[PATCH v2 04/10] index: implement notmuch_indexopts_t with try_decrypt

[Jani Nikula]

On Mon, 09 Oct 2017, Daniel Kahn Gillmor <dkg at fifthhorseman.net> wrote:
> On Sat 2017-09-23 19:10:18 +0300, Jani Nikula wrote:
>>> --- a/lib/indexopts.c
>>> +++ b/lib/indexopts.c
>>> @@ -21,9 +21,27 @@
>>>  #include "notmuch-private.h"
>>>  
>>>  notmuch_indexopts_t *
>>> -notmuch_database_get_default_indexopts (notmuch_database_t unused (*db))
>>> +notmuch_database_get_default_indexopts (notmuch_database_t *db)
>>>  {
>>> -    return NULL;
>>> +    return talloc_zero (db, notmuch_indexopts_t);
>>
>> I wonder about the lifetime of indexopts. Should default indexopts be
>> part of the db object, so that your caller above doesn't have to
>> alloc/destroy it for every file?
>
> The caller doesn't have to alloc/destroy it for every file, they can
> alloc it once and pass it in for every file.

My point was, if the caller doesn't do it, it'll get alloced and
destroyed per file.

BR,
Jani.


>
> I'd rather not have the indexopts be part of the db object itself
> explicitly, because i can imagine a longer-running program wanting to
> create two indexopts objects, configuring them differently, and re-using
> one or the other without wanting to modify the database itself.
>
>> Our library interface has a leaky abstraction of the talloc hierarchical
>> refcounting. We don't talk about it in any of the docs, some of it is
>> implied, most of it is completely surprising if the library interface
>> user assumes a traditional C memory allocation model without
>> refcounting.
>
> right, probably the most surprising thing would be if the user got a
> default indexopts from one database object, and then tried to apply it
> to another database object, after having deleted the first database
> object.
>
> However, we *do* talk about it in the docs now, so i think we've given
> fair warning:
>
> -------------
> /**
>  * get the current default indexing options for a given database.
>  *
>  * This object will survive until the database itself is destroyed,
>  * but the caller may also release it earlier with
>  * notmuch_indexopts_destroy.
>  *
>  * This object represents a set of options on how a message can be
>  * added to the index.  At the moment it is a featureless stub.
>  *
>  * @since libnotmuch 5.1 (notmuch 0.26)
>  */
> notmuch_indexopts_t *
> notmuch_database_get_default_indexopts (notmuch_database_t *db);
> -------------
>
>     --dkg