Re: [PATCH] Add option `hooks.path` for setting the directory of hooks.

Subject: Re: [PATCH] Add option `hooks.path` for setting the directory of hooks.

Date: Tue, 30 Aug 2016 20:51:27 +0300

To: Erik Rybakken

Cc: Jani Nikula,, David Bremner

From: Tomi Ollila

On Tue, Aug 30 2016, Tomi Ollila <> wrote:

> On Tue, Aug 30 2016, Tomi Ollila <> wrote:
>> On Sat, Aug 27 2016, Erik Rybakken <> wrote:
>>> Hi,
>>> Thanks Tomi and David for the feedback!
>>> On Fri, Aug 26, 2016 at 02:32:19PM +0300, Tomi Ollila wrote:
>>>> ... but I can think of one problem there (if my memory server correctly)
>>> Yeah, I didn't think of that. I have been thinking about how to make the
>>> generated configuration show only the options that differ from the
>>> default, and have the default options commented out, but it got a bit too
>>> involved for me.
>>> However, I think I have another solution, and I included a updated patch
>>> for this. There is only one (*) difference from the first patch:
>>> When reading the configuration file and "hooks.path" is either unset or
>>> set to an empty string, we set config->hooks_path to be the expanded
>>> path "database.path/.notmuch/hooks", but we set the value of hooks.path in
>>> config->key_file to be an empty string. That way, when calling
>>> "notmuch_config_get_hooks_path", the expanded path gets returned, but when
>>> saving the config file, either from "notmuch setup" or "notmuch config set",
>>> the value will still be an empty string, given that it wasn't changed.
>>> I believe this is the easiest fix, and if this sounds good, I will start
>>> working on the tests.
>> I kinda like how this would work...
>> The code looked pretty good -- when did I git am to the email content
>> I got all from the beginning of this email to the commit message --
>> so before next patches use git-format-patch and git-am... Check
>> for more information...
>> 2 things that came up after quick view
>> 2) I wonder whether calling notmuch_config_get_hooks_path() could
>> be "lazier".... ARGH no :( -- it would make notmuch_config_get_hooks_path()
>> have different code than others and ... (**)
> That said, perhaps 
> const char *
> notmuch_config_get_hooks_path (notmuch_config_t *config)
> {
>      const char * hooks_path =
>          _config_get (config, &config->hooks_path, "hooks", "path");
>      if (hooks_path == NULL || hooks_path[0] == '\0') {
>         hooks_path = talloc_asprintf (config, "%s/.notmuch/hooks",
>                                       notmuch_config_get_database_path (config));
>         _config_set(config, &config->hooks_path, "hooks", "path", hooks_path);
>      }
>      return hooks_path;
> }
> But, it takes quite a bit of careful examination to check whether that works
> as expected, and I always think whether some accidental fragileness there
> causes that stored value to be dumped to the configuration file (now, or
> in later changes).

I forgot that this risk can be mitigated by a set of suitable tests...

> But, the above may be useless crap -- just I don't have more time to check
> that out now...