lib/messages.c: Make message searches stream as well.

Xapian provides an interator-based interface to all search results.
So it was natural to make notmuch_messages_t be iterator-based as
well. Which we did originally.

But we ran into a problem when we added two APIs, (_get_replies and
_get_toplevel_messages), that want to return a messages iterator
that's *not* based on a Xapian search result. My original compromise
was to use notmuch_message_list_t as the basis for all returned
messages iterators in the public interface.

This had the problem of introducing extra latency at the beginning
of a search for messages, (the call would block while iterating over
all results from Xapian, converting to a message list).

In this commit, we remove that initial conversion and instead provide
two alternate implementations of notmuch_messages_t (one on top of a
Xapian iterator and one on top of a message list).

With this change, I tested a "notmuch search" returning *many* results
as previously taking about 7 seconds before results started appearing,
and now taking only 2 seconds.

diff --git a/lib/messages.c b/lib/messages.c
index 2f7c283ecbb220a3fcf6dca033efced805791335..54c0ab07805ab33b171a2107a7ac7196cc2e5ca5 100644 (file)
--- a/lib/messages.c
+++ b/lib/messages.c
@@ -20,13 +20,6 @@
 
 #include "notmuch-private.h"
 
-#include <glib.h> /* GList */
-
-
-struct _notmuch_messages {
-    notmuch_message_node_t *iterator;
-};
-
 /* Create a new notmuch_message_list_t object, with 'ctx' as its
  * talloc owner.
  *
@@ -85,20 +78,45 @@ _notmuch_messages_create (notmuch_message_list_t *list)
     if (unlikely (messages == NULL))
        return NULL;
 
+    messages->is_of_list_type = TRUE;
     messages->iterator = list->head;
 
     return messages;
 }
 
+/* We're using the "is_of_type_list" to conditionally defer to the
+ * notmuch_mset_messages_t implementation of notmuch_messages_t in
+ * query.cc. It's ugly that that's over in query.cc, and it's ugly
+ * that we're not using a union here. Both of those uglies are due to
+ * C++:
+ *
+ *     1. I didn't want to force a C++ header file onto
+ *        notmuch-private.h and suddenly subject all our code to a
+ *        C++ compiler and its rules.
+ *
+ *     2. C++ won't allow me to put C++ objects, (with non-trivial
+ *        constructors) into a union anyway. Even though I'd
+ *        carefully control object construction with placement new
+ *        anyway. *sigh*
+ */
 notmuch_bool_t
 notmuch_messages_has_more (notmuch_messages_t *messages)
 {
-    return (messages != NULL && messages->iterator != NULL);
+    if (messages == NULL)
+       return FALSE;
+
+    if (! messages->is_of_list_type)
+       return _notmuch_mset_messages_has_more (messages);
+
+    return (messages->iterator != NULL);
 }
 
 notmuch_message_t *
 notmuch_messages_get (notmuch_messages_t *messages)
 {
+    if (! messages->is_of_list_type)
+       return _notmuch_mset_messages_get (messages);
+
     if (messages->iterator == NULL)
        return NULL;
 
@@ -108,6 +126,9 @@ notmuch_messages_get (notmuch_messages_t *messages)
 void
 notmuch_messages_advance (notmuch_messages_t *messages)
 {
+    if (! messages->is_of_list_type)
+       return _notmuch_mset_messages_advance (messages);
+
     if (messages->iterator == NULL)
        return;
 

diff --git a/lib/notmuch-private.h b/lib/notmuch-private.h
index 2b4bf319cf8c3e683ca4c715f377c0a33c0a038d..d3f9a4c4ba4727c0860804ac30a0ddfd883565b6 100644 (file)
--- a/lib/notmuch-private.h
+++ b/lib/notmuch-private.h
@@ -290,6 +290,15 @@ typedef struct _notmuch_message_list {
     notmuch_message_node_t **tail;
 } notmuch_message_list_t;
 
+/* There's a rumor that there's an alternate struct _notmuch_messages
+ * somewhere with some nasty C++ objects in it. We'll try to maintain
+ * ignorance of that here. (See notmuch_mset_messages_t in query.cc)
+ */
+struct _notmuch_messages {
+    notmuch_bool_t is_of_list_type;
+    notmuch_message_node_t *iterator;
+};
+
 notmuch_message_list_t *
 _notmuch_message_list_create (const void *ctx);
 
@@ -304,6 +313,17 @@ _notmuch_message_list_add_message (notmuch_message_list_t *list,
 notmuch_messages_t *
 _notmuch_messages_create (notmuch_message_list_t *list);
 
+/* query.cc */
+
+notmuch_bool_t
+_notmuch_mset_messages_has_more (notmuch_messages_t *messages);
+
+notmuch_message_t *
+_notmuch_mset_messages_get (notmuch_messages_t *messages);
+
+void
+_notmuch_mset_messages_advance (notmuch_messages_t *messages);
+
 /* message.cc */
 
 void

diff --git a/lib/query.cc b/lib/query.cc
index 7d191e529ec72974a2ec5adad21fe63e87e2942f..a571a618bdd1350aa7041487af19cdf788834c36 100644 (file)
--- a/lib/query.cc
+++ b/lib/query.cc
@@ -31,11 +31,12 @@ struct _notmuch_query {
     notmuch_sort_t sort;
 };
 
