(replying to felix because it seems Stanislav's email was not
replied-to-all and I can't see it in my inbox, only in the archives)

> When people submit patches they did not write to the code they don’t
understand, the burden on maintainers increases tremendously.

100% agree with you. I think the key here is to differentiate between AI
slop and AI-assisted coding by someone that could understand and vet the
output. IE was AI used as a proper tool to get to an outcome faster? or was
it used blindly to just generate something without care?

Best,
Rolando

On Mon, Feb 9, 2026 at 2:17 AM Felix Winkelmann <[email protected]>
wrote:

> (At Peter's suggestion I forward this to the mailing list, as the
> discussion may
> be of interest to other contributors. We don't have a policy regarding LLM
> usage yet, but need to make up our minds on how we approach this in
> general.
>
> Note that this does _not_ mean a lack of trust in you or your
> contribution, but
> reviewing LLM-assisted code is a new and somewhat unsettling experience to
> me, and
> I'm not quite sure how we want to handle this in the future)
>
> On Mon Feb 9, 2026 at 5:07 AM CET, Rolando Abarca wrote:
> > Hi Felix,
> >
> > Thanks for taking the time to review this!
> >
> > I'll look into the code cleanup suggestions, the points about removing
> the
> > unnecessary #ifdefs, the _VAL macros, and the redundant clause in
> > url-protocol all make sense. I'll also look at the error handling in
> > make-ssl-input-port and the static build approach you suggested.
> > I really like the idea of including and using cond-expand to add it or
> not.
> > I'll play with that.
> >
> > To answer your question: yes, I used LLM assistance (Claude Code) for the
> > initial research to understand the codebase and figure out what the
> > implementation would entail. From there I created a manual plan for how
> to
> > approach it and had the AI run with it. There were a few manual fixes
> along
> > the way, but to my non-expert eyes it looked good enough to share as a
> > proof of concept.
> >
> > I appreciate the thorough review, I'll work through your suggestions and
> > follow up.
> >
> > Cheers,
> > Rolando Abarca
> >
> >
> > On Sun, Feb 8, 2026 at 3:42 AM Felix Winkelmann <
> [email protected]>
> > wrote:
> >
> >> Hi, Rolando!
> >>
> >> I finally got around to reviewing this, sorry again for the long delay.
> >> I have a few questions and suggestions regarding the code.
> >>
> >> Always building the ssl module in dynamic mode is problematic when
> >> CHICKEN is built statically. It would be better to compile it
> conditionally
> >> and link it to chicken-install (excusively). An alternative would be
> >> to "include" it (similar to egg-download.scm). You could then avoid the
> >> eval-trick, for example by using cond-expand, after adding a feature
> >> to the build of chicken-install.
> >>
> >> SSL support needs to be activated explicitly, I guess it would be fine
> >> to enable it by default, if the configure script succeeds in testing
> >> for ssl availability, since your checks are quite thorough.
> >>
> >> In egg-download.scm, the 2nd clause in url-protocol seems to be
> >> redundant.
> >>
> >> In ssl.scm, the #ifdefs are not needed (it is built only when SSL
> >> is available, anyway). Also the "..._VAL" macro defs are not needed,
> >> you can use the constants from ssl.h directly. Also, the function
> >> prototype for get_tcp_fd is unnecessary, I think. make-ssl-input-port
> >> does not handle other errors besides the _want_... cases, is this
> >> intentional?
> >>
> >> One question: did you use LLM assisstance for this code? Some of
> >> the redundancies seem to indicate that. That is not indented to
> >> diminish your effort (I'd very much like to add this feature to
> >> CHICKEN), just to make sure with what amount of diligence I should
> >> review the code.
> >>
> >>
> >> cheers,
> >> felix
> >>
> >>
>
>

Reply via email to