Given that all tokens extracted from one single URL are potentially
iterated multiple times in a single URL-matching cycle, it pays to
ignore extracted tokens which are known to not be used anywhere in
the static filtering engine.
The gain in processing a single network request in the static
filtering engine can become especially high when dealing with
long and random-looking URLs, which URLs have a high likelihood
of containing a majority of tokens which are known to not be in
use.
Related commit:
- 99390390fc
The token information available at compile time can be stored
in the filter to be used at match() time. This allows the use of
startsWith() rather than a more costly indexOf() call as a first
quick test to detect mismatches.
Due to how web pages typically load secondary resources and due
to how HNTrieContainer instances are used in uBO, there is a
great likelihood that the result of a previous call to
HNTrieRef.matches() can be reused in a subsequent call.
This has been confirmed by instrumenting HNTrieRef.matches().
Since uBO uses distinct HNTrieContainer instances to either
match against the request or the origin hostnames, this
means a high likelihood of repeated calls to HNTrieRef.matches()
with the same hostname as argument, hence a performance gain
when caching the argument+result -- as despite that
HNTrie.matches() is fast, comparing two short strings is even
faster if this allows to skip HNTrie.matches() altogether.
Performance- and memory-related work. Three more classes have
been created to avoid regex-based filters internally.
Purpose is to enforce filters which have only one single
wildcard in their pattern, a common occurrence. The filter
pattern is split in two literal string segments.
Similar as above, with the added condition that the filter is
hostname-anchored (`||`). The "Wildcard2" variant is a further
specialization to enforce filters where the only wildcard
is immediately preceded by the `^` special character, again
a very common occurrence.
Using two literal string segments in lieu of regexes allows to
quickly detect a mismatch by just testing the first segment.
Additionally, this reduces memory footprint as regexes are
much more expensive memory-wise than plain strings.
These three new filter classes allow to replace the use of
5276 regex-based filters internally with plain string-based
filters.
Often-called isHnAnchored() has been further fine-tuned to
avoid as much work as possible. I have also observed that
using an arrow function for closure-purpose helps measurably
performance, as per built-in benchmark.
The purpose of using a custom base128 encoder is to
convert array buffers into strings, to allow a direct
string-to-array buffer conversion at load time:
string => array buffer
Whereas a JSON array would require an extra step:
JSON array as string => JS array => array buffer
Turns out that the current use of a custom base128 encoding
results in a significantly larger selfie storage usage when
converting array buffers into strings.
Speculation: possibly the browser convert the strings to
save into JSON strings internally. Since the custom base128
encoder is likely to cause the resulting string to contain
a lot of unprintable ASCII characters, these will need to
be escaped when converted to JSON -- escaped characters
occupy more space than non-escaped ones.
Using a sequence of base 64 numbers means only printable
will be present in the output string, hence no escaping
necessary. I have observed significant reduction in
storage usage for selfie purpose.
Related issue:
- https://github.com/uBlockOrigin/uBlock-issues/issues/528#issuecomment-484408622
Following STrie-related work in above issue, I noticed that a large
number of filters in EasyList were filters which only had to match
against the document origin. For instance, among just the top 10
most populous buckets, there were four such buckets with over
hundreds of entries each:
- bits: 72, token: "http", 146 entries
- bits: 72, token: "https", 139 entries
- bits: 88, token: "http", 122 entries
- bits: 88, token: "https", 118 entries
These filters in these buckets have to be matched against all
the network requests.
In order to leverage HNTrie for these filters[1], they are now handled
in a special way so as to ensure they all end up in a single HNTrie
(per bucket), which means that instead of scanning hundreds of entries
per URL, there is now a single scan per bucket per URL for these
apply-everywhere filters.
Now, any filter which fulfill ALL the following condition will be
processed in a special manner internally:
- Is of the form `|https://` or `|http://` or `*`; and
- Does have a `domain=` option; and
- Does not have a negated domain in its `domain=` option; and
- Does not have `csp=` option; and
- Does not have a `redirect=` option
If a filter does not fulfill ALL the conditions above, no change
in behavior.
A filter which matches ALL of the above will be processed in a special
manner:
- The `domain=` option will be decomposed so as to create as many
distinct filter as there is distinct value in the `domain=` option
- This also apply to the `badfilter` version of the filter, which
means it now become possible to `badfilter` only one of the
distinct filter without having to `badfilter` all of them.
- The logger will always report these special filters with only a
single hostname in the `domain=` option.
***
[1] HNTrie is currently WASM-ed on Firefox.
As a development tool for investigation purpose. To use, enter the
following at uBO's dev console:
µBlock.staticNetFilteringEngine.filterClassHistogram()
In the static network filtering engine, `google` token is too
generic and probably leads to too many false positives, beside
causing too large filter bucket.
