Updated in response to comments.

Added additional comments to CFI_Parser<A>::decodeFDE.

To be clear, this new code parses exactly as much of each FDE as the existing __register_frame/__unw_add_dynamic_fde does, including doing the same work to compute the record length. Neither needs to parse the instructions at registration time.

In D111863#3072829, @joerg wrote:

Are you mixing up of __register_frame and __register_frame_info?

It's also worth noting that FreeBSD's version of libgcc exception handling is actually based on the libunwind code, with a local patch that implements compatibility with libgcc __register_frame by changing it to parse an entire .eh_frame section (in a slightly more ad hoc fashion than this code). Having this new entry point in-tree would simplify the FreeBSD local changes.

In D111863#3069279, @lhames wrote:

I think the ORC runtime provides a much more natural way to test this. Did you manage to come up with some ORC-runtime based tests in the end?

In D111863#3065992, @MaskRay wrote:

I looked at the libgcc mechanism at one time. I remember that in most cases it just uses PT_GNU_EH_FRAME and these eh_frame boundary registry functions are not needed.
Can ORC just use PT_GNU_EH_FRAME?

In D108081#2945526, @lhames wrote:

I know that you mentioned on discord that you do not have commit access: Would you mind if I removed the walkEHFrame and TLV functions from the runtime, then committed on your behalf?

Corrected file header comments

In D105389#2864941, @clayborg wrote:

LGTM. Jim, chime in soon if you have any other objections!

Updated based on reviewer suggestions, thanks.

Updated against r347619

I don't have commit access; could someone land this please? Thanks.

Adds an inline comment about Open Dylan debug info