move more initialization code into server core class

[totaam]

Issue migrated from trac ticket # 1123

component: server | priority: major | resolution: fixed

2016-02-17 08:35:14: antoine created the issue

---

This will allow us to cleanup the code somewhat and avoid issues like #1122, or at least solve them more cleanly.
The server knows if it is doing an upgrade, exit or stop - so the exit code stuff can go away too.

[totaam]

2016-07-12 17:52:23: antoine commented

---

Milestone renamed

[totaam]

2016-09-15 03:59:33: antoine changed status from new to assigned

[totaam]

2016-09-15 03:59:33: antoine commented

---

Too late for this release (likely to cause tricksy bugs).

[totaam]

2016-09-21 08:56:36: antoine commented

---

This would help simplify the code used for remote ssh start: we could just start a "plain" server and let the client tell the server what it wants without using command line arguments. (better cross version compatibility)
See #1319 which does this for starting sessions via the proxy.

[totaam]

2018-08-21 17:46:15: antoine commented

---

See also #1838, #1943

[totaam]

2018-09-04 04:29:00: antoine commented

---

r20286 moves in this direction.

[totaam]

2019-03-04 02:56:25: antoine commented

---

Blocker for #2187

[totaam]

2019-05-07 13:43:43: antoine commented

---

Preparatory work in r22652.

This is really hard: we can't instantiate the server class early because it (indirectly) loads GTK and X11 bindings, and for those we need to have initialized the vfb, which requires the following components to have been initialized too:

• pam
• daemonization
• logging / directories creation
• uinput virtual devices
• dbus
  etc..
  Even within those components, there are dependencies: uinput before pam, etc..

[totaam]

2019-05-08 09:54:07: antoine commented

---

Updates:

• r22654 + r22655: move some initialization / validation code to methods
• r22656 move vfb resizing to server_init()
• r22659 move dbus bits
• r22660 move dbus kill and property handling to server

Other undocumented rules I have (re)discovered:

• dbus before gtk3
• sockets before vfb?

[totaam]

2019-05-08 12:27:08: antoine commented

---

Updates:

• r22661 hide ugly gtk2 vs gtk3 code
• r22662 fixup for r22660
• r22663 import cleanup (+r22664 fixup)
• r22665 move kill xvfb to x11 server code

[totaam]

2019-05-09 12:05:48: antoine commented

---

Updates:

• r22666 close gtk display on exit now for all types of servers (+r22667 and r22668 fixups)
• r22669 always fsync and close the pipe when using displayfd
• r22670 + r22674: move socket cleanup code to the server
• r22671 fix --bind-ssl socket errors on start (backport in 22672)
• r22676 ssl setup moved
• r22677 fixup dbus refactoring

Still TODO:

• derive mdns data from socket info and move mdns setup code to server
• find a way to move display related stuff into the server class, delaying the actual bindings initialization until we have a display to connect to.. (hard)

[totaam]

2019-05-09 16:09:13: antoine commented

---

Updates:

• r22678: close_gtk_display
• r22680: pidfile
• r22679 + r22681 + r22682: minor cleanups

[totaam]

2019-05-09 18:17:32: antoine changed status from assigned to closed

[totaam]

2019-05-09 18:17:32: antoine set resolution to fixed

[totaam]

2019-05-09 18:17:32: antoine commented

---

Updates:

• mdns done in r22683

Moving the display stuff to #2294.

This will do for 3.0.