Separate interning text run

[djg]

First part of separating Primitives into separate interned data.

---

This change is 

[gw3583]

Looks like there is a reftest failure on CI:

REFTEST TEST-UNEXPECTED-FAIL | reftests\text\shadow.yaml == reftests\text\shadow-ref.yaml | image comparison, max difference: 131, number of differing pixels: 6425

Any chance you could do a talos run before review, just to give an idea what effect it has on dl_mutate (even just to make sure it doesn't regress, for now, since we probably won't get big wins from this initial work)?

[djg]

Here is the try run with talos results. I see some small improvement and regression which is curious. https://treeherder.mozilla.org/#/jobs?repo=try&revision=752295ab260070f973c22dcb167cd2bef18e48bd

[gw3583]

Reviewed 6 of 12 files at r1.
Reviewable status: 6 of 12 files reviewed, 2 unresolved discussions (waiting on @djg)

---

webrender/src/clip.rs, line 101 at r1 (raw file):

#[derive(Clone, Copy, Debug, Eq, Hash, PartialEq)]
pub struct ClipUidMarker;
pub type ClipUid = intern::ItemUid<ClipUidMarker>;

This would become simpler if we use a global uid type (mentioned below).

---

webrender/src/intern.rs, line 76 at r1 (raw file):

/// Implement for a marker struct to associate the type of Uid marker to use
/// with `intern::Handle<..>`
pub trait UidMarker: Copy + Clone {

Having the item uids require this trait and phantom marker feels a bit overengineered to me. How about we just share one uid concrete type among all interners, and don't make them generic. That might aid in debugging too, since a uid is globally unique and would never overlap with a uid from a different interner type.

[gw3583]

(partial review only so far, will work through this today).

[bors-servo]

☔ The latest upstream changes (presumably #3368) made this pull request unmergeable. Please resolve the merge conflicts.

[djg]

(partial review only so far, will work through this today).

Thanks @gw3583

[djg]

Reviewable status: 2 of 12 files reviewed, 2 unresolved discussions (waiting on @gw3583)

---

webrender/src/clip.rs, line 101 at r1 (raw file):

Previously, gw3583 (Glenn Watson) wrote…

This would become simpler if we use a global uid type (mentioned below).

Done.

---

webrender/src/intern.rs, line 76 at r1 (raw file):

Previously, gw3583 (Glenn Watson) wrote…

Having the item uids require this trait and phantom marker feels a bit overengineered to me. How about we just share one uid concrete type among all interners, and don't make them generic. That might aid in debugging too, since a uid is globally unique and would never overlap with a uid from a different interner type.

Done.

[gw3583]

Reviewed 4 of 8 files at r2.
Reviewable status: 6 of 12 files reviewed, 2 unresolved discussions (waiting on @gw3583)

[gw3583]

Reviewed 1 of 12 files at r1, 1 of 8 files at r2.
Reviewable status: 8 of 12 files reviewed, 8 unresolved discussions (waiting on @gw3583 and @djg)

---

webrender/src/display_list_flattener.rs, line 1233 at r2 (raw file):

            );

            //cur_instance.kind = PrimitiveInstanceKind::Picture { pic_index: current_pic_index };

Why is this change needed?

---

webrender/src/display_list_flattener.rs, line 1264 at r2 (raw file):

            current_pic_index = filter_pic_index;
            //cur_instance.kind = PrimitiveInstanceKind::Picture { pic_index: current_pic_index };

same

---

webrender/src/display_list_flattener.rs, line 1302 at r2 (raw file):

            current_pic_index = blend_pic_index;
            // cur_instance.kind = PrimitiveInstanceKind::Picture { pic_index: blend_pic_index };
            match cur_instance.kind {

same

---

webrender/src/picture.rs, line 573 at r2 (raw file):

        );

        let prim_rect = match prim_instance.kind {

This is not quite right for Pictures (their rect is built dynamically), but this wasn't handled correctly previously anyway - I've got a fix for it in my local branch.

---

webrender/src/picture.rs, line 1253 at r2 (raw file):

            };

            let prim_data = match prim_instance.kind {

Maybe make this a helper method on FrameResources? Seems like it'll be used in a few places.

---

webrender/src/scene_builder.rs, line 173 at r2 (raw file):

}

// Access to `DocumentResources` interners by `Internable`

What is this trait needed for?

[gw3583]

Reviewed 1 of 12 files at r1, 3 of 8 files at r2.
Reviewable status: all files reviewed, 11 unresolved discussions (waiting on @gw3583 and @djg)

---

webrender/src/display_list_flattener.rs, line 2236 at r2 (raw file):

}

pub trait AsInstanceKind<H> {

Is there a reason to have these as separate traits rather than a "Primitive" trait or similar?

---

webrender/src/prim_store/text_run.rs, line 29 at r2 (raw file):

#[derive(Debug, Clone, Eq, PartialEq, Hash)]
pub struct TextRunKey {
    pub is_backface_visible: bool,

It's probably worth having a set of "common" structs that are embedded within each key / template etc for the data that is shared between all the different types?

---

webrender/src/prim_store/text_run.rs, line 80 at r2 (raw file):

#[cfg_attr(feature = "replay", derive(Deserialize))]
pub struct TextRunTemplate {
    pub is_backface_visible: bool,

Same here for common data.

