14 Jan
2025
14 Jan
'25
11:50 a.m.
On 14/01/2025 13:56, Peter Corlett via cctalk wrote:
On Mon, Jan 13, 2025 at 01:57:44PM -0700, ben via
cctalk wrote:
[...]
As someone in the Atari 68000 camp I'd dispute the idea that 'C' wasn't
used. Most of the OS was written in 'C' apart from DAS BOOT, and most of
the shrink wrap applications (games being written in assembler, I
believe). Originally it used the Alcyon 'C' compiler (whcih I think was
produced by Motorola in some way) and it was pretty naff. There was also
Lattice, and a couple of Tiny 'C's (I'd have said BCPL with
pretensions). Then along came Megamax, which set the standard. You
could, of course, cross-compile for Unix System V but internally we ate
our own dog food!
I defected from Apple to Atari when Uncle Jack "defected" from
Commodore. The Apple II was Assembler but I did have a 'C' compiler for
it (Aztec). Strangely, in 1984 you were faced with BCPL for developing
for the Mac, as only the Lisa had a 'C' cross-compiler and no one could
afford a Lisa. Running 'C' on a single system floppy drive wasn't
practical (the second external floppy came later).
Funny when the 8 and 16 bit micros hit the
market, Algol seemed to vanish
off the face of the earth. Was 64KB too small a address space?
It's likely to be the same reasons why C never really became a thing on
typical microcomputers of the era, although the specifics vary by
architecture:
The popular 8-bit CPUs such as the 6502 and 8080/Z80 lack useful
base-plus-index addressing modes. These are useful for accessing function
parameters and local variables on the stack, and for extracting fields from
a structure. The 6502 doesn't have these modes *at all* and requires
explicit futzing around to compute the effective address, whereas the Z80
does have IX/IY-relative operations, but they are both very slow to use and
only support fixed offsets so if the index is variable, manual address
calculation like on the 6502 is required. Real-world code for these CPUs
tended to use hardwired (i.e. static/global) addresses, and/or used a
virtual machine (with a BASIC interpreter being an extreme case) which was
more capable than the base CPU.
The 6800/6809 and the 65802/65816 are somewhat better on this front, but
they were either too expensive or too late to be relevant. You might as well
just get a 16-bit system instead.
As to the 16-bits, the 8086/80286 had segmentation and so did bang up
against a 64kiB address space because once again there was a lot of futzing
around with segment registers to access more memory than that. They were
barely any better than an 8-bit system with bank switching. Realistically,
they are still basically 8-bit CPUs with 16-bit busses.
The other popular 16-bit CPU was the 68000, which what with its PDP-11
heritage, couldn't fail to be a good C target. A fair chunk of AmigaOS was
written in C, for example. But one could argue that it's actually a 32-bit
CPU with a 16-bit bus. C was still not terribly popular on these machines,
but that'll be because C compilers cost as much as the machine itself, and
generated *much* worse code than handwritten assembler. So the only
customers were businesses for whom time-to-market was more important than
the cost of tools or the performance of the end product.