Re #2485.

Unfortunately this implementation of org-roam-db results in endless recursion when file doesn't exist. When org-roam-db--setup checks if file exists, it is already created by emacsql.

Here's the updated version that works:

(defun org-roam-db ()
  "Entrypoint to the Org-roam sqlite database.
Initializes and stores the database, and the database connection.
Performs a database upgrade when required."
    (or (and-let* ((conn (org-roam-db--get-connection)))
          (and (emacsql-live-p conn) conn))
        (puthash (expand-file-name (file-name-as-directory org-roam-directory))
                 (emacsql-sqlite-open org-roam-db-location nil
                                      (if (file-exists-p org-roam-db-location)
                                          #'org-roam-db--setup
                                        #'org-roam-db--init))
                 org-roam-db--connection)))

emacsql-sqlite-open calls its SETUP function with the database as its only argument and expects it to only access the db via that binding. When I wrote 118be8a I did not notice, that, while none of the functions called in the old org-roam-db do that directly, they do call functions that in turn use org-roam-db, leading to this infinite loop.

I.e., we have to give up on using the SETUP argument, which means that Org-Roam doesn't follow the (refined) conventions. Alternatively the problematic functions could be changed to take the database as an optional argument, but I don't understand enough about what Org-Roam is doing here (and why), so it would be too risky for me to do it.

I've pushed an additional commit to address the compiler warning (which are unrelated to the changes in the other commits). Normally that should be done in a separate pull-request, but due to the radio silence, I have chosen to append that here, so I don't have to juggle with two branches when/if I decide to play with Org-Roam.

I think that org-roam-db--setup is for setting up database when it already exists and org-roam-db--init for its first creation, so if we pass the correct one, that should work. Unfortunately the org-roam-db--setup checks if upgrade is needed and that uses (org-roam-db-sync t) which actually closes and reopens the connection. While emacql approach is fine we might want to rewrite the setup function so that it detects whether an update is needed and postpones it until after return from setup/open.

@tarsius I just merged a couple of related changes in #2503 that do just enough to get CI passing. Would you be willing to update this branch based off of latest main?

I've rebased, but note that I have not tested, since I don't use this package.

It is working well for me locally. Thank you so much for this thorough PR! ❤️

Thanks for reviving this package!