[djg]

Reviewable status: all files reviewed, 11 unresolved discussions (waiting on @gw3583 and @djg)

---

webrender/src/display_list_flattener.rs, line 1233 at r2 (raw file):

Previously, gw3583 (Glenn Watson) wrote…

Why is this change needed?

Because PrimitiveInstanceKind:Picture now contains a data_handle. data_handle is Copy so I might try adding it the Picture { ... } here, instead of updating pic_index in-place.

[djg]

Reviewable status: all files reviewed, 11 unresolved discussions (waiting on @gw3583)

---

webrender/src/display_list_flattener.rs, line 2236 at r2 (raw file):

Previously, gw3583 (Glenn Watson) wrote…

Is there a reason to have these as separate traits rather than a "Primitive" trait or similar?

Because there implemented on different types. I can move BuildKey into the the Internable trait as is it related to generating the key for interning. AsInstanceKind is used for getting a PrimitiveInstanceKind and is implemented on the key that's the source, eg TextRunKey. CreateShadow and IsVisible are properties of the primitive and can be combined into a Primitive trait.

---

webrender/src/picture.rs, line 573 at r2 (raw file):

Previously, gw3583 (Glenn Watson) wrote…

This is not quite right for Pictures (their rect is built dynamically), but this wasn't handled correctly previously anyway - I've got a fix for it in my local branch.

Do you need to post me the fix to incorporate?

---

webrender/src/picture.rs, line 1253 at r2 (raw file):

Previously, gw3583 (Glenn Watson) wrote…

Maybe make this a helper method on FrameResources? Seems like it'll be used in a few places.

Done.

---

webrender/src/scene_builder.rs, line 173 at r2 (raw file):

Previously, gw3583 (Glenn Watson) wrote…

What is this trait needed for?

It's used by the generic version of DisplayListFlattener::create_primitive to obtain the correct interner for primitive.

        let interner = self.resources.interner_mut(); // Uses type deduction to work out the which store +
        let prim_data_handle =                        //                                                 |
            interner                                  //                                                 |
            .intern(&prim_key, || { ... });           // << From the types required here ----------------+

---

webrender/src/prim_store/text_run.rs, line 29 at r2 (raw file):

Previously, gw3583 (Glenn Watson) wrote…

It's probably worth having a set of "common" structs that are embedded within each key / template etc for the data that is shared between all the different types?

OK, I had that in a version but on IRC you advised against it. I think having common structs can help clean up some pattern matching mess.

---

webrender/src/prim_store/text_run.rs, line 80 at r2 (raw file):

Previously, gw3583 (Glenn Watson) wrote…

Same here for common data.

Done.

[djg]

Reviewable status: 3 of 12 files reviewed, 11 unresolved discussions (waiting on @gw3583)

---

webrender/src/display_list_flattener.rs, line 2236 at r2 (raw file):

Previously, djg (Dan Glastonbury) wrote…

Because there implemented on different types. I can move BuildKey into the the Internable trait as is it related to generating the key for interning. AsInstanceKind is used for getting a PrimitiveInstanceKind and is implemented on the key that's the source, eg TextRunKey. CreateShadow and IsVisible are properties of the primitive and can be combined into a Primitive trait.

One thing I found that is having CreateShadow being a separate trait means that we can use the compiled to help us. ie.

impl CreateShadow for LinearGradient {
    fn create_shadow(&self, shadow: &Shadow) -> Self {
        PrimitiveKeyKind::LinearGradient { .. } |
        panic!("bug: this prim is not supported in shadow contexts");
    }
}

We don't need the panic! implementation and can have the compiler error when trying to create a shadow for LinearGradient because the CreateShadow trait doesn't exists for LinearGradient.

[bors-servo]

☔ The latest upstream changes (presumably #3381) made this pull request unmergeable. Please resolve the merge conflicts.

[djg]

@gw3583 Ping

[gw3583]

Sorry, all-hands jetlag and that - I was going to take a more detailed look in the morning, but I think it probably looks good in general now, or the remaining bits could be done as follow ups. Did we work out what effect this has on talos?

[gw3583]

Probably worth getting @kvark to take a quick look too, if he has time.

[djg]

I'm currently working through removing the conflict with servo/master. Update shortly.

@gw3583 It had a slight improvement on talos, but I wasn't expecting much of a change yet.

[gw3583]

@djg Yea, I wouldn't expect a large improvement until we separate solid rects. As long as it doesn't regress, we're good for now 👍

[bors-servo]

☔ The latest upstream changes (presumably #3384) made this pull request unmergeable. Please resolve the merge conflicts.

[gw3583]

Sorry, this got conflicts again - I think it should be good to go once those conflicts are resolved?

[djg]

Gecko try run: https://treeherder.mozilla.org/#/jobs?repo=try&revision=b66118f88f9306e0d2fd6a94df9da24fb71a7135

[djg]

@gw3583 Apparently there is one comment to resolve?

[gw3583]

I think it's fine to get this merged now, to avoid any more rebase issues. We should be able to fix up any remaining changes as follow ups.

[gw3583]

@bors-servo r+

[bors-servo]

📌 Commit e3ccda7 has been approved by gw3583

[bors-servo]

⌛ Testing commit e3ccda7 with merge 6db2d6c...

[bors-servo]

☀️ Test successful - status-appveyor, status-taskcluster
Approved by: gw3583
Pushing 6db2d6c to master...