Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!unmvax!pprg.unm.edu!hc!lanl!jlg
From: jlg@lanl.gov (Jim Giles)
Newsgroups: comp.arch
Subject: Re: Sorting struct members for alignment (was Re: Unaligned Accesses)
Message-ID: <11541@lanl.gov>
Date: 5 Apr 89 22:11:42 GMT
References: <583@geovision.UUCP>
Organization: Los Alamos National Laboratory
Lines: 16

From article <583@geovision.UUCP>, by gd@geovision.uucp (Gord Deinstadt):
> And it's about time there was such a keyword.  We've got a million plus
> lines of C code here, and hardly a line of it needs to know what a structure
> looks like internally.  For our purposes (as for most application programmers)
> structure member reordering should be the default.


I disagree.  One of the main reasons I use structures is to describe tables
that will be used to share data between machines and/or between languages.
For this reason, I want all the fields to be in the order I specify _AND_
I want them contiguous (_NO_ compiler generated padding).  Any automatically
generated padding or alignment may vary from compiler to compiler and almost
certainly _will_ vary from one machine to another.  If I design a table that
tends to span word boudaries with fields, or has fields that begin at clumsy
places - that's my fault and I'll take the performance penalty - at least
the data will be shared correctly.