Stack.Layout and Flex.Layout caused a lot of heap allocations / escapes. The
reason was that scratch space for dims and call and was inside Stack/FlexChild.
child.call.Add(gtx.Ops) confused the go escape analysis and caused the entired
children slice to escape to the heap, including all widgets in it. This caused a
lot of heap allocations. Now the scratch space is separate from children, and
for cases len(children) <= 32, we will allocate the scratch space on the stack.
For cases len(children) > 32, only the scratch space gets allocated from the
heap, during append.
Signed-off-by: vsariola <5684185+vsariola@users.noreply.github.com>
It's relatively common to create a widget and then add a background to
it. Using layout.Stack causes bunch of heap allocs, which we would like
to avoid whenever we can.
This adds layout.Background which is roughly the same as:
layout.Stack{Alignment: layout.C}.Layout(gtx,
layout.Expanded(background),
layout.Stacked(widget)
)
goos: windows
goarch: amd64
pkg: gioui.org/layout
cpu: AMD Ryzen Threadripper 2950X 16-Core Processor
│ Stack │ Background │
│ sec/op │ sec/op vs base │
*-32 203.80n ± 1% 83.36n ± 3% -59.09% (p=0.000 n=10)
│ Stack │ Background │
│ B/op │ B/op vs base │
*-32 48.00 ± 0% 0.00 ± 0% -100.00% (p=0.000 n=10)
│ Stack │ Background │
│ allocs/op │ allocs/op vs base │
*-32 2.000 ± 0% 0.000 ± 0% -100.00% (p=0.000 n=10)
Signed-off-by: Egon Elbre <egonelbre@gmail.com>
op.Offset is a convenience function most often used by layouts. Layouts
usually operate in integer coordinates, and the float32 version of op.Offset
needlessly force conversions from int to float32. This change makes op.Offset
take integer coordinates, to better match its intended use.
Signed-off-by: Elias Naur <mail@eliasnaur.com>
The op.Save and Load methods exist to support the need for
transformation, clip, pointer area state to behave as stacks. For
example, layout needs to apply an offset to its children but not
subsequent operations.
Before this change, op.Save and Load were used to save and restore the
state:
ops := new(op.Ops)
// Save state.
state := op.Save(ops)
// Apply offset.
op.Offset(...).Add(ops)
// Draw with offset applied.
draw(ops)
// Restore state.
state.Load()
A drawback with the op.Save mechanism is that there is no direct
connection between the state change and the saving and loading of state.
This causes confusion as to when a Save/Load is needed and who is
responsible for performing them, which leads to subtle bugs and over-use
of Save/Loads.
This change gets rid of the general state stack and replaces it with
per-state stacks. There is now a stack for transformation, clip, pointer
areas, and they can only be restored by the code pushing state to them.
The example above now becomes:
ops := new(op.Ops)
// Push offset to the transformation stack.
stack := op.Offset(...).Push(ops)
// Draw with offset applied.
draw(ops)
// Restore state.
stack.Pop()
For convenience, transformation also be Add'ed if the stack operation is
not required.
Simple state such as the current material no longer has a way to be
restored; it is assumed the client of a PaintOp adds their desired
material operation before it.
API change: replace op.Save/Load with explicit Push/Pop scopes for
op.TransformOps, pointer.AreaOps, clip.Ops.
To ease porting, this change retains a version of op.Save/Load that
saves and restores the transformation and clip stacks. It also retains
an Add method for clip.Op.
Signed-off-by: Elias Naur <mail@eliasnaur.com>
The semantics were relaxed in a previous commit; this change renames
to operations accordingly.
API change. Use gofmt to adjust your code accordingly:
gofmt -r 'op.Push(a).Pop() -> op.Save(a).Load()'
gofmt -r 'op.Push(a) -> op.Save(a)'
gofmt -r 'v.Pop() -> v.Load()'
gofmt -r 'op.StackOp -> op.StateOp'
Signed-off-by: Elias Naur <mail@eliasnaur.com>
Use op.Offset instead, or create and manipulate a f32.Affine2D.
