GeezerVison

Can you avoid having an LLM doing those suggestions please?

Sure, AFK today but can do that tonight. Emacs doesn't work so well on phone. You want PR clean without Claude?

What I (personally) want usually is:

• human interaction (I don't want to communicate with an agent, that's out the question. I don't care whether the text has been written by the AI as long as it's not a C/C without any human review before sending it but I don't want an automated conversation)
• commits that are always reviewed by a human before even committing them; in general this means that the AI is giving you the rough idea or contents but you are responsible for cleaning it up, and addressing things that could help the next review. In this case, I considered the name = entry.name to be un-necessary so reducing the diff only required to keep the same code and inlining that access. It didn't mean to rewrite all the occurrences with entry.name.

Our policy is that AI is a tool that can be used but it's not something meant to generate PRs for users that would later take credit for (I'm not saying you are, I'm saying that just how AI PRs usually go: most of the time, people just want to have a contribution and so they simply leave their agent in automode).

So, the problem generally is not the AI, it's more that whenever there is an LLM, that LLM tends to choose things that are either useless, weird, and that would have been avoided if a human had directly written the lines. If you prefer writing your code through Claude, it's ok, but keep in mind that reviewers are not responsible for guiding the AI, it's the contributor. If we keep telling their agent through our review "no, not like that", it's as if we were directly using an agent in the least efficient way, and hence we are both losing our time.

Ah. FWIW, Claude pointed out that your suggestion resulted in a NameError at string concat below, so it proposed this as easiest way to avoid name global and I accepted it

Oh, my bad actually it should have been displayname = linkname = name = entry.name. But yeah, in this case, I would have preferred to know why it decided to do so. And then I would have suggested:

displayname = linkname = name = entry.name

instead of the two lines. That's why I prefer that you tell me this first before changing otherwise I would get confused.

So to reiterate: there is nothing wrong with using Claude (the only think I don't like about it is the long summaries in every PR and trivial things it reports like "all tests pass" which is something I would expect for a PR otherwise it doesn't make sense). It's just that having a 3rd actor usually creates more back and forth between the reviewer and the reviewee.