Quoting Paul Edwards to All:

PE>Can someone tell me what the deal is with structure padding?

PE>Does it apply to structures when they are DEFINED or when they are
PE>USED? 

I would assume that the definition is where the padding is taken into
account, otherwise 'twould be possible to have two supposedly identical
but incompatible structs:

#pragma pack(1)
struct cow moo;
#pragma pack(4)
struct cow milk;

Both structs must be identical, so the compiler must make the determination
based on the #pragma pack setting in effect at the time the structure was
defined.

PE>ie if I define a struct fred { char x; int y; }; and then do a #pragma
PE>pack(1) (previously was pack(4)), and then declare a variable x, ie
PE>struct fred x;, does x come out 5 or 8? 

Huh.

PE>My experimentation with Watcom says that it is the *structure
PE>definition* not the variable, that gets defined with the appropriate
PE>packing. 

PE>I was wondering whether this was universal?

It'd have to be; otherwise the compiler would need to be able to
differentiate between which style of padding was used. That'd be silly.
You can't have two different types of "struct moo".

PE>Is it true that gcc under Linux has no compile-time option to pack
PE>structures, same as EMX for OS/2?

I don't know about gcc in general, but EMX supports the #pragma pack(n)
directive, and defaults to #pragma pack(4).


... Oxymoron: Press Release.
--- FMail/Win32 1.22