Performance and memory usage tradeoffs to give the best experience to most users most of the time

[mjethani]

Background

We have made a few changes to Adblock Plus recently to optimize the performance and memory usage of the product. Most of these changes are based on insights from filter list data, the performance characteristics of the most common platforms (currently ​V8), and the default settings in the Adblock Plus browser extension.

Our primary goal is to optimize for platforms with limited resources. As a bonus, these optimizations also benefit desktop users on ​Chrome (primarily), ​Firefox, and ​Edge.

Most of these optimizations are based on conscious tradeoffs with best and worst cases.

The rationale behind each change may be found in the individual tickets listed in the table below.

The list of changes

Ticket	Status	Resolution	Summary	Component	Owner
#6437	closed	fixed	Optimize element hiding emulation for DOM modifications	Core	mjethani
#6727	closed	fixed	Use string rather than map for single-domain filters	Core	mjethani
#6735	closed	fixed	Store domains in lower case	Core	mjethani
#6815	closed	fixed	Avoid multiple copies of domain maps	Core	mjethani
#6856	closed	fixed	Remove FilterStorage.moveSubscription	Core	jsonesen
#6916	closed	fixed	Avoid Set object for filters with only one subscription	Core	mjethani
#6957	closed	fixed	Combine element hiding selector matching with style sheet generation	Core	mjethani
#6967	closed	fixed	Hold on to only non-standard style sheets	Platform	mjethani
#6989	closed	fixed	Optimize CSS rule generation	Core	mjethani
#6992	closed	fixed	Remove keyword-by-filter map and associated dead code	Core	mjethani
#6994	closed	fixed	Use shortcut matching for filters with no content type, no domain/sitekey, and no third-party flag	Core	mjethani
#6999	closed	fixed	Always generate style sheets in the background page	Platform	mjethani
#7003	closed	fixed	Implement alternative filter matching algorithm deprioritizing whitelist filters	Core	mjethani
#7021	closed	rejected	Compress filter text in memory	Core
#7029	closed	fixed	Remove subscriptions property of Filter object	Core	mjethani
#7045	closed	rejected	Optimize V8 memory layout of filter text	Core	mjethani
#7046	closed	fixed	Defer caching of domain maps for element hiding filters and exceptions	Core	mjethani
#7052	closed	fixed	Use string-based matching for literal patterns	Core	mjethani
#7089	closed	fixed	Use type-specific maps for matching non-default types	Core	mjethani
#7096	closed	rejected	Make it possible to lazy initialize a filter's subscriptions	Core	mjethani
#7097	closed	rejected	Make filter objects ephemeral	Core	mjethani
#7208	closed	fixed	Drop superfluous wildcards before processing pattern	Core	mjethani
#7209	closed	fixed	Treat patterns with trailing separators as literals	Core	mjethani
#7244	closed	fixed	Maintain cache of domain-specific style sheets	Core	mjethani
#7245	closed	fixed	Move protocol keyword to end of candidate list	Core	mjethani
#7253	closed	fixed	Pre-render icons for badge on Chromium	Platform	mjethani
#7254	closed	fixed	Expand matcher result cache to ten times the current capacity	Core	mjethani
#7257	closed	fixed	Throttle badge updates	Platform	mjethani
#7265	closed	fixed	Orgnanize request blocking filters by domain	Core	mjethani
#7267	closed	fixed	Optimize memory layout of filter domain maps in matcher	Core	mjethani
#7296	closed	fixed	Implement lightweight URL parsing for canonicalized URLs	Core	mjethani
#7321	closed	fixed	Lower-case non-literal patterns for case-insensitive matching	Core	mjethani
#7435	closed	fixed	Use request object for filter matching	Core	mjethani
#7441	closed	rejected	Add benchmark command to package.json	Core	mjethani
#7444	closed	fixed	Compile patterns for simple filters	Core	mjethani
#7445	closed	rejected	Implement faster parsing of generic element hiding filters	Core	mjethani

Additional notes

This is a meta ticket and should be updated as things evolve.