Pale Moon: Release notes

v32.4.0.1 (2023-09-14)

• Fixed a WebP decoder issue (CVE 2023-4863)

v32.4.0 (2023-09-05)

• Implemented the BigInt primitive type for JavaScript. See implementation notes.
  
• Implemented Big(U)Int64 array support.
• Implemented ergonomic brand checks for JavaScript class fields.
  
• Aligned the Performance API with the Timeline v2 spec.
• Aligned the handling of flex/grid percentages resolving against the parent with other browsers. See implementation notes.
  
• Added or updated several user-agent overrides for problematic websites.
• Added 2 preferences to allow users to disable CSS animations and transitions. See implementation notes.
  
• Improved compatibility with MacOS 14.
• Fixed an important, intermittent JavaScript crash related to garbage collection.
  
• Fixed several crashes.
• Fixed several debug build related issues.
• Fixed an issue building on SunOS related to the spelling library.
  
• Developer: Added ASan support for building with MSVC.
• Added the .xll file extension to the executable extensions list.
• Security issues addressed: several potential security issues that do not have a CVE number. DiD
• UXP Mozilla security patch summary: 1 fixed, 3 DiD, 17 not applicable.

• The BigInt primitive (base number format) in JavaScript allows JavaScript to handle excessively large integers (whole numbers). This primitive is especially useful for specialized scientific applications that need very large yet accurate numbers, but has seen widespread adoption for an as of yet unknown reason as part of web frameworks, causing general web compatibility issues for Pale Moon when scripts expect BigInt support and instead have an error thrown. We have now implemented this primitive for use so we no longer have compatibility issues with these frameworks. It is still unknown why BigInt is in use there and for what. Critical note: BigInt might be tempting to consider for JS-backed cryptography but this is very ill-advised, as BigInt operations are, by their nature, not constant-time and allow timing and side-channel attacks.
• Flex and grid item sizes in percentages would previously be resolved against the parent like other elements, according to a very long-standing practice that stems from the Internet Explorer days. Mainstream browsers have, however, made an exception for flex items and grid items to no longer do this. We have now made the same exception for these types of elements which should solve layout issues on some websites (notably reserving too much space for items, often resulting in very large areas of whitespace or items being pushed out of view).
• Two preferences were added (layout.css.animation.enabled and layout.css.transition.enabled) to allow users to completely disable CSS-based animations and transition effects. This was a request by users as both a performance and accessibility consideration. Please note that in some cases, disabling animations and transitions may have an impact on final web page layout, so you may run into some issues when disabling these animations and transitions as the web pages were designed to use them.

v32.3.1 (2023-07-18)

• Fixed intermittent crashes related to the performance API.
• Fixed intermittent issues with JavaScript malfunctioning in chrome scripts (causing faults in the UI and extensions).

v32.3.0 (2023-07-11)

• Added the (hidden) preference browser.history.menuMaxResults to allow users to control how many history entries are listed in the menu. Setting this to 0 will hide history menu entries altogether, and any positive number configures how many entries the entries are limited to. The default if not defined is 15.
• Switched C++ language level used to C++14 on all platforms.
  
• Web compatibility and scripting improvements:
  
• Implemented geometry .from* static constructors for web compatibility.
• Implemented partial support for CSS calc() in color keywords.
• Implemented Array "find from last" feature (findLast and findLastIndex).
• Implemented Object.hasOwn(object,property).
• Implemented several additional Intl API methods and functions. This improves web compatibility with sites making use of things like hourCycle, advanced DateTimeFormat, Intl.Locale, and Intl as a constructor.
  
• Cleaned up some unused code.
• Removed support for Mozilla "experiment" type extensions.
• Improved the JavaScript garbage collector's sweeping. This should fix a few intermittent crashes and improve performance.
• Implemented some structural changes to the source to make future porting easier, and preparing for switching to C++17.
  
