Respect notmuch-config(1) FILES specs.

[ferdinandyb]

WIP

Questions and comments:

• I would make --config supersede any environment variables (the same as the notmuch command itself)
• To actually conform to the specs, I'd need to check at each stage whether the file actually exists and go to the next one if it doesn't, otherwise we'll never get to $HOME/.notmuch-config. I did a quick search of how you do this in C, but as there are multiple ways and nothing really popped out as appropriate, what do you prefer for that? In the end I need a function that returns a true or false :)
• I'd probably add a G_OPTION_ARG_NONE type --debug flag which prints which config file / database is actually used in the end, or maybe at that point maybe the "check file exists" func could print each file it tries, otherwise I think it might get confusing if things don't work as expected.

[aperezdc]

@ferdinandyb I agree that following the same steps as Notmuch to find the condiguration file would be great, thanks for looking into this. That being said, this patch needs a bit of work before we can merge it. I have written a couple of comments with some hints on how to continue.

Looking forward to an updated patch ⚒️

[aperezdc]

You should use g_get_user_config_dir() which will work even if the XDG_CONFIG_HOME environment variable is undefined. Then you don't need to check if g_getenv() returned a non-NULL value.

To check whether the file exists you can use g_file_new_build_filename() to calculate the path and create a GFile for it at the same time, then use g_file_query_exists(). Finally, if the file is present, its full path can be obtained with g_file_get_path().

[aperezdc]

I know the path was already being calculated in this way, but it would be better to use g_build_filename() instead:

Suggested change

	config_path = g_strdup_printf ("%s/.notmuch-config", g_get_home_dir ());
	config_path = g_build_filename (g_get_home_dir (), ".notmuch-config", NULL);

[ferdinandyb]

Thanks for the followup! It is indeed just half-done. Could you also check my question about the config override and the debug flag?

[aperezdc]

Thanks for the followup! It is indeed just half-done. Could you also check my question about the config override and the debug flag?

My two cents:

• Yes, it makes sense that --config is more important and takes precedence.
• I think we do not need a --debug option, you can use g_debug() and by default debug messages are not shown, and setting e.g. G_MESSAGES_DEBUG=all in the environment can then be used to have them shown. As a bonus, GLib will handle adding timestamps, process identifier, and a few other goodies.

[ferdinandyb]

perfect, I'll likely take a week or two to follow up