This commit also changes the behaviour of willPatch and patched: they no
longer transfer data from one to the other, because it can be done
cleanly by hooks and a closure.
closes#307
If we want to generate dynamically svg components, it is useful to be
able to define components with a <g> tag as root, and then getting the
correct namespace.
closes#302
With this commit, we can now do this:
const SUBTEMPLATE = xml`<span><t t-esc="state.n"/></span>`
class Parent extends Widget {
static template = xml`<div><t t-call="${SUBTEMPLATE}"/></div>`
state = {n: 42};
}
Big change! This commit introduces an xml function tag to easily define
inline templates.
This is a pretty big change toward single file owl components
Part of #284
Before this commit, QWeb crashed when it had to deal with a t-call with
multiple sub nodes on the same line, and one text node:
<t t-call="SomeTemplate">
<span>hey</span> <span>hey</span>
</t>
This was because we need to compile the content to be able to extract
all possible t-set t-value statements. But doing so means that we had a
multiple root templates, which caused a crash in this case.
Solving this issue is simple: the sub template is compiled with the flag
allowMultipleRoots set to true.
Of course, the templateId computation is kind of tricky. Here, an issue
occurred because the templateId did not take the componentId into
account when we were in a loop, but with no keyed parent.
This meant that multiple sub components shared the same templateId,
confusing the vdom algorithm.
This is a difficult part of owl: we need to be able to reconcile the
vdom generated (this is done with our virtual dom algorithm), but also
to be able to reconcile the proper components.
The issue here is that when we are in a list, with a t-key attribute on all
list nodes, and those list nodes contains sub component, then the
component system used the index of the list, and did not take into
account the key from the parent. The fix is to keep track of the
current key in the context, and uses that as a part of the templateId.
When components were initially rendered more than once, in consecutive
microtask ticks, the order of two actions was reversed: the assignation
of the vnode to the __owl__.vnode property, and the assignation of
__owl__.vnode to the final rendered vdom of the component.
As a result, the widget was rendered as null (so, not present).
The reason for this issue was that the renderPromise key was reassigned
at some point.
Aaron Bohy