Spurious updates to registered content scripts would occur when
using some regional lists, for example DEU would cause spurious
updates to registered script `css-procedural`.
Network filters with csp= option will now be enforced.
Caveat: DNR API does not have support for exception csp= rules,
so excepted csp= filters are currently rejected at conversion time.
This commit adds the ability to inject entity-based plain CSS
filters and also a set of the most commonly used entity-based
scriptlet injection filters.
Since the scripting API is not compatible with entity patterns,
the entity-related content scripts are injected in all documents
and the entity-matching is done by the content script themselves.
Given this, entity-based content scripts are enabled only when
working in the Complete filtering mode, there won't be any
entity-based filters injected in lower modes.
Also, since there is no way to reasonably have access to the
Public Suffix List in the content scripts, the entity-matching
algorithm is an approximation, though I expect false positives
to be rare (time will tell). In the event of such false
positive, simply falling back to Optimal mode will fix the
issue.
The following issues have been fixed at the same time:
Fixed the no-filtering mode related rules having lower priority
then redirect rules, i.e. redirect rules would still be applied
despite disabling all filtering on a site.
Fixed improper detection of changes to the generic-related CSS
content script, potentially causing undue delays when for example
trying to access the popup panel while working in Complete mode.
The scripting MV3 can be quite slow when registering/updating
large content scripts, so uBOL does its best to call the API only
if really needed, but there had been a regression in the recent
builds preventing uBO from properly detecting unchanged content
script parameters.
This adds support for `redirect=` filters. As with `removeparam=`
filters, `redirect=` filters can only be enforced when the
default filtering mode is set to Optimal or Complete, since these
filters require broad host permissions to be enforced by the DNR
engine.
`redirect-rule=` filters are not supported since there is no
corresponding DNR syntax.
Additionally, fixed the dropping of whole network filters even though
those filters are still useful despite not being completely
enforceable -- for example a filter with a single (unsupported) domain
using entity syntax in its `domain=` option should not be wholly
dropped when there are other valid domains in the list.
With the new csstree-based parser, it should now be
safe to parse `-abp-has` as declarative. There are over
a hundred such cosmetic filters in EasyList, and we want
to have these filters declaratively enforced whenever
possible in order to let the browser do the work natively
rather than rely on JS code.
The addition of those scriptlets will help uBOL to properly
filter more anti-blocker mechanisms on various site and also prevents
popups on many sites.
Also, fixed properly injecting scriptlets in embedded frames -- it
turns out I misunderstood the `allFrames` settings in the `scripting`
API.
Fixed trusted sites not being excluded from declarative
scripting.
Assign "uBOL_"-prefixed name to anonymous scripting functions
so that they can be easily found in performance profiler results
in dev tools.
Imrpove spread of chunks of filters across declarative scripting
files.
Consequently, AdGuard URL Tracking Protection (AUTP) has been
added to the set of available filter lists.
However, removeparam= equivalent DNR rules can only be enforced
when granting uBOL broad permissions. If broad permissions are
not granted, removeparam= equivalent DNR rules are ignored.
Exception removeparam= filters are not supported, and these are
present in AUTP and meant to unbreak some websites which are
known to break as a result of removing query parameters.
This is issue might be mitigated in the future by making the
conversion from filters to DNR rules more complicated but this
can never replace the accuracy of uBO's filtering engine being
able to fully enforce arbitrary exception removeparam= filters.
Also, it is not possible to translate regex-based removeparam=
values to DNR rules, so these are dropped at conversion time.
As with other filters to DNR rules conversion, the converter
coallesce many distinct removeparam= filters into fewer DNR
rules.
The new parser no longer uses the browser DOM to validate
that a cosmetic filter is valid or not, this is now done
through a JS library, CSSTree.
This means filter list authors will have to be more careful
to ensure that a cosmetic filter is really valid, as there is
no more guarantee that a cosmetic filter which works for a
given browser/version will still work properly on another
browser, or different version of the same browser.
This change has become necessary because of many reasons,
one of them being the flakiness of the previous parser as
exposed by many issues lately:
- https://github.com/uBlockOrigin/uBlock-issues/issues/2262
- https://github.com/uBlockOrigin/uBlock-issues/issues/2228
The new parser introduces breaking changes, there was no way
to do otherwise. Some current procedural cosmetic filters will
be shown as invalid with this change. This occurs because the
CSSTree library gets confused with some syntax which was
previously allowed by the previous parser because it was more
permissive.
Mainly the issue is with the arguments passed to some procedural
cosmetic filters, and these issues can be solved as follow:
Use quotes around the argument. You can use either single or
double-quotes, whichever is most convenient. If your argument
contains a single quote, use double-quotes, and vice versa.
Additionally, try to escape a quote inside an argument using
backslash. THis may work, but if not, use quotes around the
argument.
When the parser encounter quotes around an argument, it will
discard them before trying to process the argument, same with
escaped quotes inside the argument. Examples:
Breakage:
...##^script:has-text(toscr')
Fix:
...##^script:has-text(toscr\')
Breakage:
...##:xpath(//*[contains(text(),"VPN")]):upward(2)
Fix:
...##:xpath('//*[contains(text(),"VPN")]'):upward(2)
There are not many filters which break in the default set of
filter lists, so this should be workable for default lists.
Unfortunately those fixes will break the filter for previous
versions of uBO since these to not deal with quoted argument.
In such case, it may be necessary to keep the previous filter,
which will be discarded as broken on newer version of uBO.
THis was a necessary change as the old parser was becoming
more and more flaky after being constantly patched for new
cases arising, The new parser should be far more robust and
stay robist through expanding procedural cosmetic filter
syntax.
