Compatibility of textually identical structs/unions

[Henry Spencer]

In article <288 at edg1.UUCP> jsa at edg1.UUCP (J. Stephen Adamczyk) writes:
>Is the following legal, according to the May 13 (88-001) draft?
>
>struct {int a;} v1;
>struct {int a;} v2 = {1};	...
>  v1 = v2;     /* <--- Are the structs compatible? */

No.  The two structs are different types.

>I know that 3.1.2.6 says that "two structure, union, or enumeration
>types declared in separate translation units are compatible if they have
>the same number of members, the same member names, and compatible member
>types".  However, this says nothing about types within one translation
>unit.

This rule is necessary because otherwise there is no way that structs etc.
declared in different translation units could ever be compatible.  There
is no such thing as having "the same declaration" appear in two different
translation units -- by definition, it's two different declarations that
happen to be textually identical.  This is basically a concession to the
usual C compilation environment, in which the compiler does not save
state from one translation unit to be used in compiling another.  Within
a single translation unit, this concession is not necessary and is not
made.

It would arguably be cleaner to have the "same declaration" rule apply
throughout in theory, but with an implementation-defined possibility of
relaxing the rule in the event that the implementation can't do that.
That way, an implementation that did save state across compilations
could enforce the stricter rule.  Doing the details right could be
really tricky, though, and in any case it's too late.
-- 
SunOSish, adj:  requiring      |     Henry Spencer at U of Toronto Zoology
32-bit bug numbers.            | uunet!attcan!utzoo!henry henry at zoo.toronto.edu