RE: alignment errors on armel: what to do?

[Weidong Li <wli@marvell.com>]

> From: Daniel Kahn Gillmor [mailto:dkg@fifthhorseman.net] 
> Sent: Tuesday, September 29, 2009 5:09 PM
> To: debian-arm@lists.debian.org
> Subject: Re: alignment errors on armel: what to do?

> true, i'm not saying it's non-compliant that it's char-aligned.  it's
> just interesting to observe that in i386, it seems to be guaranteed to
> be word-aligned, even if i do:
>
> char g;
> char f[4];
>
> both &g and f are 32-bit aligned on i386.  not so when i build on armel.
>
> given armel's failures on alignment errors (and i386's apparent ability
> to handle them gracefully -- try passing f+1 to test() ) i find it
> curious that gcc would be *more* cautious about over-alignment on i386
> than on armel.

For armel, this is the expected behavior according to eabi, which uses "natural" alignment.  Gcc for i386, I guess, uses the default alignment of 4 bytes.  The compilers do not seem to be wrong.

-- Weidong