[patch, libgfortran] Merge of fortran-dev IO to trunk

[Steve Ellcey]

On Mon, 2009-04-06 at 18:33 +0200, Daniel Kraft wrote:

> Hi Steve,
> 
> thanks for testing this!  That was caused by my floating-point reading 
> patch, where I changed this to not use a memcpy; I thought there were 
> already other points in the library that use this casting...  I will try 
> to locate them, so we can see if those fail, too, and need a fix.
> 
> Daniel

Daniel,

I have had alignment problems before, looking back through the
libfortran ChangeLog I see these changes I made in 2005 that I think
were done to fix alignment bugs much like the one we are seeing now.  It
is possible that not all the uses of unaligned buffers trigger problems
because the buffer might 'accidently' be aligned enough for things to
work.

Steve Ellcey
sje@cup.hp.com


2005-09-07  Steve Ellcey  <sje@cup.hp.com>

        PR libfortran/23419
        * io/write.c (extract_int): Use memcpy to access buffer.
        (extract_uint): Ditto.
        (extract_real): Ditto.

2005-08-31  Steve Ellcey  <sje@cup.hp.com>

        * io/read.c (set_integer): Use memcpy to fill buffer.

2005-08-31  Steve Ellcey  <sje@cup.hp.com>

        PR target/23556
        * io/read.c (convert_real): Use memcpy to fill buffer.