More granular script thread memory reporting - #46931

…ervo#47531)

Building on top of servo#46931, this adds a new `out-of-tree` entry to the
DOM memory report which automatically groups all memory for each
previously-unmeasured WebIDL interface that is encountered when
measuring the objects in the GC graph. Any object measured when
traversing the tree is excluded from this, so we won't count any objects
twice even if we add new grouping categories.

Testing: No automated tests for about:memory at this time.

---------

Signed-off-by: Josh Matthews <josh@joshmatthews.net>

So far, all Rust memory for DOM objects ended up in a single category
(MallocHeap).

To get a more fine-grained view into memory usage, we are now counting
DOM object sizes:

```rust
struct DocumentSizes {
    element_nodes_size: usize,
    text_nodes_size: usize,
    attribute_nodes_size: usize,
    other_nodes_size: usize,
}
```

For flexibility reasons, we do the counting within Servo instead of
within SpiderMonkey.

When collecting memory reports from the script thread, we traverse each
document and accumulate node sizes into `DocumentSizes`. We remember,
which nodes we already visited (`ALREADY_COMPUTED_OBJECTS`), and ignore
these in `get_size()` to avoid counting them twice.

<img width="1354" height="1568" alt="image"
src="https://github.com/user-attachments/assets/fc217398-fd99-4665-b9bf-9ea0f6ed140b"
/>

---

Testing: manual
Fixes: servo#44995

Signed-off-by: pylbrecht <pylbrecht@mailbox.org>

…ervo#47531)

Building on top of servo#46931, this adds a new `out-of-tree` entry to the
DOM memory report which automatically groups all memory for each
previously-unmeasured WebIDL interface that is encountered when
measuring the objects in the GC graph. Any object measured when
traversing the tree is excluded from this, so we won't count any objects
twice even if we add new grouping categories.

Testing: No automated tests for about:memory at this time.

---------

Signed-off-by: Josh Matthews <josh@joshmatthews.net>

So far, all Rust memory for DOM objects ended up in a single category
(MallocHeap).

To get a more fine-grained view into memory usage, we are now counting
DOM object sizes:

```rust
struct DocumentSizes {
    element_nodes_size: usize,
    text_nodes_size: usize,
    attribute_nodes_size: usize,
    other_nodes_size: usize,
}
```

For flexibility reasons, we do the counting within Servo instead of
within SpiderMonkey.

When collecting memory reports from the script thread, we traverse each
document and accumulate node sizes into `DocumentSizes`. We remember,
which nodes we already visited (`ALREADY_COMPUTED_OBJECTS`), and ignore
these in `get_size()` to avoid counting them twice.

<img width="1354" height="1568" alt="image"
src="https://github.com/user-attachments/assets/fc217398-fd99-4665-b9bf-9ea0f6ed140b"
/>

---

Testing: manual
Fixes: servo#44995

Signed-off-by: pylbrecht <pylbrecht@mailbox.org>

…ervo#47531)

Building on top of servo#46931, this adds a new `out-of-tree` entry to the
DOM memory report which automatically groups all memory for each
previously-unmeasured WebIDL interface that is encountered when
measuring the objects in the GC graph. Any object measured when
traversing the tree is excluded from this, so we won't count any objects
twice even if we add new grouping categories.

Testing: No automated tests for about:memory at this time.

---------

Signed-off-by: Josh Matthews <josh@joshmatthews.net>