Since in uBOL filter lists from various sources are combined into
a single list, there must be a way to turn on/off trust level
inside the resulting combined filter list so as to be able to
validate the trust level of filters requiring trust.
This commit adds new parser directives understood only by MV3
compiler to turn on/off trust flag internally.
See `managed_storage.json` for available settings. Currently
only `noFiltering` setting is availale.
`noFiltering` is an array of strings, each being a domain for
which no filtering should occur.
Related discussion:
- https://github.com/uBlockOrigin/uBOL-issues/discussions/35
Avoid using updateContentScripts() as it suffers from an unexpected
behavior, causing injected content scripts to lose proper order
at injection time. The order in which content scripts are injected
is key for uBOL content scripts. Potential out of order injection
was causing cosmetic filtering to be broken.
Use actual storage API to persist data across service worker
wake-ups and browser launches. uBOL was trying to avoid using
storage API, at the cost of somewhat hacky code (using DNR API
to persist settings).
Make use of session storage if available, to speed up
initialization of waking up the service worker (which at this
point is necessary to properly implement cosmetic filtering).
Related issues:
- https://github.com/uBlockOrigin/uBOL-issues/issues/5#issuecomment-1575425913
- https://github.com/w3c/webextensions/issues/403
Currently, there is no other way to inject CSS user styles than to
wake up the service worker, so that it can inject the CSS styles
itself using the `scripting.insertCSS()` method.
If ever the MV3 API supports injecting CSS user styles directly
from a content script, uBOL will be back to be fully declarative.
At this point the service worker is very lightweight since the
filtering is completely declarative, so this is not too much of
an issue performance-wise except for the fact that waking up the
service worker for the sole purpose of injecting CSS user styles
and nothing else introduces a pointless overhead.
Hopefully the MV3 API will mature to address such inefficiency.
Specifically, avoid long list of hostnames for the `matches`
property[1] when registering the content scripts, as this was causing
whole browser freeze for long seconds in Chromium-based browsers
(reason unknown).
The content scripts themselves will sort out which cosmetic filters to
apply on which websites.
This change makes it now possible to support annoyances-related lists,
and thus two lists have been added:
- EasyList -- Annoyances
- EasyList -- Cookies
Related issue:
- https://github.com/uBlockOrigin/uBOL-issues/issues/5
These annoyances-related lists contains many thousands of specific
cosmetic filters and as a result, before the above change this was
causing long seconds of whole browser freeze when simply modifying
the blocking mode of a specific site via the slider in the popup
panel.
It is now virtually instantaneous, at the cost of injecting larger
cosmetic filtering-related content scripts (which typically should
be garbage-collected within single-digit milliseconds).
Also, added support for entity-based cosmetic filters. (They were
previously discarded).
---
[1] https://developer.mozilla.org/en-US/docs/Mozilla/Add-ons/WebExtensions/API/scripting/RegisteredContentScript
Source code of scriplets is now fetched directly from uBO
project, so there is no longer the need to keep duplicate
versions of scriplet code.
All scriplet filters are now supported.
Some filters with entity-based domain option can be salvaged
when there are non-entity-based domain option, but since we are
throwing away the entity-based entries, we are only partially
converting to DNR. This commit will log a warning about this
in log.txt. Before this commit, only non-salvageable filters
were logged.
This reflects the _world_ of the MV3 scripting API:
https://developer.mozilla.org/en-US/docs/Mozilla/Add-ons/WebExtensions/API/scripting/ExecutionWorld
MAIN: page's world
ISOLATED: extension's content script world
Some scriptlets are best executed in either world, so this
commit allows to pick in which world a scriptlet should execute
(default to MAIN).
For instance, the new sed.js scriptlet will now execute in
the ISOLATED world.
Also added additional safeguard against sticky unprocessed-request
status: all unprocessed-request statuses will be cleared after a
minute elapsed following intialization. The idea is that if the
user hasn't care to force a reload of the page, then it's assumed
to be by choice and uBO should stop informing about the status.
What does not work at the time of commit:
Cosmetic filtering does not work:
The content scripts responsible for cosmetic filtering fail when
trying to inject the stylesheets through document.adoptedStyleSheets,
with the following error message:
XrayWrapper denied access to property Symbol.iterator
(reason: object is not safely Xrayable).
See https://developer.mozilla.org/en-US/docs/Xray_vision for more
information. ... css-declarative.js:106:8
A possible solution is to inject those content scripts in the
MAIN world. However Firefox scripting API does not support MAIN
world injection at the moment.
Scriptlet-filtering does not work:
Because scriptlet code needs to be injected in the MAIN world,
and this is currently not supported by Firefox's scripting API,
see https://bugzilla.mozilla.org/show_bug.cgi?id=1736575
There is no count badge on the toolbar icon in Firefox, as it
currently does not support the `DNR.setExtensionActionOptions`
method.
Other than the above issues, it does appear uBO is blocking
properly with no error reported in the dev console.
The adoptedStyleSheets issue though is worrisome, as the
cosmetic filtering content scripts were designed with ISOLATED
world injection in mind. Being forced to inject in MAIN world
(when available) make things a bit more complicated as uBO
has to ensure it's global variables do not leak into the page.
uBO will now verify that at least one unprocessed network requests
at launch should have been blocked in order to warn users of
unprocessed network requests through the `!` toolbar icon badge.
For example, with default filter lists, there is nothing to block
on `wikipedia.org`, and hence in this case it's not useful to
present the user with the `!` badge.
Therefore uBO will not show the badge *only* when at least one
unprocessed network requests should have been blocked had uBO been
ready when it was fired by the browser.
Related commit:
- https://github.com/gorhill/uBlock/commit/769b8da664be
Related feedback:
- 769b8da664 (commitcomment-104695781)
The incomplete filtering status of a given tab at browser launch
will be carried over visually as a yellowish `!` badge until the
web page in the tab is force reloaded, navigated away, or closed.
The purpose is to make it obvious to end users that a web page
has not been filtered properly and to avoid issue reports
related to this.
It is expected that Firefox should never be affected by cases of
yellowish badge -- that is unless the setting "Suspend network
activity [...]" has been disabled, in which case the new behavior
will also be useful to those who disabled the setting.
Raymond Hill