On Sat, 30 Dec 2000, Jason R Thorpe wrote:

> On Sat, Dec 30, 2000 at 11:47:53AM -0800, Matthew Jacob wrote:
> 
>  > Read my update. It is a bug somewher and not the code.
>  > 
>  > char *foo = "XXX"
>  > 
>  > is not
>  > 
>  > const char *foo = "XXX";
> 
> The Alpha compiler generates the following code for 'char *foo = "XXXX";':
> 
> 	.globl foo
> .section        .rodata
> $LC0:
> 	.ascii "XXXX\0"
> .data
> 	.align 3
> foo:
> 	.quad $LC0
> 	.ident  "GCC: (GNU) egcs-2.91.66 19990314 (egcs-1.1.2 release)"
> 
> ...and the same code for 'const char *foo = "XXXX";'.
> 
> If you change it to 'char foo[] = "XXXX";':
> 
> 	.globl foo
> .data
> foo:
> 	.ascii "XXXX\0"
> 	.ident  "GCC: (GNU) egcs-2.91.66 19990314 (egcs-1.1.2 release)"
> 
> ...and if you compile the first version with -fwritable-strings:
> 
> 	.globl foo
> .data
> $LC0:
> 	.ascii "XXXX\0"
> 	.align 3
> foo:
> 	.quad $LC0
> 	.ident  "GCC: (GNU) egcs-2.91.66 19990314 (egcs-1.1.2 release)"
> 
> This is one of the subtle semantic differences between * and [].
> 

[ you have an alpha again? Great! ]

Yes. It's really hard for me to shake myself from early training with C which
said that * && [] are semantically equal.

I'm certainly convinced of this now. As I said, I wouldn't have gone this
direction with the language., but that hardly matters.

-matt