[TUHS] speaking of early C compilers

[crossd@gmail.com (Dan Cross)]

An interesting digression....  A few years ago, the architects of MIT's
6.828 course ("Operating Systems Engineering") were unsatisfied with the
current stable of systems for teaching, so they did a re-implementation of
6th Edition in modern ANSI C (with a couple of GNU extensions for things
like assigning names to registers) targeting a multiprocessor x86.  It's an
interesting, accessible piece of work; a modern take on a classic.
http://pdos.csail.mit.edu/6.828/2014/xv6.html

        - Dan C.

On Mon, Oct 27, 2014 at 10:48 AM, Noel Chiappa <jnc at mercury.lcs.mit.edu>
wrote:

>     > From: Jason Stevens
>
>     > has anyone ever tried to compile any of the old C compilers with a
>     > 'modern' C compiler?
>     > ...
>     > How did any of this compile? How did this stuff run without
> clobbering
>     > each-other?
>
> As Ron Natalie said, the early kernels are absolutely littered with all
> sorts
> of stuff that, by today's standards, are totally unacceptable. Using a
> variable declared as an int as a pointer, using a variable declared as a
> 'foo' pointer as a 'bar' pointer, yadda-yadda.
>
> I ran (tripped, actually :-) across several of these while trying to get my
> pipe-splicing code to work. (I used Version 6 since i) I am _totally_
> familiar with it, and ii) it's what I had running.)
>
> For example, I tried to be all nice and modern and declared my pointer
> variables to be the correct type. The problem is that Unix generated unique
> ID's to sleep on with code like "sleep(p+1, PPIPE)", and the value
> generated
> by "p+1" depends on what type "p" is declared as - and if you look in
> pipe.c,
> you'll see it's often declared as an int pointer. So when _I_ wrote
> "sleep((p + 1), PPIPE)", with "p" declared as a "stuct file pointer", I got
> the wrong number.
>
> I can only speculate as to why they wrote code like this. I think part of
> it
> is, as Brantley Coile points out, historical artifacts due to the evolution
> of C from (originally) BCPL. That may have gotten them used to writing code
> in a certain way - I don't know. I also expect the modern mindset (of being
> really strict about types, and formal about coverting data from one to
> another) was still evolving back then - partly because they often didn't
> _have_ the tools (e.g. casts) to do it right. Another possibility is that
> they were using line editors, and maintaining more extensive source is a
> pain
> with an editor like that. Why write "struct file *p" wnen you can just
> write
> "*p"? And of course everyone was so space-concious back then, with those
> tiny
> disks (an RK05 pack is, after all, only 2.5MB - only slightly larger than a
> 3.5" floppy!) every byte counted.
>
>
> I have to say, though, that it's really kind of jarring to read this stuff.
>
> I have so much respect for their overall structure (the way the kernel is
> broken down into sub-systems, and the sub-systems into routines), how they
> managed to get a very powerful (by anyone's standards, even today's) OS
> into
> such a small amount of code... And the _logic_ of any given routine is
> usually quite nice, too: clear and efficient. And I love their commenting
> style - no cluttering up the code with comments unless there's something
> that
> really needs elucidation, just a short header to say, at a high level, what
> the routine does (and sometimes how and why).
>
> So when I see these funky declarations (e.g. "int *p" for something that's
> _only_ going to be used to point to a "struct file"), I just cringe - even
> though I sort of understand (see above) why it's like that. It's probably
> the
> thing I would most change, if I could.
>
>         Noel
> _______________________________________________
> TUHS mailing list
> TUHS at minnie.tuhs.org
> https://minnie.tuhs.org/mailman/listinfo/tuhs
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://minnie.tuhs.org/pipermail/tuhs/attachments/20141027/c06fe681/attachment.html>