donn (09/09/82)
The only complete way to solve the problem of portability of binary data gen-
erated by C programs is to create and enforce restrictions on the alignment of
the various data types, and the format of long integers and floating point
numbers. The standard would have to be that of the most restrictive/bizarre
machine. I can imagine people writing horribly distorted compilers that use
data formats and alignments completely unsuited to the machine they are based
on... No one in their right mind would ever do such a thing.
The next least ridiculous thing that can be done, short of writing out only
clumsy voluminous ASCII data, is to find a reasonable arrangement that is not
too gruesome and that can be handled by at least a large subset of the avail-
able machines, or of the machines of a given family. At UCSD this has pro-
duced two varieties of compiler: a mildly tweaked pcc for the VAX which does
16-bit alignment of 32-bit quantities (but no adjustment of longs); and a
severely abused Ritchie compiler for the PDP which does VAX longs adjustment
(and may soon with a little tweaking do 32-bit alignments). The latter pro-
duces systems which boot off of VAX filesystems on dual-ported disk drives and
controllers. Since our applications involve lots of binary files and we
expect to be able to make them on both VAXen and PDPs, such portability is
indispensable. I was mildly surprised to find that it hadn't been attempted
before...
Donn Seeley UCSD Chemistry Dept. RRCF ucbvax!sdcsvax!sdchema!donn