Re: New "genAI" zig scheme implementation

--000000000000ff8dd80657a4c463
Content-Type: text/plain; charset="UTF-8"

It's interesting since the author of Zig rejects AI generated code, he
insists on coding by himself. But his high quality code is trusted and used
to be the product of vibe coding like such a project.

I was managed to implement r7rs-small with my brand new optimized compiler.
But I didn't use LLVM as optimizing layer. I saw this Zig Scheme
implementation contains LLVM converter. This may reduce large complexity.

I wouldn't say r7rs-small is hard for AI. But I think AI should try
r7rs-large, especially for the testing work.

I've been trying total vibe coding in a complex C++ project for 1 month.
It's not so simple to generate code with a functional described prompt. But
a bunch of design spec with carefully designed data structures,
conventions, exceptions handling logics, and testing hints. Even so, I feel
so tired to discuss with Claude on many corner cases, compromises and
architecture alignments. But it's still good to enjoy the corner cases
checking and auto testing by AI. It's hard to imagine that I can manage to
do this project on my own, but it's true now.

However, this is the only project I use pure vibe coding, yet.
In vibe coding, I feel like my hacking-power was drained quickly by the
thirsty AI. I have to save some for my rest life.

The key of software is robustness and interpretability. The key to Internet
application business is fast-iterating, not a key to software itself.
I imagine one day, people will find me on my software issue, and I can tell
them the design idea, and why this part works like that, or why this is a
bug.

After all, the value of software is established in the trust to its
designer. I respect vibe coders, but I will ask them, do you believe in the
thing you just generated? To myself, I would say yes, and if they don't
trust me, I will show my design spec docs (it's almost accumulated to a
book now) and tell them why.

On Wed, Jul 8, 2026, 01:24 Nicolas Graves <[email protected]> wrote:

>
>
> https://www.reddit.com/r/scheme/comments/1upm0sb/kaappi_yet_another_implementation/
>
> I don't know how to react to this in our ecosystem, it's almost entirely
> AI generated in 3 weeks total, and looks (untested) and claims to be
> competitive with current scheme implementations, and Guix aside, has the
> breadth of a good part of the Guile ecosystem.
>
> I'm currently myself a (much much lighter) user, and want to stand by
> the values of free software ; but I'm not sure what is the right way to
> stand up for them anymore, in this context.
>
> I'm still excited to contribute to projects I value, but I lost my
> vision / theory of action which guided how I spent my time in FOSS.  I
> don't know how I should spend my time anymore.
>
> The conversation in Guix is GCD focussed and is already huge, this is a
> Scheme implementation, but I wanted to share my uncertainties on how our
> small but faithful ecosystem should navigate these troubled times.
>
> --
> Best regards,
> Nicolas Graves
>
>

--000000000000ff8dd80657a4c463
Content-Type: text/html; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

<div dir=3D"auto">It&#39;s interesting since the author of Zig rejects AI g=
enerated code, he insists on coding by himself. But his high quality code i=
s trusted and used to be the product of vibe coding like such a project.<di=
v dir=3D"auto"><br></div><div dir=3D"auto">I was managed to implement r7rs-=
small with my brand new optimized compiler. But I didn&#39;t use LLVM as op=
timizing layer. I saw this Zig Scheme implementation contains LLVM converte=
r. This may reduce large complexity.</div><div dir=3D"auto"><br></div><div =
dir=3D"auto">I wouldn&#39;t say r7rs-small is hard for AI. But I think AI s=
hould try r7rs-large, especially for the testing work.</div><div dir=3D"aut=
o"><br></div><div dir=3D"auto">I&#39;ve been trying total vibe coding in a =
complex C++ project for 1 month. It&#39;s not so simple to generate code wi=
th a functional described prompt. But a bunch of design spec with carefully=
 designed data structures, conventions, exceptions handling logics, and tes=
ting hints. Even so, I feel so tired to discuss with Claude on many corner =
cases, compromises and architecture alignments. But it&#39;s still good to =
enjoy the corner cases checking and auto testing by AI. It&#39;s hard to im=
agine that I can manage to do this project on my own, but it&#39;s true now=
.</div><div dir=3D"auto"><br></div><div dir=3D"auto">However, this is the o=
nly project I use pure vibe coding, yet.=C2=A0</div><div dir=3D"auto">In vi=
be coding, I feel like my hacking-power was drained quickly by the thirsty =
AI. I have to save some for my rest life.</div><div dir=3D"auto"><br></div>=
<div dir=3D"auto">The key of software is robustness and interpretability. T=
he key to Internet application business is fast-iterating, not a key to sof=
tware itself.=C2=A0</div><div dir=3D"auto">I imagine one day, people will f=
ind me on my software issue, and I can tell them the design idea, and why t=
his part works like that, or why this is a bug.=C2=A0</div><div dir=3D"auto=
"><br></div><div dir=3D"auto">After all, the value of software is establish=
ed in the trust to its designer. I respect vibe coders, but I will ask them=
, do you believe in the thing you just generated? To myself, I would say ye=
s, and if they don&#39;t trust me, I will show my design spec docs (it&#39;=
s almost accumulated to a book now) and tell them why.</div></div><br><div =
class=3D"gmail_quote gmail_quote_container"><div dir=3D"ltr" class=3D"gmail=
_attr">On Wed, Jul 8, 2026, 01:24 Nicolas Graves &lt;<a href=3D"mailto:ngra=
[email protected]">[email protected]</a>&gt; wrote:<br></div><blockquote clas=
s=3D"gmail_quote" style=3D"margin:0 0 0 .8ex;border-left:1px #ccc solid;pad=
ding-left:1ex"><br>
<a href=3D"https://www.reddit.com/r/scheme/comments/1upm0sb/kaappi_yet_anot=
her_implementation/" rel=3D"noreferrer noreferrer" target=3D"_blank">https:=
//www.reddit.com/r/scheme/comments/1upm0sb/kaappi_yet_another_implementatio=
n/</a><br>
<br>
I don&#39;t know how to react to this in our ecosystem, it&#39;s almost ent=
irely<br>
AI generated in 3 weeks total, and looks (untested) and claims to be<br>
competitive with current scheme implementations, and Guix aside, has the<br=
>
breadth of a good part of the Guile ecosystem.<br>
<br>
I&#39;m currently myself a (much much lighter) user, and want to stand by<b=
r>
the values of free software ; but I&#39;m not sure what is the right way to=
<br>
stand up for them anymore, in this context.<br>
<br>
I&#39;m still excited to contribute to projects I value, but I lost my<br>
vision / theory of action which guided how I spent my time in FOSS.=C2=A0 I=
<br>
don&#39;t know how I should spend my time anymore.<br>
<br>
The conversation in Guix is GCD focussed and is already huge, this is a<br>
Scheme implementation, but I wanted to share my uncertainties on how our<br=
>
small but faithful ecosystem should navigate these troubled times.<br>
<br>
-- <br>
Best regards,<br>
Nicolas Graves<br>
<br>
</blockquote></div>

--000000000000ff8dd80657a4c463--