Computing `total_messages` isn't cheap but done on every keypress

[dominikh]

Describe the bug

My notmuch database contains almost 2 million messages. Any alot buffer with a status bar, such as a thread, is unusably slow, with any keypress lagging noticeably.

According to perf, 88% of time is spent in _notmuch_query_count_documents. Applying the following hack removes this source of slowness:

diff --git i/alot/ui.py w/alot/ui.py
index 4c44809a..311ff4bf 100644
--- i/alot/ui.py
+++ w/alot/ui.py
@@ -677,7 +677,7 @@ class UI:
             btype = cb.modename
             info['buffer_no'] = self.buffers.index(cb)
             info['buffer_type'] = btype
-        info['total_messages'] = self.dbman.count_messages('*')
+        # info['total_messages'] = self.dbman.count_messages('*')
         info['pending_writes'] = len(self.dbman.writequeue)
         info['input_queue'] = ' '.join(self.input_queue)

Unfortunately, just removing total_messages from all statusbar settings isn't sufficient, because the information gets calculated unconditionally.

I believe that count_messages runs in O(n) time and has to iterate over the result set and can't trivially retrieve the count.

Another way of reproducing it is with notmuch count "*", which finishes in 270ms on my good system and in 1s on my weak system, obviously not something you want to run on every keypress.

Software Versions

• Python version: 3.12.10
• Notmuch version: 0.40
• Alot version: 40c57d6, also present in v0.11

[lucc]

I can understand but not reproduce the issue because I only have ~50k messages and "notmuch count" runs in ~34ms. Still I think it is worth it to do something about this.

If we remove the computation from the code that runs on every key stroke the question is: do we compute it on buffer load and cache it (so it is an approximate value) or do we just remove it (then the statusbar would just say "x messages"). I am fine with both solutions.