Spec of align attribute is a mess

[Stewart Gordon]

"The great inapplicable attribute debate"
http://www.digitalmars.com/webnews/newsgroups.php?art_group=digitalmars.D&article_id=87915
drifted into a discussion on alignment in relation to unions.  However, 
I've looked around the spec and found that quite a bit of work is needed 
on the specification of align as a whole.


1. ctod.html says
"Clearly, since much of the point to setting alignment is for 
portability of data, a portable means of expressing it is necessary."

However, attribute.html says
"AlignAttribute is meant for C ABI compatiblity [sic], which is not the 
same thing as binary compatibility across diverse platforms. For that, 
use packed structs:"

which is basically claiming that align isn't good for portability of 
data after all.


2. Still on attribute.html, the basic documentation of align says
"Specifies the alignment of struct members. align by itself sets it to 
the default, which matches the default member alignment of the companion 
C compiler. Integer specifies the alignment which matches the behavior 
of the companion C compiler when non-default alignments are used."

What if the companion C compiler:
- doesn't exist?
- has no alignment control facilities?
- has alignment control only on whole structs - IWC how should alignment 
applied to individual members be interpreted?
- doesn't support a particular align value that's used in the D code?

Moreover, if the sole purpose of align is to provide an interface to 
C-compiler-dependent alignment features, then this is incongruous with 
alignment control being given as one of D's improvements over C.  It may 
be the case that _a standardised notation for_ alignment control is an 
improvement, but that's different.


3. What kind of alignment is it meant to be anyway?  Absolute (based on 
raw memory addresses) or relative (to the beginning of the struct)?  It 
seems that various people have assumed various answers to this.

Together with the inapplicable attribute debate, this affects the 
logicality of the statement on attribute.html
"AlignAttribute is ignored when applied to declarations that are not 
structs or struct members."

If alignment is relative, then it's naturally inapplicable to anything 
that isn't a struct or a member thereof.  If OTOH alignment is absolute, 
some attempts to use align become cases of an attribute that makes sense 
in the context being silently ignored.


There are probably other issues with the align documentation, but I 
can't think of them at the mo....

Thoughts?

Stewart.