[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

0.5 released

From: Libor Polčák
Subject: 0.5 released
Date: Wed, 15 Sep 2021 14:11:54 +0200
User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:60.0) Gecko/20100101 Firefox/60.0 SeaMonkey/2.53.9

Hello all,

JSR0.5 was released. All the development is merged to master and tagged. 
Firefox extension is already available in AMO, other stores should follow.

Thank you all for the effort.


* Add fingerprinting defenses based on Farbling developed by the Brave browser 
(improved or added
        wrappers for Canvas, Audio, Web GL, device memory, hardware 
concurrency, enumerateDevices). Most
        wrappers support provisioning of white lies that differ between origins 
and sessions (the
        fingeprint is different across origins and across sessions).
 * We claimed to generate white image fake Canvas value but instead generated 
fully transparent black image. We now generate the white image as it is more 
common in other anti-canvas fingerprinting tools (level 3).
 * toDataUrl() no longer destructs the original canvas.
* We use NoScript Commons Library to simplify some tasks like cross-browser 
 * More reliable early content script configuration.
   * CSP headers no longer prevents the extension from wrapping JS APIs in 
Firefox (Github issue #25)
   * Wrappers should be injected reliably before page scripts start to operate 
(Github issue #40)
 * We use NSCL to wrap APIs in iframes and workers
   * It is no longer possible to access unwrapped functions from iframes and 
workers (Pagure issue #2, Github issue #56)
* Ignore trailing '.' in domain names when selecting appropriate custom level.
* Do not freeze wrappers to prevent fingeprintability of the users of JSR. We 
wrap the correct function
        in the prototype chain instead.
* navigator.getGamepads() wrapper added
* navigator.activeVRDisplays() and navigator.xr wrappers added
* Limit precision of high resolution timestamps in the Event, VRFrameData, and 
Gamepad interface to be consistent
        with Date and Performance precision


reply via email to

[Prev in Thread] Current Thread [Next in Thread]