-struct _notmuch_messages {
+typedef struct _notmuch_mset_messages {
+    notmuch_messages_t base;
     notmuch_database_t *notmuch;
     Xapian::MSetIterator iterator;
     Xapian::MSetIterator iterator_end;
-};
+} notmuch_mset_messages_t;
 
 struct _notmuch_threads {
     notmuch_query_t *query;
@@ -75,19 +76,42 @@ notmuch_query_set_sort (notmuch_query_t *query, notmuch_sort_t sort)
     query->sort = sort;
 }
 
+/* We end up having to call the destructors explicitly because we had
+ * to use "placement new" in order to initialize C++ objects within a
+ * block that we allocated with talloc. So C++ is making talloc
+ * slightly less simple to use, (we wouldn't need
+ * talloc_set_destructor at all otherwise).
+ */
+static int
+_notmuch_messages_destructor (notmuch_mset_messages_t *messages)
+{
+    messages->iterator.~MSetIterator ();
+    messages->iterator_end.~MSetIterator ();
+
+    return 0;
+}
+
 notmuch_messages_t *
 notmuch_query_search_messages (notmuch_query_t *query)
 {
     notmuch_database_t *notmuch = query->notmuch;
     const char *query_string = query->query_string;
-    notmuch_message_list_t *message_list;
-    Xapian::MSetIterator i;
+    notmuch_mset_messages_t *messages;
 
-    message_list = _notmuch_message_list_create (query);
-    if (unlikely (message_list == NULL))
+    messages = talloc (query, notmuch_mset_messages_t);
+    if (unlikely (messages == NULL))
        return NULL;
 
     try {
+
+       messages->base.is_of_list_type = FALSE;
+       messages->base.iterator = NULL;
+       messages->notmuch = notmuch;
+       new (&messages->iterator) Xapian::MSetIterator ();
+       new (&messages->iterator_end) Xapian::MSetIterator ();
+
+       talloc_set_destructor (messages, _notmuch_messages_destructor);
+
        Xapian::Enquire enquire (*notmuch->xapian_db);
        Xapian::Query mail_query (talloc_asprintf (query, "%s%s",
                                                   _find_prefix ("type"),
@@ -130,22 +154,8 @@ notmuch_query_search_messages (notmuch_query_t *query)
 
        mset = enquire.get_mset (0, notmuch->xapian_db->get_doccount ());
 
-       for (i = mset.begin (); i != mset.end (); i++) {
-           notmuch_message_t *message;
-           notmuch_private_status_t status;
-
-           message = _notmuch_message_create (message_list, notmuch,
-                                              *i, &status);
-           if (message == NULL)
-           {
-               if (status == NOTMUCH_PRIVATE_STATUS_NO_DOCUMENT_FOUND)
-                   INTERNAL_ERROR ("A message iterator contains a "
-                                   "non-existent document ID.\n");
-               break;
-           }
-
-           _notmuch_message_list_add_message (message_list, message);
-       }
+       messages->iterator = mset.begin ();
+       messages->iterator_end = mset.end ();
 
     } catch (const Xapian::Error &error) {
        fprintf (stderr, "A Xapian exception occurred performing query: %s\n",
@@ -154,7 +164,55 @@ notmuch_query_search_messages (notmuch_query_t *query)
        notmuch->exception_reported = TRUE;
     }
 
-    return _notmuch_messages_create (message_list);
+    return &messages->base;
+}
+
+notmuch_bool_t
+_notmuch_mset_messages_has_more (notmuch_messages_t *messages)
+{
+    notmuch_mset_messages_t *mset_messages;
+
+    mset_messages = (notmuch_mset_messages_t *) messages;
+
+    return (mset_messages->iterator != mset_messages->iterator_end);
+}
+
+notmuch_message_t *
+_notmuch_mset_messages_get (notmuch_messages_t *messages)
+{
+    notmuch_message_t *message;
+    Xapian::docid doc_id;
+    notmuch_private_status_t status;
+    notmuch_mset_messages_t *mset_messages;
+
+    mset_messages = (notmuch_mset_messages_t *) messages;
+
+    if (! _notmuch_mset_messages_has_more (&mset_messages->base))
+       return NULL;
+
+    doc_id = *mset_messages->iterator;
+
+    message = _notmuch_message_create (mset_messages,
+                                      mset_messages->notmuch, doc_id,
+                                      &status);
+
+    if (message == NULL &&
+       status == NOTMUCH_PRIVATE_STATUS_NO_DOCUMENT_FOUND)
+    {
+       INTERNAL_ERROR ("a messages iterator contains a non-existent document ID.\n");
+    }
+
+    return message;
+}
+
+void
+_notmuch_mset_messages_advance (notmuch_messages_t *messages)
+{
+    notmuch_mset_messages_t *mset_messages;
+
+    mset_messages = (notmuch_mset_messages_t *) messages;
+
+    mset_messages->iterator++;
 }
 
 /* Glib objects force use to use a talloc destructor as well, (but not