Re: _obstack_begin alignment / gnu-smalltalk bug

[Benno]

Hi all,

I orginally wrote about this over two weeks ago and have received
no reply. Can someone in the know please at least tell me if this
is a bug or not, or provide me with some feedback here?

Benno

On Sat May 29, 2004 at 15:55:03 +1000, Benno wrote:
>Hi all,
>
>I have a bug with either gnu-smalltalk, or glibc on ia64 platform.
>It stems from the fact that ia64's long doubles are 128bit aligned, not
>64bit.
>
>The problem is that _obstack_begin from glibc returns things that are
>not aligned enough to put long doubles in.
>
>This could be because gnu-smalltalk calls it with '0' alignment and hence gets
>a default alignment, which is only 8 bytes, not 16 bytes, however, I think
>that _obstack_begin should in this case allocate memory that is 16 byte aligned
>so that it can hold a long double. (But I can't find any documentation on
>what _obstack_begin is *meant* to do).
>
>Assuming the bug is glibc, which I think it is then the patch is something
>along the lines of:
>
>struct fooalign {char x; double d;};
>                         ^--- this should be long double.
>
>(near the top of malloc/obstack.h)
>
>Of course smalltalk has this same code in their copy of obstack.c
>
>If obstack is only meant to return 8 byte aligned stuff then the fix needs
>to be somewhere in gnu-smalltalk's obstack.h
>
>Cheers,
>
>Benno