[lld] r230157 - [ELF] Teach GNU Driver about --stats.

Joerg Sonnenberger joerg at britannica.bec.de
Wed Feb 25 03:57:05 PST 2015


On Tue, Feb 24, 2015 at 07:13:48PM -0800, Davide Italiano wrote:
> I think that the whole point of having a GNU compatible driver is that
> of being fully compatible with the existing behaviour.

I completely disagree with that notion. Heck, even GNU ld is not
compatible with GNU ld in pretty fundamental ways.

> I'm not sure why you're considering debugging options as second class
> citizen.

lld works fundamentally different than GNU ld. If you want to debug the
behavior of lld, it follows that the interesting debug data is
fundamentally different than GNU lds. So how do you want to be fully
*output* compatible? It doesn't make sense. If we are not fully
compatible in the output, it is better not to pretend to support the
option in first place. That makes it at least easy to test for the
option.

> We had a related discussion about --export-dynamic VS -export-dynamic
> and what you proposed was to convince all the consumers to switch to
> the "right" format. I don't think this is a reasonable request.
> Your mileage may vary.

My request was to make sure that the user asked for the right thing. It
is bad enough that they had the brilliant idea to support two different
long option formats in the same program. They made it worse by creating
a long option that is only distinguished by a short option with an
argument due to order rules. That's insane. Just because we want to
reduce the pain of switch from GNU ld to lld by providing compatibility
doesn't mean we have to support all the ridiculous interface bugs.

Joerg



More information about the llvm-commits mailing list