I think the current code is currently okay because:
- initialStyle() returns a const reference
- CSSToLengthConversionData takes a raw ptr to a ComputedStyle and doesn't store it in a RefPtr.

But from a system design perspective these classes can only be used from main, for example we might lazy allocate the nested fields inside it, or something inside CSSToLengthConversionData could call a function that refs the style.

Removing Blink>DOM because DOMMatrixReadOnly is not in the scope of Blink>DOM.  We should have something like Blink>Geometry.

The geometry types are not a part of the bindings. What team owns DOMClientRect? What team did the code reviews for the new geometry types code?

I think this feature needs an owner before shipping and that all the geometry types should belong to one of DOM, CSS, Paint or Layout teams. Adding a more specific bug component like Blink>Geometry doesn't address the team ownership problem. 

I think it should be the CSS team that owns at least DOMRect, because the main API using it is getBoundingClientRect() defined in https://drafts.csswg.org/cssom-view/#extension-to-the-element-interface

It makes sense for DOMMatrix too, given the entanglement with the parser.

Just my 2c.

Not a P1 because we do not need it for a current milestone, and as I understand it we have not shipped the geometry types yet. Correct me (and adjust the priority) if I'm wrong.

When a decision is made to ship the priority needs to be bumped, a release version needs to be attached, and probably also needs a ReleaseBlock if it really is such a serious concern.

Okey.
So 3 things about this.

1. According to https://drafts.fxtf.org/geometry/#dom-dommatrixreadonly-dommatrixreadonly, we should never call SetMatrixValueFromString on the worker. I'm preparing a CL to fix this today. This bug will be a non-issue.

2. Seeing this code reminded me that FilterOperationResolver does have a use case for resolving Canvas2D filters on Workers that have a similar code: https://cs.chromium.org/chromium/src/third_party/WebKit/Source/core/css/resolver/FilterOperationResolver.cpp?type=cs&q=CreateOffscreenFilterOperations&sq=package:chromium&l=208

(ignore the viewport_size part, we are still debating what's the proper viewport spec-wise). This code has the exact same problem pointed out by this bug.

3. As Elliott pointed out, this should be currently safe unless CSSToLengthConversionData ever decides to start changing something global. At this point, I'd say that we probably need to fork CSSToLengthConversionData (and CHECK the proper threads) and keep a thread-safe version of it. Design-wise, I don't know how to make this explicit, except for having what we currently do: have tests on separate threads and hope that when the function becomes unsafe there will be a new CHECK that will fail.

Ideas?

The following revision refers to this bug:
  https://chromium.googlesource.com/chromium/src.git/+/d72fe5c3a271ab3e63f2ca5c247b3e4e686e275b

commit d72fe5c3a271ab3e63f2ca5c247b3e4e686e275b
Author: fserb <fserb@chromium.org>
Date: Fri May 12 02:04:03 2017

Stop accepting string constructor of DOMMatrix on workers

This conforms to the spec:
https://drafts.fxtf.org/geometry/#dom-dommatrixreadonly-dommatrixreadonly

And ensures that we don't call SetMatrixValueFromString inside a worker

BUG= 703908 

Review-Url: https://codereview.chromium.org/2868953002
Cr-Commit-Position: refs/heads/master@{#471182}

[modify] https://crrev.com/d72fe5c3a271ab3e63f2ca5c247b3e4e686e275b/third_party/WebKit/Source/core/geometry/DOMMatrix.cpp
[modify] https://crrev.com/d72fe5c3a271ab3e63f2ca5c247b3e4e686e275b/third_party/WebKit/Source/core/geometry/DOMMatrix.h
[modify] https://crrev.com/d72fe5c3a271ab3e63f2ca5c247b3e4e686e275b/third_party/WebKit/Source/core/geometry/DOMMatrix.idl
[modify] https://crrev.com/d72fe5c3a271ab3e63f2ca5c247b3e4e686e275b/third_party/WebKit/Source/core/geometry/DOMMatrixReadOnly.cpp
[modify] https://crrev.com/d72fe5c3a271ab3e63f2ca5c247b3e4e686e275b/third_party/WebKit/Source/core/geometry/DOMMatrixReadOnly.h
[modify] https://crrev.com/d72fe5c3a271ab3e63f2ca5c247b3e4e686e275b/third_party/WebKit/Source/core/geometry/DOMMatrixReadOnly.idl

The geometry specific issue is closed.
I've created  http://crbug.com/721768  to follow up on this outside Geometry.