On Sep 27, 2013, at 11:33 AM, Rafael Espíndola <[email protected]> 
wrote:

>> I see. If driver always passes -fobjc-default-synthesize-properties then our 
>> users will be OK.
>> But, we still need to use legacy property synthesis for testing purposes. 
>> But we can always use cc1 directly.
>> LGTM them.
> 
> Sorry, it looks like there is a bit of confusion.
> 
> the current status is:
> * Driver alway passes -fobjc-default-synthesize-properties
> * -cc1 defaults to false, but only 3 tests expect the behavior of not
> having -fobjc-default-synthesize-properties.
> * There is no -fno-objc-default-synthesize-properties, so the only way
> to get the off behavior is to run -cc1 directly.
> 
> With the patch we get:
> * Driver doesn't pass -fobjc-default-synthesize-properties.
> * -cc1 always has behavior we currently get by passing
> -fobjc-default-synthesize-properties.
> * The 3 tests that were not using -fobjc-default-synthesize-properties
> are updated (super-class-protocol-conformance.m,
> unimplemented-protocol-prop.m, property-category-1.m)
> * -fobjc-default-synthesize-properties  is removed from -cc1 too, so
> there is no way to turn it off.

>From a user perspective this is all fine. But from a testing perspective we 
>cannot turn off the default property synthesis
any more. So, we are permanently removing none-auto property synthesis in 
clang. I guess with other -fxx options there is an -fno-xx option
available and I need to add -fno-objc-default-synthesize-properties  next time 
we need one. I guess I am ok with this.

- Fariborz

> 
> Cheers,
> Rafael


_______________________________________________
cfe-commits mailing list
[email protected]
http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits

Reply via email to