On Tue, 16 Jun 2026, Aaron Wohl via Freedos-devel wrote:
<snip>
It is similar to the great industrial revolution and the shift from
handcrafting to the assembly line.
Like I said elsewhere, I have a love-hate relationship with LLMs. There's
ups and downs to them.
My biggest concern is actually that the current system means unless you
train the bot yourself or are very specific with your directives, you
can't be sure of a clean chain of provenance - that's where replacing the
clanked compiler and its tooling will be vital to a proper release,
although I certainly feel OK releasing the initial version as, well, an
initial version.
The libc is a different story. For most of it I directed the clanker to
look very specifically at 3 sources, all of which are versions of BSD. So
I feel pretty comfortable saying that if it's in 4.4, FreeBSD or NetBSD,
and it's in the "Fanta C" runtime, it has a definite provenance. As for
the rest, it's mostly DOS-specific support, and how many ways can you
write gotoxy(), clrscr() or abswrite() and have them be compatible with
Borland's? So for the libc, I have absolutely no qualms that it's
clanked.
On the use of automation in compilers, there has been a push since the
1970s to automate as much as possible. Yacc (and bison) are widely used
to make parsers from formal language specifications rather than
hand-written recursive descent parsers. See
https://github.com/avwohl/uplox/tree/main/examples for grammars for
various languages. The grammar for c is
https://github.com/avwohl/uplox/blob/main/examples/c23.uplox
I've used flex and byacc as examples of the kind of automation that's been
going on longer than I've been around. These days I highly doubt anything
but a toy compiler is still written from the ground up without some sort
of computer program - even if it's nowhere near the level of complexity of
an LLM - developing a significant amount of the guts.
As for how I feel about it, it's great to automate more. I pushed to
move away from assembly code to higher and higher level languages, to
have as much test automation as possible. LLMs seem the next step
I tend to prefer to keep at a very specific sweet spot - where low-level
and high-level meet.
That's called C.
Now I know there's no way in hell that a finished version of Fanta C would
even be desirable to the FreeDOS project until at least the compiler
itself is declanked. I'm still going to *offer* it, not for the FreeDOS
Project, but for anyone else here who might want it. I think it fills a
specific niche - one where many of my projects fall, and one where FreeDOS
falls.
Before someone throws my "Three Rules" back at me: technically this is a
reverse engineering of Turbo C/C++ 1.01; it just happens to have gone
black-box instead of white-box. And I deliberately restrained the clanker
when writing libc/libm (libm is literally just ported over from NetBSD
with no more change than necessary, because the clanker realized who its
author was and decided that I had made the perfect choice).
-uso.
_______________________________________________
Freedos-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/freedos-devel
