Dagger Cloud

[ionic cape]

It looks like everything is cached, do you have another dagger shell or other command still running?

Not that it should be cached if you changed your host dir, just checking

There's a known problem involving contextual dir args, lemme see if that could be involved here

urgh yeah:

• +private field https://github.com/sipsma/dagger/blob/c613b8264a7b60c190b2918d075122ddec5acd59/.dagger/engine.go#L27-L27
• that gets initialized w/ a contextual dir: https://github.com/sipsma/dagger/blob/c613b8264a7b60c190b2918d075122ddec5acd59/.dagger/main.go#L51-L51

Working on the fix to that here: https://github.com/dagger/dagger/pull/10175

In the meantime you can get rid of // +private on this line

[stoic idol]

do you have another dagger shell or other command still running
lol many, we're working on mcp servers so i've got at least 2 of those running at any given time, exec'd by gui clients

initially i was like "no, i tried that, all my terminals are closed" but but my mcp configs are all over the place

[ionic cape]

Hmm actually, this doesn't fix it. I think you hit a different version of this problem @stoic idol that doesn't involve +private; it's contextual dirs more generally.

The issue is contextual dirs don't count for the cache key of function calls. I previously assumed this was mostly ok since function calls are cached by the module source digest and it would be extremely obscure to have identical module sources that have different contextual dirs/files. But now that I think about it, the case of:

1. load dagger-dev module
2. change a file/dir not part of the module source (i.e. an integ test)
3. re-load dagger-dev module

Is exactly that... so not obscure at all

Fortunately I was addressing this is my wip dagql cache changes, so I think I may just need to extract a chunk of that out

Thank you for hitting it though @stoic idol 😅

[stoic idol]

here to help

i also just hit another weird x-session persistence thing with llm tool state... that one i have no idea if it's caching or just llm/shell weirdness

https://github.com/dagger/dagger/pull/10118/commits/832807b6b09f3b4fec38e9f667cb2484c0f5e2fd

[ionic cape]

is that PR rebased off https://github.com/dagger/dagger/commit/9eb69cbf8f0d4b419fd63814c400117a8c13c42c ?

[stoic idol]

it is, yes

that one's interesting because it doesn't repro at all manually, so feels like something to do with the connect() helper

[ionic cape]

likely a nested exec thing

I can put that in the queue to look at

[stoic idol]

yeah exactly, lmk if you want a gh issue. i'm hoping to merge that PR with the test workaround and then the repro is just switching around what's commented

[ionic cape]

yeah an issue never hurts, your workaround is fine afaict. If the problem is what I think it is then it's highly specific to nested execs and thus not that relevant to IRL

[ionic cape]

WIP fix here: https://github.com/dagger/dagger/pull/10187, passes the new integ test that repro'd what you were seeing