Eliminating Timeout

[flatcap]

Current State

NeoMutt uses getch() to get a keypress from the user.
$timeout determines how long ncurses should wait for a response.

If that timeout is reached, then OP_TIMEOUT is returned to the caller, e.g. Index.
The Index's Event Loop can now do some background tasks, like checking for new mail.

But... there are lots of dialogs.

NeoMutt wants to check for new mail in the Index, Pager and Browser, created duplicated code.
Also, IMAP doesn't want its connection to timeout -- more important than the new mail checks.
To make sure that this happens, wherever the user happens to be, means some real ugliness.

Source: graphviz svg

• dlg_index() - Index Event Loop wants an event
• km_dokey_event() - It will be interpreted using MENU_INDEX
• mutt_getch() - Check for queued keys/macros
• getch() - Get a character from the user (ncurses)

---

mutt_getch() has a lot to do.

It handles the key and macro queue, SIGINT (Ctrl-C) and the Abort Key (Ctrl-G).
If inotify is enabled (Linux only), then it's given a chance to poll for updates.

inotify doesn't belong here.

---

km_dokey_event() has a lot to do, too.

Its main job is to convert keys d into functions <delete>.
This involves a looking in global tables (see also #3916).
There's also some magic to lookup macros.

$timeout is handled here.
It determines how long ncurses blocks waiting for a keypress.

It also calls imap_keep_alive().
This really doesn't belong here.

Who Needs the Timeout?

There are four users that need a timeout.
They need to be called regardless of which dialog the user is in.
Ideally, they want to be called when nothing else is happening.

• Imap keep alive
• timeout-hook
• Checking for new mail / stats
• inotify

Anything else, like redrawing the screen should be triggered by other events.

A Global Solution

My solution is a global timeout notification.
In mutt_getch(), the curses timeout would be set to 1 second.
If ncurses returns ERR (no key waiting), then a notification gets sent.

Each recipient would register an observer.

IMAP can handle its own timing for the keep alive message.

Checking for new mail currently happens in the Index, Pager and Browser.
Making it happen globally, means changing the new_mail_command to respond to events.

The timeout-hook already happens at a global level -- unaware of the current Mailbox.

inotify doesn't involve the keys and communicates via global variables, so using an observer is easy.

[flatcap]

Checking for new mail / stats

Hmm... this is getting interesting.

Consider the default config:

set mail_check = 5
set mail_check_stats_interval = 60
set imap_keep_alive = 300
set pop_check_interval = 60

Every 5s do a quick check for new mail in all mailboxes.
Every 60s do full check for stats in all mailboxes.
Ping the IMAP server every 5 minutes.
Check for POP mail every minute.

Sounds good until you consider the default for $timeout -- 10 minutes!

set timeout = 600

This means all the other checks no occur at 10 minute intervals.
The only rational choice for $timeout is 1 second.

---

The Event Loops in Index and Pager assume that when mutt_getch() returns, a timeout must have occurred.
Currently, resizing the screen will send a flurry of mailbox checks.

Eliminating $timeout from mutt_getch() means moving the checks upwards.
But where should the common config be checked?

• $mail_check
• $mail_check_stats_interval

They're currently checked in mutt_mailbox_check(), in the front-end.
This means $pop_check_interval cannot be lower (more frequent) than $mail_check.

[flatcap]

Completed in 0cf1ab9