Hi Ludo!

> “Something” needs to build the file-to-package database (which is what
> you’re working on), and then there needs to be a way for users to fetch
> that database.  This is all orthogonal to substitutes, as I see it,
> which is why I think we need to think about integrating it maybe with
> ‘guix publish’ on the server side and (guix channels) on the client
> side.

If the database is filled on =guix build=, then the substitute server
would automatically have a filesearch database.

Question: How do I hook onto =guix build=?

The only thing left to do after that is to expose the database in =guix
publish= so that the client can fetch it with a =guix pull=.

>>> If we download a fresh database every time, we might as well simply
>>> overwrite the one we have?
>>
>> But then we would miss the database when switching Guix generation.
>
> Ah got it; having a database known to correspond to a specific commit is
> even better.
>
> However, note that it could take time for the server to provide the
> complete database for a commit (the time for as many packages as
> possible to be built), so clients may want to refresh it anyway, or even
> perhaps to use an older database.

Indeed, and that's why I think we need to include a timestamp: if the
client's database timestamp is older than that of the substitute server
database, then we re-fetch it.

Sounds good?

-- 
Pierre Neidhardt
https://ambrevar.xyz/