, which I didn't. Since you've written a web page on DWARF I imagined checking the range with you would be the best idea. Anyway the ARM regression is gone.
>

Thank you for calling the failure out and for the investigation. I
believe you described the bug correctly.

The coercion to 32 bits is legal *at this point* since DWARF64 is not
supported. In DWARF32 (the only version LLVM emitters and parsers
support today) all offsets are restricted to 32 bits, so the value is
fully contained in the low 32 bits. I do plan to expand the frame
parsing implementation a bit to give more support to DWARF64, although
at this point it would be difficult to exercise - see
http://llvm.org/bugs/show_bug.cgi?id=15173

Eli