• Removed handling of symlinks for directory listings to prevent potential security issues by walking symlinks when uploading. This effectively reverts a change made in Firefox 50 where this functionality was introduced. A case of "Not such a good idea after all" ;-)
  
• Updated the list of extensions on Windows treated as "executable".
• Security issues addressed: CVE-2023-37208.
• Made preparations for requiring Authorization in CORS ACAH preflight.
  Since no browser honors this part of the spec at the moment this is left disabled until there is consensus among browsers.
  
• UXP Mozilla security patch summary: 2 fixed, 2 rejected, 20 not applicable.

v32.2.1 (2023-06-19)

• Fixed a crash in devtools in some OOM situations.
  
• Fixed crashes when internal script execution was blocked by extensions.
  
• Fixed crashes with WebComponents' ::slotted selector.
  
• Disabled incremental cycle collector by default to avoid performance regressions.
  
• Updated the default override for chase.com to work around issues.

v32.2.0 (2023-05-16)

• Implemented dynamic module imports. See implementation notes.
• Implemented exporting of async functions in modules.
• Implemented JavaScript class fields. See implementation notes.
  
• Implemented logical assignment operators ||=, &&= and ??=.
  
• Implemented a solution for websites using the officially deprecated ambiguous window.event. This is disabled by default but can be enabled through about:config's dom.window.event.enabled preference. See implementation notes.
• Implemented self.structuredClone() (this may be very obscure to anyone except web developers. Apologies ;-) )
• Implemented Element.replaceChildren. Once again primarily a web developer note.
  
• Improved Shadow DOM :host matching.
• Implemented WebComponents' CSS ::slotted() and related functionality.
• Improved page caching in our memory allocator.
  
• Added support for FFmpeg 6.0, especially important for bleeding-edge Linux distros.
• Fixed a potential drawing deadlock for images, specifically SVG. This solves a number of hang-on-shutdown scenarios.
  
• Fixed various crashes related to WebComponents and our recent JavaScript work.
• Fixed various build-from-source issues on secondary target platforms.
• Fixed various small browser front-end scripting issues that could lead to errors or broken functionality.
  
• Fixed handling of async (arrow) functions declared inside constructors.
• Fixed various small JavaScript conformance issues.
• Fixed an issue where JavaScript (only in modules) would not properly create async wrappers.
  
• Updated the DOM Performance API to the current spec (User Timing L3).
  See implementation notes, especially if you intend to use this in web content for critical functionality.
  
• Updated keypress event handling to send keypress events on Ctrl+Enter.
• Updated internal JavaScript structures to make future porting easier, as well as improve JavaScript performance.
• Updated window handling and styling on Mac.
• Updated the Freetype lib to 2.13.0.
• Updated the Harfbuzz lib to 7.1.0.
• Updated our DNS lookup calls to use inet_ntop() instead of the deprecated inet_ntoa().
• Updated the Fetch API to use the global's base URL instead of the entry document's base URL for spec compliance.
  
• We no longer support the outmoded fontconfig on GTK systems.
• We no longer parse or return the body of known-empty responses from servers (content-length of 0, or in case of HEAD or CONNECT methods).
  
• Implemented scaled font caching on GTK, improving performance.
  
• Fixed a build issue when building for Linux on ARM64 on later distros.
  
• Split out more parts of the browser into separate .dll files on Windows to reduce compiler strain and an oversized xul.dll
• Removed mozilla::AlignedStorage (code cleanup).
• Builds for FreeBSD now use xz for packaging instead of bzip2. By request, we now also offer GTK2 builds for FreeBSD.
  
• Merged the preference dom.getRootNode.enabled into the dom.webcomponents.enabled pref. See implementation notes.
• Fixed a potential DoS issue with JPEG decoding.
• Fixed a potential issue in Windows widget code that could lead to crashes.
• Disabled potentially hazardous external protocols on Windows.
• Added known-problematic .dlls to the internal blocklist.
• Security issues addressed: CVE-2023-32209, CVE-2023-32214 and several others that do not have a CVE designation.
• UXP Mozilla security patch summary: 4 fixed, 1 rejected, 27 not applicable.

