Ellama attempts to kill current buffer when there is a backend error

[sunng87]

Steps to reproduce:

1. Configure ollama to an unreachable host
2. Start ellama-ask-line and input the question

Emacs will kill current buffer or prompt for save if it's dirty.

[tusharhero]

Can confirm this, is this the intended behaviour @s-kostyaev ?

[s-kostyaev]

No, it's a bug. I will try to fix it soon.

[s-kostyaev]

@ahyatt looks like it's a bug in an llm library. Steps to reproduce:

(llm-chat-streaming (make-llm-ollama :chat-model "llama3" :host "unreacheable")
		    (llm-make-chat-prompt "test") #'ignore #'ignore #'ignore)

[ahyatt]

Interesting find. @r0man, can you take a look at this? After some initial investigation, plz--kill-buffer is killing the wrong buffer. I think you should have a better idea of why this is happening than I do. It probably does not have to do with streaming, I can also reproduce with llm-chat-async.

[r0man]

Hi @s-kostyaev and @ahyatt,

this is a know issue, that @leotaku reported a while ago. He also contributed a fix:
r0man/plz-media-type#14

I just tried the snippet @s-kostyaev provided and I think it solves the issue. Since I believe the PR counts as a non-trivial contribution we are waiting for Leo's paperwork to complete. He told be that he did all of it and the ball is at the FSF right now. So I expect the paperwork to be complete in the next weeks or so.

Is that good enough for you? Should we do something else?

Roman

[s-kostyaev]

Thank you @r0man

I think we need to wait.

[r0man]

Hi @s-kostyaev and @ahyatt ,

this should be fixed now. I cut a release which should be on ELPA soonish.
https://github.com/r0man/plz-media-type/releases/tag/v0.2.3