Add callback for document element creation

[magreenblatt]

Original report by Anonymous.

---

Original issue 1454 created by finn.hughes1 on 2014-11-26T10:58:02.000Z:

What steps will reproduce the problem?

1. Upgrade to cef 1750 thus losing support for user style sheets
2. Try to use OnLoadStart to inject css to colour scrollbar red
3. Sometimes scrollbars are red sometimes not.

What is the expected output? What do you see instead?
On simple but long pages e.g.
http://munsterdirectory.com/aboutus.htm
Sometimes the red styling isn't applied.

What version of the product are you using? On what operating system?
1750 on fedora/ubuntu 32bit.

Please provide any additional information below.
I've attached a patch to cefclient to demonstrate the problem (scrollbar_style_test.diff)

I imagine this is a fundemental limitation of OnLoadStart being asynchronous and to reliably inject css to style the main window scrollbar we'll have to do it in the render process.

Reading through the chrome issues from when they dropped support for the user style sheet it seems chrome extensions are the replacement. To inject css in a chrome extension you use 'content_scripts'. Looking at the source for that I tracked the injection to:
chrome/renderer/extensions/user_script_scheduler.cc
Triggered by UserScriptScheduler::DidCreateDocumentElement it uses the insertUserStyleSheet call that used to be used for user style sheet injection.

When removing user style sheets they created a insertStyleSheet function and made it so the second parameter setting the type of style sheet would be ignored so all sheets are author sheets. I had a look at instertStyleSheet in trunk and found it unchanged so this interface has remained stable from 33-40+.

I created a very simple patch to emulate the css injection of the extension, but only took it far enough for my needs. If there are specific steps required to make this something cef can accept upstream I'll see if I can find time.

I patched:
content/renderer/render_view_impl.cc
adding one simple call to RenderViewImpl::didCreateDocumentElement
'frame->document().insertStyleSheet(WebString::fromUTF8(css_));'
where css_ was initialized by reading a css file.

[magreenblatt]

Original comment by Anonymous.

---

Comment 1. originally posted by czarek.tomczak on 2014-11-26T11:47:13.000Z:

I'm +1 to restore user stylesheet option to CefSettings.

Regarding "limitation of OnLoadStart being asynchronous" issue. CefLoadHandler with OnLoadStart callback is exposed to both the browser process and the renderer process. It's available in the renderer process handler since branch 1650, see issue #1077.

[magreenblatt]

Comment 2. originally posted by magreenblatt on 2014-11-26T19:47:18.000Z:

@ finn: Please try OnLoadStart in the render process and see if that satisfies your use case.

[magreenblatt]

Original comment by Anonymous.

---

Comment 3. originally posted by finn.hughes1 on 2014-11-30T21:53:44.000Z:

Thanks for the feedback, I wasn't aware the handler was available in the render process. I didn't take the time to understand the delegates in cefclient instead I implemented this in cefpython since that's what I use anyway.

The javascript I injected started:
console.log('boo');
console.log(document.head.innerHTML);

And the log showed "boo" then:
CONSOLE(0)] "Uncaught TypeError: Cannot read property 'innerHTML' of null"

So it seems that there is no DOM if you call this function in the render process, it's possible I've done something wrong but that seems to be the case. Any more ideas?

Finn

[magreenblatt]

Comment 4. originally posted by magreenblatt on 2014-12-04T09:29:51.000Z:

@ comment 3.: How did you inject the JavaScript? You're correct that there is no DOM when OnLoadStart is called.

[magreenblatt]

Original comment by Anonymous.

---

Comment 5. originally posted by finn.hughes1 on 2014-12-04T11:07:06.000Z:

Hi Marshall,

I'm not aware of any way to inject css before the dom exists?

I used "frame->ExecuteJavaScript" exactly the same as in the previously attached diff. That fails as there's no head to append to.

If there's a way to inject css before the DOM exists then I'd also have the option of using CefRegisterExtension in the render process, not sure if that would have any advantages or disadvantages over OnLoadStart.

[magreenblatt]

Comment 6. originally posted by magreenblatt on 2015-01-08T18:01:00.000Z:

[magreenblatt]

• set title to "Add callback for document element creation"
• set state to "open"

[magreenblatt]

• edited description
• set component to "Framework"
• changed kind from "bug" to "enhancement"

[magreenblatt]

In the browser process we can expose WebContentsObserver::DocumentAvailableInMainFrame.

[magreenblatt]

Original comment by Alex Maitland (Bitbucket: a-maitland).

---

Would it also be possible to add WebContentsObserver:: DOMContentLoaded at the same time? That would be quite useful also.

Happy to create a new issue rather than hijack this one.

[magreenblatt]

Add support for direct DevTools protocol messaging (fixes issue #2961).

This change allows the client to directly send and receive DevTools
protocol messages (send method calls, and receive method results and
events) without requiring a DevTools front-end or remote-debugging
session.

This change includes additional supporting changes:

• Add a new CefRequestHandler::OnDocumentAvailableInMainFrame
  callback (see issue Add callback for document element creation #1454).
• Add a CefParseJSON variant that accepts a UTF8-encoded buffer.
• Add a --devtools-protocol-log-file=<path> command-line flag for
  logging protocol messages sent to/from the DevTools front-end
  while it is displayed. This is useful for understanding existing
  DevTools protocol usage.
• Add a new "libcef_static_unittests" executable target to support
  light-weight unit tests of libcef_static internals (e.g. without
  requiring exposure via the CEF API). Files to be unittested are
  placed in the new "libcef_static_unittested" source_set which is
  then included by both the existing libcef_static library and the
  new unittests executable target.
• Linux: Remove use_bundled_fontconfig=false, which is no longer
  required and causes unittest build errors (see issue Linux Ubuntu 16.04: Hang during startup in FT_Load_Glyph #2424).

