Re: make existing mips files multi-ABI

[Andreas Jaeger <aj at suse dot de>]

Alexandre Oliva <aoliva at redhat dot com> writes:

> I've been working on using the stat data structures used by the kernel
> directly on n32 and n64, but I'm wondering...  Do we really want to do
> this to avoid the copying?  At the expense of a little bit of POSIX
> compatibility, that o32 already has? (not dev_t, that's easy to fix,
> it's time_t: even though time_t is a 64-bit data type on n64, both in
> the kernel and userland, the kernel uses a 32-bit number to hold
> time_t in struct state)  Also, should padding ints be added at the end,

Please make the userland have space for a 64-bit time_t type and
convince the kernel developers to do the same.

> like we do on o32 and many other ports?
>
> Or are we better off defining userland data structures independent of
> the kernel, which requires the use of xstatconv, degrading
> performance?  What's the general policy in this regard, if there is
> any?

Use the kernel only if it supports (or has space to support)
everything we think of ;-) (space for 64-bit dev_t, 64-bit time_t
etc),

Andreas
-- 
 Andreas Jaeger
  SuSE Labs aj at suse dot de
   private aj at arthur dot inka dot de
    http://www.suse.de/~aj