-m(no-)align-double compatible libgfortran

[Tobias Burnus]

Hi,

Steve Kargl wrote:
> On Tue, Feb 20, 2007 at 06:57:08PM +0100, Tobias Schl?ter wrote:
>   
>>>>> This is a "m"achine specific option which 
>>>>> should not be touched in the front-end at all.
>>>>>           
I agree that it should be primarily the responsibility of the compiler
user to make sure that the options make sense; and thus tweaking the
machine specific options in the front end don't make sense.

Thus in principle we shouldn't care. However, if we can make libgfortran
compatible with -malign-double/-mno-align-double with little effort, I
think we should really consider doing this. Currently, this seems to
affect only a single structure and "fixing" it would cost only little.
The only problem I see is that we can get to the slippery slope of
adding more and more such tweaks.
 
>>> Have you checked libgomp?
>>>       
The question is where to draw the line. We want to make it hard for the
user to shoot into his own feet, but we cannot prevent it completely.

- Do we ignore -malign-double issues completely?
This is cleanest solution.

- Do we only care for libgfortran?
Here we have only one structure to take care of and this seems to be
most affected part. (Judging from the rather large number of bug reports.)

- Do we also care of libgomp? (Seems to be not affected?)

> I think someone will also need to audit SEQUENCE in derived types
> and the interaction with ISO C Binding.
>   
I'm not sure about this. Many Fortran programs don't depend on any
library except of the Fortran Compiler's one. (Plus lapack/blas maybe,
which don't seem to be affected by this.) And for ISO C Binding they
have to make sure themselves that the Fortran program is compatible with
the C library.

I think drawing the line at libgfortran or at libgfortran+libgomp should
be enough. I also think we should continue advocating against using
-malign-double/-mno-align-double and regard those bugs as INVALID.
(The problem with libgfortran for the users is: They don't realize that
it is there, they are only aware of the libraries they add via -l
explicitly.)

Tobias