completely skip server base classes

[Antoine Martin]

Similar to #1836 but for mixins (#1778 / #1761), we should be able to completely skip some of the server base classes.
ie:

• if remote-logging is disabled, don't inherit from LoggingServer
• if mmap is disabled, don't inherit from MMAP_Server

etc

This would reduce the memory footprint, and increase the security (decreasing the attack surface): it is impossible to attack code which isn't there.

Could be done for the server by using a dynamic type for "server base" (see example patch).
The client is not modular enough to support this sort of refactoring. (see #1796 for authentication handlers)

[Antoine Martin]

dynamic list of base classes

[Antoine Martin]

Huge commits making things modular in r19283 + r19282.

This may help with #1123, we just need a new "upgrade" argument for the cleanup method.

[Antoine Martin]

As of r19284, the server can completely skip a number of modules:

xpra start --notifications=no --webcam=no --speaker=no --microphone=no \
    --file-transfer=no --printing=no --dbus-proxy=no \
    --remote-logging=no --mmap=no --clipboard=no --av-sync=no

(this will disable most of the optional mixins, both in the server base class and in the "client connection" instance)

Still TODO:

• add test script running every combination possible (there are 2**10=1024!) to check for invalid attribute dependencies
• ideally, deal with the "mmap_size" check more cleanly - meh
• add switch for "idle", "client info" mixins
• make control commands mixin optional (some subclasses assume this is always defined)
• make it possible to disable the "commands", "encoding" and "network state" switches? (hard, especially for "encoding")
• this probably caused #1841

[Antoine Martin]

Regression: #1841, fixup: r19363

[Antoine Martin]

Automated test script added in r19371, with some bug fixes included.

This script exposes lots of ugly dependency issues. (in particular with read-only mode, keyboard, etc)
So many more bugs are left to fix..

[Antoine Martin]

Updates:

• r19373 + r19381 + r19388 + r19389 + r19467: fix readonly mode
• r19400 skip shadow pointer polling when input devices are disabled
• r19374 client bug when the desktop-size is not provided
• r19375 fix wrong location for packet-queue
• r19401 window model management refactoring
• r19376 improve client warning message when the server does not forward any windows (--windows=no)
• r19440 move encode work queue and thread to superclass (used by both clipboard and picture encoding)
• r19379 + r19395 + r19406 major unit tests improvements
• minor fixes: r19380, r19407, r19413, r19439, r19443, r19446 + r19447 + r19450, r19462, r19476, r19484
• r19410 + r19452: audio vs av-sync fix
• r19414 + r19415 + r19416 + r19417 + r19441 + r19442 + r19451 + r19482 + r19483 + r19485: run shadow unit tests on win32 and macos
• r19475: test desktop and shadow servers with vnc client
• r19448 + r19454: make it easier to run without first compiling with cython + installing
• r19460 + r19461 + r19463 + r19469 + r19471 + r19473: rfb fixes
• r19486: run fewer tests by default (slows down the build too much)
• r19599 actually add test for "start-desktop"

Still TODO:

• send_cursor might belong in the display rather than windows mixin?
• the "desktop_size_unscaled" / "desktop_size" code in x11 server base is a bit ugly
• same for "suspended" attribute used in source mixins: r19409
• keys_pressed should be moved to the keyboard config: r19466
• maybe verify that the window we forward is on the client's vfb screen (checking pixel colour at window location)
• use super() more
• add new build options to exclude some subsystems completely and avoid failures when they cannot be imported
• batch delay and other data is missing from session info

[Antoine Martin]

For the client, see #1861

[Antoine Martin]

Ugly dependency utility cursor functions added: ticket:1658#comment:8.

[Antoine Martin]

• as of r19723 + r19724, we can run the server with some modules completely missing. Either not installing them (ie: build with --without-dbus - new switches added in r19725) or by nuking them afterwards (ie: rm -fr /usr/lib64/python2.7/site-packages/xpra/keyboard/).

This is true for the following xpra submodules: notifications, keyboard, clipboard, sound and dbus.
The server runs but prints a warning if the settings or command line try to enable functionality that is not available, ie:

Warning: missing notifications module

• r19726: the server can run without any codecs at all, this also means without any webcam or window forwarding
• r19727: the client avoids sending keyboard and pointer events if the server doesn't know how to handle them
• r19728: the client still runs without any of the optional modules (but those are still loaded - not the same as #1861)

[Antoine Martin]

@maxmylyn: FYI, feel free to close. (see comment:8)

Will follow up in #1913 (high memory usage).

[J. Max Mena]

Noted and closed.

[Antoine Martin]

r20282 does the same for RFB in shadow and desktop servers.

[Norman Rasmussen]

Would you backport a fix to the v2.3.x branch (and maybe others?) to include calls to IdleMixin and ClipboardConnection in ClientConnection.get_info? I noticed that the data is missing :-(

[Antoine Martin]

Would you backport a fix to the v2.3.x branch (and maybe others?) to include calls to IdleMixin and ClipboardConnection in ClientConnection.get_info? I noticed that the data is missing :-(

Done in r20284.
@norman: does that work for you?

[Antoine Martin]

See also #2344 and #2351

[migration script]

this ticket has been moved to: ​https://github.com/Xpra-org/xpra/issues/1838