Falk Hueffner wrote:
<snip>
> > struct PACKET_HEADER
> > {
> > le length;
> > MD5Hash setid;
> > } __attribute__ ((packed));
> >
> > struct MAINPACKET
> > {
> > PACKET_HEADER header;
> > MD5Hash fileid[0];
> > } __attribute__ ((packed));
> >
> > const MD5Hash& FileId(const char *packetdata, int filenumber)
> > {
> > return ((const MAINPACKET*)packetdata)->fileid[filenumber];
> > }
> >
> > const MD5Hash& SetId(const char *packetdata)
> > {
> > return ((const MAINPACKET*)packetdata)->header.setid;
> > }
> >
> > // END
> >
> > SetId provokes the error; FileId doesn't.
>
> The problem is that the packedness of MAINPACKET::header isn't
> ignored.
>
> Note also that the program will not work as intended, because (as soon
> as gcc is fixed) all the packedness is ignored. So I clone this bug.Thanks for the confirmation. Personally I think packed structures are a horrible idea. Presumably if all the classes involved are changed to be POD, they'll become packed again? (In practice I don't think it makes a difference, as the class members appear to be naturally aligned without any need for padding bytes.) Ben. -- Ben Hutchings One of the nice things about standards is that there are so many of them.
signature.asc
Description: This is a digitally signed message part
