[llvm-commits] [PATCH] An option to selectively enable parts of ARM EHABI support

[Evgeniy Stepanov]

On Wed, Jan 18, 2012 at 9:01 PM, Anton Korobeynikov <asl at math.spbu.ru> wrote:
>> Enabling it by default would be nice, but we are not quite ready for
>> that. Besides my other CL, I've found another case of an unsupported
>> instruction. So, sometimes -arm-enable-ehabi just breaks compilation.
> Ok. Then it's a good thing to add support for missed instructions :)

Yeah. And I plan to do so until it works on real-world programs. It
seems we are pretty close, though.

>
>> So, for now, we need a way to (a) enable just unwinding, and (b)
>> enable both, with both disabled by default.
> Well, but you need this for your local debugging, no?

I'm not sure I understand you.

ASan needs a working backtrace on ARM. Not for debugging, but rather
for "production". But still, I'm sure, there will be a period of time
when backtrace support (EHABI without descriptors) will work fine for
us, but will not yet be stable enough to enable by default for
everyone.

In other words, I plan to develop the EHABI w/o descriptors mode. To
be able to do it without forking LLVM, I need a switch to enable it,
which is off by default. For other people to develop a much more
complex full EHABI support, there must be another switch to enable the
whole thing. (btw, what are your plans regarding this?)

WDYT we have a single tri-switch like
-arm-enable-ehabi=no|backtrace(unwind)|full(yes)?

>
> --
> With best regards, Anton Korobeynikov
> Faculty of Mathematics and Mechanics, Saint Petersburg State University