Re: [PATCH] SPU use a 16 byte fpos_t

[Patrick Mansfield <patmans at us dot ibm dot com>]

On Thu, Sep 06, 2007 at 05:20:59PM -0400, Jeff Johnston wrote:
> Patrick Mansfield wrote:
> >fpos_t is 12 bytes for 32 bit ppc glibc, and 16 bytes for 64 bit.
> >
> >Currently SPU newlib uses 4 bytes for fpos_t. This is always a problem
> >with 64 bit ppc, since the fgetpos() assist call writes 8 bytes back for
> >the offset, and is a potential problem with 32 bit ppc.
> >
> >So for SPU always use 16 bytes for fpos_t and also make currently unused
> >(in SPU) _fpos64_t 16 bytes.
> >
> 
> Patrick,
> 
>   This needs to be cleaned up.  Not your fault, but I don't want cpu 
> checks in sys/reent.h and I don't think that _fpos_t should be magically 
> defined in sys/reent.h.  The following proposal makes this change and 
> others easy for you to do.  It is pretty straightforward.
> 
>   So, this is what I propose:
> 
>   1. copy machine/_types.h to machine/_default_types.h
>   2. make machine/_types.h simply include machine/_default_types.h
>   3. have sys/_types.h include machine/_types.h
>   4. have sys/_types.h look for flags before defining _ types.  For
>      example, #ifndef _OFF_T_DEFINED).  You need to add
>      _fpos_t and  _fpos64_t default defs which weren't there previously
>   5. create an machine/spu/machine/_types.h which defines all of these
>      to your liking...also set the flags on so sys/_types.h won't try
>      and define them
>   6. remove _fpos_t and _fpos64_t defs from sys/reent.h
> 
>   This will be clean and allow other platforms to do tweaking as
> desired without cluttering up the common headers.
> 
>   Comments?

I'm trying this, but machine/spu/machine/_types.h is not being used - at
least it is not installed, full path is
newlib/libc/include/machine/spu/machine/_types.h.

Do I need a configuration change too, or did you mean a different path?

And machine/spu/machine/_types.h would replace (and so has to supply
everything included by) machine/_types.h ?

-- Patrick Mansfield