[PATCH] D15455: [Driver] Let -static override the toolchain default PIC setting.
Bob Wilson via cfe-commits
cfe-commits at lists.llvm.org
Fri Dec 18 12:45:47 PST 2015
> On Dec 17, 2015, at 10:59 AM, Bob Wilson via cfe-commits <cfe-commits at lists.llvm.org> wrote:
>
>
>> On Dec 17, 2015, at 10:16 AM, Joerg Sonnenberger via cfe-commits <cfe-commits at lists.llvm.org> wrote:
>>
>> On Wed, Dec 16, 2015 at 11:59:10PM +0000, Bob Wilson via cfe-commits wrote:
>>> We can change this to be Darwin-specific if you prefer, but we should
>>> maintain compatibility with GCC and previous Clang releases in this behavior.
>>
>> Who is really affected by this? I don't care too much about obscure
>> Darwin hacks, but I really wonder why it isn't better to just explicitly
>> add -fno-PIC (e.g. when building a kernel module). It's not like that
>> will break on older versions of GCC or Clang.
>
> Apple has internal projects that are failing to build. This behavior has been in places for many years and I don’t even know how we could find all the people relying on this behavior. Yes, we could break them and force everyone to add -fno-PIC, but typically when we make disruptive and incompatible changes like that, we need to stage the changes and give people a transition plan. For example, we could keep the old behavior but add a warning about the change, something like “warning: -static may be changed in future versions of clang to stop implying -fno-PIC”. After a year or two, we could then go ahead with the change. That is all a lot of work and there needs to be some significant benefit to justify breaking compatibility with older compilers. I don’t see any significant benefit here. It’s a 2-line change to the driver.
Joerg, I’m going to interpret your “I don’t care too much” comment as an indication that you’re not opposed to moving forward with the Darwin-specific change to restore the previous behavior. I went ahead and committed the change in r256026.
I can also add another data point on the impact. We just spent several days tracking down a problem that turned out to be caused by this. The code built successfully but crashed at run-time. It was extremely difficult to figure out what was going wrong.
More information about the cfe-commits
mailing list