API change. Update your code with a gofmt rule:
gofmt -r 'op.TransformOp{}.Offset -> op.Offset'
Signed-off-by: Elias Naur <mail@eliasnaur.com>
Converting
macro := op.Record(ops)
...
macro.Stop()
macro.Add()
to
macro := op.Record(ops)
...
call := macro.Stop()
call.Add(ops)
Which is more general (call.Add can take a different ops than the op.Record
that started it), and enforced the order between Stop and the subsequent Add.
Signed-off-by: Elias Naur <mail@eliasnaur.com>
The funcs replace stack.Push and macro.Record, which become private.
This makes stack and macro faster to write, in particular for stacks
where you can just write the following line to save and restore the
state :
defer op.Push(ops).Pop()
This usage requires Push to return a pointer (since Pop has a pointer
receiver), or else the code doesn't compile.
For consistancy, I tried to do the same for op.Record, but this implied
to turn all the MacroOp fields into pointers, and this caused some
panics. As a result, op.Record doesn't return a pointer.
An other side effect pointed by Larry Clapp: StackOp and MacroOp are not
re-usable any more, you have to allocate a new one for each usage, using
the described funcs above.
Signed-off-by: Thomas Bruyelle <thomas.bruyelle@gmail.com>
Change the definition of Widget from the implicit
type Widget func()
to the explicit functional
type Widget func(gtx layout.Context) layout.Dimensions
The advantages are numerous:
- Clearer connection between the incoming context and the output dimensions.
- Returning the Dimensions are impossible to omit.
- Contexts passed by value, so its fields can be exported
and freely mutated by the program.
The only disadvantage is the longer function literals and the many "returns".
What tipped the scales in favour of the explicit Widget variant is that type
aliases can dramatically shorten the literals:
type (
C = layout.Context
D = layout.Dimensions
)
widget := func(gtx C) D {
...
}
Note that the aliases are not part of the Gio API and it is up to each user
whether they want to use them.
Finally the Go proposal for lightweight function literals,
https://github.com/golang/go/issues/21498, may remove the disadvantage
completely in future.
Context becomes a plain struct with only public fields, and its Reset is
replaced by a NewContext convenience constructor.
Signed-off-by: Elias Naur <mail@eliasnaur.com>
Instead of
type Contraints struct {
Width, Height Constraint
}
use
type Constraints struct {
Min, Max image.Point
}
which leads to simpler use. For example, the Min method is trivally replaced by
the field, and the RigidConstraints constructor is no longer a net win.
API Change. Rewrites:
gofmt -r 'gtx.Constraints.Min() -> gtx.Constraints.Min'
gofmt -r 'gtx.Constraints.Width.Min -> gtx.Constraints.Min.X'
gofmt -r 'gtx.Constraints.Height.Min -> gtx.Constraints.Min.Y'
gofmt -r 'gtx.Constraints.Height.Max -> gtx.Constraints.Max.Y'
gofmt -r 'gtx.Constraints.Width.Max -> gtx.Constraints.Max.X'
Signed-off-by: Elias Naur <mail@eliasnaur.com>
Similar to what a previous commit did for Flex, this change simplifies
Stack to just one Layout call:
layout.Stack{}.Layout(gtx,
layout.Stacked(func() {...}),
layout.Expanded(func() {...}),
)
Signed-off-by: Elias Naur <mail@eliasnaur.com>
The ability to invoke other operation lists belongs in the new CallOp.
While we're here, make MacroOp.Add use a pointer receiver to match the
other methods.
Signed-off-by: Elias Naur <mail@eliasnaur.com>
Overlapping layouts such as
outer := layout.Flex{}
inner := layout.Stack{}
child := inner.Rigid(gtx, ...)
outerChild := outer.Rigid(gtx, func() {
inner.Layout(gtx, child)
})
outer.Layout(gtx, outerChild)
runs but result in a wrong layout.
This change use empty StackOps to ensure that the Stack and Flex
child layout methods are called in the same scope as their Layout
methods:
outer := layout.Flex{}
inner := layout.Stack{}
outerChild := outer.Rigid(gtx, func() {
child := inner.Rigid(gtx, ...)
inner.Layout(gtx, child)
})
outer.Layout(gtx, outerChild)
Signed-off-by: Elias Naur <mail@eliasnaur.com>
With Context containing all the necessary information, separate
Init methods no longer makes much sense. Delete them and thereby
remove a source of runtime panics.
Signed-off-by: Elias Naur <mail@eliasnaur.com>