Additionally, in the MV3 version, filters are pre-compiled
using a Nodejs script, i.e. outside the browser, so validating
cosmetic filters using a live DOM no longer made sense.
This new parser will have to be tested throughly before stable
release.
This solves the following remaining issues regarding specific cosmetic
filtering:
- High rate of false positives in last build
- High number of generated content css files in the package
First iteration of adding scriptlet support. As with cosmetic
filtering, scriptlet niijection occurs only on sites for which
uBO Lite was granted extended permissions.
At the moment, only three scriptlets are supported:
- abort-current-script
- json-prune
- set-constant
More will be added in the future.
Specific plain CSS cosmetic filters are now supported.
Cosmetic filtering will occur only after the user explicitly
grant uBO extended permissions for a given site, so that it
can inject CSS on the site.
A new button in the popup panel allows a user to grant/revoke
extended permissions to/from uBO Lite for the current site.
More capabilities will be carefully added for when extended
permissions are granted on a site, so specific cosmetic
filtering through plain CSS is the first implemented capability.
Generic and procedural cosmetic filtering is not implemented.
The current implementation for plain CSS cosmetic filters is
through declarative content injection, which does not require
the service worker to be alive, the browser takes care to
inject the cosmetic filters.
However declarative CSS injection does not support user
styles, so the injected cosmetic filters are "weak". I consider
this is a browser issue, since user styles are supported by
Chromium, there is just no way in the API to specify user
styles for the injected content.
Also:
- Fixed dark theme issues
- Added Steven Black's hosts file
Keep in mind all this is very experimental and implementation
details in this release may (will) greatly change in the future.
Related discussion:
- a0a9497b4a (commitcomment-62560291)
The new setting, when disabled (enabled by default), allows a user
to prevent uBO from waiting for all filter lists to be loaded
before allowing network activity at launch. The setting is enabled
by default, meaning uBO waits for all filter lists to be loaded in
memory before unsuspending network activity. Some users may find
this behavior undesirable, hence the new setting.
This gives the option to potentially speed up page load at launch,
at the cost of potentially not properly filtering network requests
as per filter lists/rules.
For platforms not supporting the suspension of network activity,
the setting will merely prevent whatever mechanism exists on the
platform to mitigate improper filtering of network requests at
launch. For example, in Chromium-based browsers, unchecking the
new setting will prevent the browser from re-loading tabs for
which there was network activity while in "suspended" state at
launch.
Fixed serious regression in previous dev build in applying
`csp=` filters. Reported internally by uBO team.
Promote usage of `removeparam` in code instead of `queryprune`,
which is to be deprecated.
Removed test against previously tested hostname in
FilterHostnameDict since as per various benchmark, the
test does not really help.
Remove serialization API in Node.js code as the API is now
present in SNFE itself.
Not sure this can really happen, but if ever Math.random() would
return `0.9999999999999999`, the attribute name would start with
`{`, i.e. an invalid attribute name.
Related issue:
- https://github.com/uBlockOrigin/uBlock-issues/issues/1732
The regression affect filter with the `important` option when
the following conditions were fulfilled:
- The filter pattern is pure hostname
- The filter has not one of the following options:
- domain
- denyallow
- header
- strict1p, strict3p
- csp
- removeparam
- There is a matching exception filter
Related commit:
- a2a8ef7e85
A related mocha test has been added in order to detect this
specific regression in the future through `make test`.
This is a replacement for the dubious approach when the
extension itself was used to run benchmarks to detect
performance and filtering behavior regressions.
The main nodejs flavor is "npm", which is to be used to
lint/test and the publication of an official npm
package -- and by design it has dependencies on mocha,
eslint, etc.
A new flavor "dig" has been created with minimal
dependencies and which purpose is to easily allow to
write specialized code to investigate local code changes
in uBO -- and it's not meant for publication.
Consequently, "make nodejs" has been replaced with
"make npm", and a new "dig" target has been added to the
makefile, to be used for instrumenting local code changes
for investigation purpose.
Whereas before the string segment was encoded as:
LL OOOOOOOOOOOO
where L are the upper 8 bits and used to encode the length
of the segment, and O are the lower 24 bits and used to
encode the offset of the string data in the character
buffer, the new code encode as follow:
OOOOOOOOOOOO LL
And furthermore the most significant bit of the length
LL is now used to mark whether the current string segment
is a label boundary.
This means a cell can't reference a segment longer then
127 characters. To work around this limitation for when a
segment is longer than 127 characters (a rare occurrence),
the algorithm will simply split the segment into multiple
adjacent cells.
As a result, there is no longer a need to encode
"boundariness" into special cells, which simplifies
both the storing and matching algorithms.
Additionally, added minimal documentation for the NPM
package on how to import and use HNTrieContainer as a
standalone API.
For clients who may wish to persist the intermediate compiled form
in order to be able to skip costly parsing operation when the
list is fed to the static network filtering engine.
In the static network filtering engine (snfe), the
compiling-related code was spread across two classes.
This commit makes it so that all the compiling-related
code is in FilterCompiler class, which clear purpose is
to compile raw filters into a form which can be persisted
and later fed to the snfe with no parsing overhead.
To compile raw static network filter, the new approach is:
snfe.createCompiler(parser);
Then for each single raw filter to compile:
compiler.compile(parser, writer);
The caller is responsible to keep a reference to the
compiler instance for as long as it is needed. This removes
the need for the clunky code used to keep an instance of
compiler alive in the snfe.
Additionally, snfe.tokenHistograms() has been moved to
benchmarks.js, as it has no dependency on the snfe, it's
just a utility function.
