Search K
Appearance
Appearance
mailbox_search_*()
functions should be used always when you're iterating through multiple messages. The search queries can be complex or as simple as "all messages". Without searching there's also a way to directly switch to a specific message by its sequence number or UID, but this should be avoided usually since they prevent mail prefetching (which can reduce latency).
Search is initialized with mailbox_search_init()
on top of a transaction. Although it would appear that a search is a read-only operation, it can actually write data to Dovecot's cache file. For example if you search for a specific header not yet in cache file, the results are saved to cache file so the next search will be fast. This is why you should always commit search transactions, even if the rest of your operation fails (you should use separate transactions for search and for updates if necessary).
You'll need a search query. src/lib-storage/mail-search.h
and mail-search-build.h
contain all the functions and structures related to it. Usually you should start with mail_search_build_init()
and then start adding the search parameters, either manually or with one of the existing mail_search_build_add_*()
helper functions. The same search query structures can be saved and used for other searches later, but because search state is stored inside the structs you need to be careful that multiple searches aren't using them at the same time. This is usually more trouble than worth, so avoid doing it.
Search results can be sorted by giving a sort program. Dovecot optimizes this by keeping sort indexes in the index files.
While mailbox_search_next()
returns TRUE, a new search result is found and it changes the given mail to point to the search result. The mail's "wanted fields/headers" parameters don't need to include anything needed by the search query, Dovecot optimizes them internally.
If the search needs to parse message bodies and the mailbox is large, this call can take a long time. If you want to do other things while searching, you can use mailbox_search_next_nonblock()
that does only a bit of work and then returns either with a result or "try again later" status. Dovecot attempts to keep each non-matching call to this function between 200 and 250 milliseconds, although the upper bound can't be guaranteed.
It's possible that messages are being expunged while Dovecot is searching them, so it can't determine if they would have matched the search rule or not. In this case it skips over them, but if you want to know if this has happened, you can see if mailbox_search_seen_lost_data()
returns TRUE.
mailbox_search_deinit()
finishes the search. If it returns -1, some error occurred and some search results might not have been returned.
Iterating through all messages in a mailbox goes like:
/* build search query */
search_args = mail_search_build_init();
mail_search_build_add_all(search_args);
search_ctx = mailbox_search_init(trans, search_args, NULL, 0, NULL);
/* search context keeps args referenced */
mail_search_args_unref(&search_args);
while (mailbox_search_next(ctx, &mail)) {
printf("matched uid %u\n", mail->uid);
}
if (mailbox_search_deinit(&search_ctx) < 0)
i_error("search failed");
Search results can be saved for future use by calling mailbox_search_result_save()
just after initializing the search. The results as returned as UIDs with mailbox_search_result_get()
and may contain UIDs that are already expunged. Once you're done with the saved result, free it with mailbox_search_result_free()
.
The search result can also be automatically updated whenever mailbox is synced if MAILBOX_SEARCH_RESULT_FLAG_UPDATE
is set. The update is optimized, so Dovecot doesn't do a full re-search, but matches only new and changed messages. If MAILBOX_SEARCH_RESULT_FLAG_QUEUE_SYNC
is also set, search result additions and removals are also tracked and can be retrieved with mailbox_search_result_sync()
, i.e. with this you can implement "what changed in search results since last time I checked".