Indeed. Somehow, when I gave the agent the link and asked it to update
the patch it decided it made more sense to recreate the hunks manually
instead of just downloading the patch exactly. I thought the AIs
weren't
so dumb.
In my opinion, humans are still responsible (and still need to be
capable) to oversee and understand what any "AI" is doing.
And, I personally don't understand why you need "AI" to copy a patch
file?
I don't *need* it, it's just easier that way. I'm just reducing some
cognitive load.
And I'm capable. I just assumed it cannot trip on such a basic task,
when the agents are capable of so much more.
What would you have done in this circumstance? Just copy pasted the
whole mail into a .patch file?
I'm not very used to email workflows when gcc is the only project I ever
use them for.
If you want a more impressive example, the patch for this branch was
generated in one shot.
https://github.com/skeeto/w64devkit/pull/41
I frequently debug and patch all sorts of issues in the toolchain using
agents. Including this one.
Even if what I suggested wasn't perfect, figuring out the cause myself
without using an agent would have been very difficult. And at the very
least the generated patch worked for my own circumstances.
The good thing that I got out of this is that I learned to use your
'Dockerfile' locally with Podman, and learned to build Windows-hosted
GCC, and run under Wine that one as well as the resulting
multi-threaded
GCC/OpenMP executable.
So: we're good again. :-)
That's great. I've wasted weeks before trying to replicate this outside
of docker before, so I can use a trunk gcc clone.
I then got an agent to write a script that does the whole process on
Linux end-to-end, and it has spared me a ton of effort.
Maybe you'd be interested in it too.
https://gist.github.com/Peter0x44/9aca4f3be97e74dcae004a967da360a9
Thanks for everything,
Peter