[TUHS] MacOS X is Unix (tm)
clemc at ccc.com
Wed Jan 4 04:33:24 AEST 2017
On Tue, Jan 3, 2017 at 1:04 PM, Joerg Schilling <schily at schily.net> wrote:
> "True64" did use the ILP64 model
DEC's Tru64 for Alpha defaulted to LP64 (ex-DECie, Tru64 hacker et al).
You could get ILP64 with compiler switches, but Tru64 was the first of the
LP64 systems. We chaired the 64 bit team between the vendors and ISV
etc, but I'll not drag that dirt here.
As I like to point out, the greatest gift DEC had to the industry was their
64 bit (LP64) DEC C and C++ compiler, after 3-5 years before SGI, Sun et
al. (much of that work lives today as the Intel compiler BTW). As a
result the ISV had to make their code clean. When Sun, IBM et al followed
suite a few years later, that had better code to work with***. Besides
the Gimpel Flexelint <http://www.gimpel.com/html/flex.htm> (a very cool
program and I recommend it highly for serious programmers in the key of
what Larry has been discussing); Judy Ward's incredible diagnostics in the
DEC C++ front-end was the only tool that I know of that really told you
what was going on in your code (funny, I literally am typing this message
after I just returned from lunch with her today, and we we talking about
what we had to do in those days to help people).
Since, I've taken on the topic in Quora so I'll not overly repeat myself
here - there in more detail in that post. But it really comes down to the
this: early (v7 and before) C code assumed sizeof(int) == sizeof(*int).
When the Vax was introduced, it was just easier to go ILP32 when moving
code from the old compilers and in the case of the Vax that made sense but
was not always true. I personally built an early 68000 compiler that was
LP32 and the MIT guys did it ILP32 in their compiler. I had no tools like
we would have later, so porting code to the 68000 could >>sometimes<< be an
issue. The MIT compiler (or its children) became the standard C compiler
for most of the 68000 code [we both were "right" depending on your design
preference - speed of port vs speed of execution].
When we did Alpha, we had >>long discussions<< about word size. It was
agreed that LPxx was not reasonably safe (and that proved to be true). The
problem was that -1 @ 64bits is not 0xFFFFFFFF and those sorts of errors
>>were<< riddled through much code.
Today, I think many of those error have been found, and the fact that most
Intel*64 compilers use LP64 also and code pretty much works is a tribute to
that work. That said, when we start using a 128bit int, I'm sure "dirty"
code will appear. If I recall, the Gimpel guys will warning you when you
have a bit mask.
*** Picking on a Sun and the Sun ISV's here a little. I love to tell the
stories (<-- plural) of ISV's that found their bugs rate on Solaris drop
after the Tru64 port; because we forced them to clean up their act some.
The Sun compiler would accept most anything and generate code, and as Larry
says, much of the resultant code was garbage. While I can write crappy
program in any language, a compiler with excellent warnings or tools like
Flexelint, will not rid you of structural stuff, it will at least make what
it written unambiguous and thus less likely to take a surprising nose dive
in the field.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the TUHS