This is the mail archive of the cygwin mailing list for the Cygwin project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [ANNOUNCEMENT] [1.7] pl-0.10.2-1 - the Parma Polyhedra Library

Charles Wilson wrote:
> Dave Korn wrote:
>>   I'm not sure.  It's the compiler that knows what it needs, which is the same
>> regardless of binutils' preferences, I'm leaning toward the idea that it
>> should be in the linker specs for the benefit of anyone using older binutils.
> The only reason the compiler NEEDS auto-import is to link against the
> shared runtime libraries, since they are not declspec(dllimport)
> decorated, right?  

  That's the only reason the compiler would need auto-import itself, but there
are an awful lot of user libraries out there that need it too, and it does
seem to be very common that that warning flies past in someone's build log and
they don't notice it.

> However, if you're linking against the static
> runtimes, then the compiler doesn't care, and it should be up to the user.
> However, if the compiler unconditionally forces enable-auto-import in
> the specs file, then the end-user has no choice (or, MIGHT have no
> choice depending on the vagarious/order-of-precedence of argument parsing).
> Maybe the right compromise here is for gcc-4's specs to force
> enable-auto-import when linking against the shared runtimes, but to
> leave it up to the default binutils behavior for static?

  Hmm, you raise a point.  We should probably only enable it by default in ld
when -Bdynamic.  And yes, for the compiler's specs that would equate to
passing it down when using shared runtimes.


Unsubscribe info:
Problem reports:

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]