This change also adds a cefclient demo for configuring offline mode
using the DevTools protocol (fixes issue #245). This is controlled
by the "Offline mode" context menu option and the --offline
command-line switch which will launch cefclient in offline mode. When
cefclient is offline all network requests will fail with
ERR_INTERNET_DISCONNECTED and navigator.onLine will return false when
called from JavaScript in any frame. This mode is per-browser so
newly created browser windows will have the default mode. Note that
configuring offline mode in this way will not update the Network tab
UI ("Throtting" option) in a displayed DevTools front-end instance.

→ <<cset 39aed35644dc (bb)>>

[magreenblatt]

@{557058:5abcbcf9-b372-4ecb-830e-ddb451b76481} Can you explain the use case for DOMContentLoaded? How does it differ from the timing of OnLoadEnd (which is triggered from RenderFrameObserver::DidFinishLoad)? Thanks.

[magreenblatt]

Add support for direct DevTools protocol messaging (fixes issue #2961).

This change allows the client to directly send and receive DevTools
protocol messages (send method calls, and receive method results and
events) without requiring a DevTools front-end or remote-debugging
session.

This change includes additional supporting changes:

• Add a new CefRequestHandler::OnDocumentAvailableInMainFrame
  callback (see issue Add callback for document element creation #1454).
• Add a CefParseJSON variant that accepts a UTF8-encoded buffer.
• Add a --devtools-protocol-log-file=<path> command-line flag for
  logging protocol messages sent to/from the DevTools front-end
  while it is displayed. This is useful for understanding existing
  DevTools protocol usage.
• Add a new "libcef_static_unittests" executable target to support
  light-weight unit tests of libcef_static internals (e.g. without
  requiring exposure via the CEF API). Files to be unittested are
  placed in the new "libcef_static_unittested" source_set which is
  then included by both the existing libcef_static library and the
  new unittests executable target.
• Linux: Remove use_bundled_fontconfig=false, which is no longer
  required and causes unittest build errors (see issue Linux Ubuntu 16.04: Hang during startup in FT_Load_Glyph #2424).

This change also adds a cefclient demo for configuring offline mode
using the DevTools protocol (fixes issue #245). This is controlled
by the "Offline mode" context menu option and the --offline
command-line switch which will launch cefclient in offline mode. When
cefclient is offline all network requests will fail with
ERR_INTERNET_DISCONNECTED and navigator.onLine will return false when
called from JavaScript in any frame. This mode is per-browser so
newly created browser windows will have the default mode. Note that
configuring offline mode in this way will not update the Network tab
UI ("Throtting" option) in a displayed DevTools front-end instance.

→ <<cset 0fb8a5f1eada (bb)>>

[magreenblatt]

Add support for direct DevTools protocol messaging (fixes issue #2961).

This change allows the client to directly send and receive DevTools
protocol messages (send method calls, and receive method results and
events) without requiring a DevTools front-end or remote-debugging
session.

This change includes additional supporting changes:

• Add a new CefRequestHandler::OnDocumentAvailableInMainFrame
  callback (see issue Add callback for document element creation #1454).
• Add a CefParseJSON variant that accepts a UTF8-encoded buffer.
• Add a --devtools-protocol-log-file=<path> command-line flag for
  logging protocol messages sent to/from the DevTools front-end
  while it is displayed. This is useful for understanding existing
  DevTools protocol usage.
• Add a new "libcef_static_unittests" executable target to support
  light-weight unit tests of libcef_static internals (e.g. without
  requiring exposure via the CEF API). Files to be unittested are
  placed in the new "libcef_static_unittested" source_set which is
  then included by both the existing libcef_static library and the
  new unittests executable target.
• Linux: Remove use_bundled_fontconfig=false, which is no longer
  required and causes unittest build errors (see issue Linux Ubuntu 16.04: Hang during startup in FT_Load_Glyph #2424).

This change also adds a cefclient demo for configuring offline mode
using the DevTools protocol (fixes issue #245). This is controlled
by the "Offline mode" context menu option and the --offline
command-line switch which will launch cefclient in offline mode. When
cefclient is offline all network requests will fail with
ERR_INTERNET_DISCONNECTED and navigator.onLine will return false when
called from JavaScript in any frame. This mode is per-browser so
newly created browser windows will have the default mode. Note that
configuring offline mode in this way will not update the Network tab
UI ("Throtting" option) in a displayed DevTools front-end instance.

→ <<cset fd6631bd1812 (bb)>>

[magreenblatt]

Original comment by Alex Maitland (Bitbucket: a-maitland).

---

Can you explain the use case for DOMContentLoaded? How does it differ from the timing of OnLoadEnd (which is triggered from RenderFrameObserver::DidFinishLoad)?

Sorry for the slow reply. I understand that technically the difference between OnLoadEnd and DOMContentLoaded in terms of timing is potentially quite small.

The reason for the request to add DOMContentLoaded is more from a support perspective. The DOMContentLoaded is known to anyone familiar with Javascript and a logical place for Javascript execution. It’s largely self documenting.

The cost of implementation looks quite low and the maintenance burden would appear to be small, adding it seems like an easy win in many ways.

I think it falls into the category of nice to have as obviously you can execute javascript in OnLoadStart to hook the event in javascript and use the native javascript event.

It’s not the end of the world if it’s not implemented or you choose to delegate (via PR) to someone else.

Thanks for taking the time to follow up.

‌

‌

[magreenblatt]

Marking issues that we're unlikely to address as WontFix. If it still applies with currently supported CEF versions, and someone would like to submit a PR fix, then we can reopen.

[magreenblatt]

• changed state from "new" to "wontfix"