This fixes a race condition flicker when unfloating a view which uses
client side decorations.
When the view is floated it has using_csd = true, so the decorations are
not drawn. When unfloating it it changes to false, but this change
wasn't part of transactions so it could potentially render the
decorations around the view while it's waiting for the transaction to
apply.
Commit 4b8e3a885b makes it so only one
transaction is committed (ie. configures sent) at a time. This commit
removes the now-unnecessary code which was used to support concurrent
committed transactions.
* Instead of containers storing a list of instructions which they've
been sent, it now stores a single instruction.
* Containers now have an ntxnrefs property. Previously we knew how many
references there were by the length of the instruction list.
* Instructions no longer need a ready property. It was used to avoid
marking an instruction ready twice when they were in a list, but this is
now avoided because there is only one instruction and we nullify the
container->instruction pointer when it's ready.
* When a transaction applies, we no longer need to consider releasing
and resaving the surface, as we know there are no other committed
transactions.
* transaction_notify_view_ready has been renamed to
view_notify_view_ready_by_serial to make it consistent with
transaction_notify_view_ready_by_size.
* Out-of-memory checks have been added when creating transactions and
instructions.
This fixes an issue where views might commit to a transaction ahead of
the first one, and applying the first transaction causes us to save a
buffer of the wrong size.
This fixes two issues which were both introduced in #2396.
First issue:
The PR changes the location of the buffer save to transaction_apply, but
puts it inside the should_configure block. For unmapping (destroying)
views, should_configure returns false so it wasn't saving the buffer. If
a frame was rendered between the unmap and the transaction applying then
it would result in a crash.
Second issue:
If a destroying view is involved in two transactions, we must not
release the buffer between the transactions because there is no live
buffer to grab any more.
Fixes#2364.
Suppose a view is 600px wide, and we tell it to resize to 601px during a
resize operation. We create a transaction, save the 600px buffer and
send the configure. This buffer is saved into the associated
instruction, and is rendered while we wait for the view to commit a
601px buffer.
Before the view commits the 601px buffer, suppose we tell it to resize
to 602px. The new transaction will also save the buffer, but it's still
the 600px buffer because we haven't received a new one yet.
Then suppose the view commits its original 601px buffer. This completes
the first transaction, so we apply the 601px width to the container.
There's still the second (now only) transaction remaining, so we render
the saved buffer from that. But this is still the 600px buffer, and we
believe it's 601px. Whoops.
The problem here is we can't stack buffers like this. So this commit
removes the saved buffer from the instructions, places it in the view
instead, and re-saves the latest buffer every time the view completes a
transaction and still has further pending transactions.
As saved buffers are now specific to views rather than instructions, the
functions for saving and removing the saved buffer have been moved to
view.c.
The calls to save and restore the buffer have been relocated to more
appropriate functions too, favouring transaction_commit and
transaction_apply rather than transaction_add_container and
transaction_destroy.
The solution used in 073ac425d5 doesn't
work in all cases because the freed instruction might be ahead in the
list, not necessarily behind.
The new solution delays running the queue until after the loop has
finished iterating, thus avoiding the problem completely.
In set_instructions_ready, calling set_instruction_ready may cause any
number of transactions to get applied, which removes them from the list
being iterated. The iteration variables need to be adjusted
accordingly.
When interactively resizing some views (eg. Nautilus), new transactions
are added to the queue faster than the client can process them.
Previously, we would wait for the entire queue to be ready before
applying any of them, but in this case the transactions would time out,
giving the client choppy performance.
This changes the queue handling so it applies the transactions up to the
first waiting transaction, without waiting for the entire queue to be
ready.
Rather than maintain copies of the entire focus stack, this PR
transactionises the focus by introducing two new properties to the
container state and using those when rendering.
* `bool focused` means this container has actual focus. Only one
container should have this equalling true in its current state.
* `struct sway_container *focus_inactive_child` points to the immediate
child that was most recently focused (eg. for tabbed and stacked
containers).
We currently have several ways of setting debug flags, including command
line arguments, environment variables, and compile-time macros. This
replaces the lot with command line flags.
This PR changes the way we handle transactions to a more simple method.
The new method is to mark containers as dirty from low level code
(eg. arranging, or container_destroy, and eventually seat_set_focus),
then call transaction_commit_dirty which picks up those containers and
runs them through a transaction. The old methods of using transactions
(arrange_and_commit, or creating one manually) are now no longer
possible.
The highest-level code (execute_command and view implementation
handlers) will call transaction_commit_dirty, so most other code just
needs to set containers as dirty. This is done by arranging, but can
also be done by calling container_set_dirty.
If we set an instruction as ready twice, it decreases the transaction's
num_waiting a second time and applies the transaction earlier than it
should. This no doubt has undesired effects, probably resulting in a use
after free.
Hopefully fixes the first part of #2207.
A flash of background was happening for two reasons:
1) We were using the xsurface's dimensions to check if the surface is
ready, but these are pending dimensions.
2) In my particular setup, the default geometry of the xsurface does not
intersect any output, which prevented it from receiving a frame done
event. This made the transaction time out and the client would only
redraw once it's been rendered.
* The arrange_foo functions are now replaced with arrange_and_commit, or
with manually created transactions and arrange_windows x2.
* The arrange functions are now only called from the highest level
functions rather than from both high level and low level functions.
* Due to the previous point, view_set_fullscreen_raw and
view_set_fullscreen are both merged into one function again.
* Floating and fullscreen are now working with transactions.
