13 Feb
2012
13 Feb
'12
7:53 p.m.
On Monday 13 February 2012 13:25:13 Timothy Pearson wrote:
This would have caused the modal system notifications to lose their window
handles/decorations and drag/resize abilities, unless I have been reading
the FreeDesktop WM specifications incorrectly. Example:
http://standards.freedesktop.org/wm-spec/wm-spec-latest.html#id2528706 In the case
you want them to have window decorations, yes override redirect
does not work.
Still it's wrong what you did there :-) It's not how KWin works.
yes I noticed but
that is quite severe as it can result in KWin no longer
starting (not this commit but other ones). So you have to be aware of ABI
compatibility or at least have to take care to not load incompatible
libraries.
Care to elaborate? I am willing to listen. Your
original message was
disregarded to some extent as you linked to changes that were made on
purpose, and I usually expect people who claim something is wrong to
make
an attempt at stating *why( they think said something is wrong.
yes I know they are made on purpose, nevertheless they are wrong. It is
quite
common that developers not knowing a codebase do things incorrectly. I
will
now only elaborate on the two commits I outlined. In fact all commits I
have
seen so far would not pass a review request for KWin and as I mentioned
there
is at least one commit with the potential to prevent KWin from starting at
all.
Let's start with 1f40ada: you modify the inline getter for keepAbove. This
is
not how KWin internally works to have window being as keep above. The
proper
method to go through is Client::setKeepAbove() which would also tell other
interested parties that the window is in fact kept above. This method is
quite
important to use as it also takes care of putting the window into the
correct
layer of the stacking order. I think you solved that by hacking the
stacking
order.
The simplest way to achieve what you actually wanted would have been to
make
your "modal system notification" an override redirect window. The second commit I pointed out was 9cc1e2c1: I
think others already
commented
in my blog comments why this one is rather bad from a users point of view
(introducing new config options without removing the obsoleted ones). But
well
the main issue from my point of view is that it modified an enum in a
public
header by not appending to the end, but in the middle. I think you can
imagine
what happens to 3rd party offerings compiled against the previous version.
We are not too concerned about ABI compatibility here, but since the
requested change is so trivial I suppose I can push it through.
Anything else?
Sure, as I mentioned all commits would not pass review, hardly
anyone is only
close to correct. You are not a window manager developer - that's nothing bad.
Hardly anyone is actually developing window managers. It took me more than a
year to understand KWin. It is no surprise that you - given the amount of code
you have to handle, are not able to properly develop it. Think about that we
develop the window manager in a peer reviewed style and hardly any commit just
enters the tree. Also we have hundreds of developers testing our code right
from the day it enters master, have thousands of beta testers and much much
more infrastructure to handle the development of critical code pathes.
That's why it would be important to switch to KWin. It would seriously improve
the quality for your users and as it has been stated here more than once,
that's what you care about :-)
Cheers
Martin