[TUHS] Work I've done with a PDP-11 simulator

Noel Chiappa jnc at mercury.lcs.mit.edu
Tue May 13 03:06:17 AEST 2014

    > From: John Cowan <cowan at mercury.ccil.org>

    > Well, provided the compiler is honest, contra [Ken].

A thought on this:

The C compiler actually produces assembler, which can be (fairly easily)
visually audited; yes, yes, I know about disassembly, but trust me, having
done some recently (the RL bootstrap(s)), disassembled code is a lot harder
to grok!

So, really, to find the Thompson hack, we'd have to edit the binaries of the

For real grins, we could write a program to convert .s format assembler to
.mac syntax, run the results through Macro-11, and link it with the other
linker... :-)

Also, I found on what's going on here:

    > What was wierd was that in the new one, the routine csv is one word
    > shorter (and so is csv.o). So now I don't understand what made them the
    > same sizes!? The new ones should have been one word shorter!? Still
    > poking into this...

The C compiler is linked with the -n flag, which produces pure code. What
the linker documentation doesn't say (and I never realized this 'back in the
day') is that when this option is used, it rounds up the size of the text
segment to the nearest click (0100).

So, in c2 (which is what I was looking at), the last instruction is at
015446, _etext is at 015450, but if you look at the executable header, it
lists a text size of 015500 - i.e. 030 more bytes. And indeed there are 014
words of '0' in the executable file before the data starts.

And if you link c2 _without_ the -n flag, it shows 015450 in the header as
the text size.

So that's why the two versions of all the C compiler phases were the same
size (as files); it rounded up to the same place in both, hiding the one-word
difference in text size. 


More information about the TUHS mailing list