• JavaScript modules have various methods of being loaded into web page content. One of the later introduced methods is a function-style import() declaration, so-called "dynamic module imports" that has been used by various web frameworks, causing issues for Pale Moon resulting in blank pages in most cases (since the websites would not actually use document structure HTML, but rather JavaScript to create content, all from imported modules). This has been a major web compatibility issue lately and we're pleased to announce that this complex bit of machinery has been implemented.
• JavaScript's language specification is continuing to be watered down from a prototyping language towards a more "C-like" hybrid. As part of that effort, JavaScript classes were introduced in ECMAScript 6, and now further expanded in ES2022 with class fields and private class fields/methods, as well as statics. We should have a complete implementation of this now, which constitutes the more important parts of the ES2022 language update.
• The use of the outdated Microsoft Internet Explorer global window.event has been a pervasive web compatibility issue for us, especially since it was officially deprecated and we never implemented this ambiguous and unreliable property that is highly-context sensitive. Websites should use the event as passed into the event handler to get the event source instead. However, since neither Chrome nor Firefox have dropped this and seem to be playing a game of "chicken", it remains in use on the web. To deal with this conflict, we have now implemented the equivalent behind a preference to enable users to (temporarily) use the global window.event while webmasters update their websites. We hope the Google camp will finally drop this one soon so we can be done with this legacy quirk.
• The DOM Performance API was updated to the User Timing level 3 spec. It should be critically noted that the DOM Performance API was never designed to be used as a matter of course on published content, and was designed only for page performance analysis use by web designers. Of course, as part of making dev tools available to the web, a lot of abuse ensued because of the accurate navigation and timing measurements that this API can provide (looking at you, Google!). Because of tight integration with web content analysis, the older spec implementation we had was causing issues and actually breaking some services, so we updated it, but with a few important key differences:
• In Pale Moon, we keep navigation timing disabled because it's a notable privacy issue for the data it can gather (exact navigational events and timings). If you're a web dev and need these timing measurements, you can enable them with dom.enable_performance_navigation_timing.
  
• Our implementation, contrary to the spec, does not allow unlimited recording of performance events (effectively logging every page event!) which can also rapidly eat up memory. Instead we enforce a sane default quota that should be roomy enough for all legitimate use, but prevents runaway resource use or extensive logging of user actions.
• If the set quota is reached, a warning will be printed in the console and the recorded performance events will be thrown away. If you (foolishly) rely on Performance API events for your web application to function, be aware this may cause compatibility issues as the API was, again, not designed to be used in such a fashion. For event handling, there are much better alternatives available which do not involve extensive recording of user data or relying on a developer tool API.
• We've historically implemented the DOM getRootNode function as it was being used in the wild as a standalone function, however its main intent has always been to be a helper function part of Shadow DOM/WebComponents. As such we have now merged the preference into the WebComponents preference, enabling and disabling it along with the rest of our WebComponents implementation.

v32.1.1 (2023-04-18)

• Fixed a crash in CompareDocumentPosition with Shadow DOM.
• Fixed a crash with display:contents styling.
• Added a preference to disable the TLS 1.3 protocol downgrade sentinel (see implementation notes).
• Changed the way large clipboard copy/paste operations are handled, improving privacy (see implementation notes).
• Improved filename safety when saving files to prevent potential environment leaks (bis).
• Improved sanity checks of MIME type headers.
• Security issues addressed: CVE-2023-29545 and CVE-2023-29539.
• UXP Mozilla security patch summary: 2 fixed, 1 rejected, 49 not applicable.

• Some proxies and middleware boxes improperly handle the TLS 1.3 protocol handshake causing an insecure downgrade to TLS 1.2. With our recent update of NSS, Pale Moon no longer allows this kind of protocol downgrade when trying to establish a TLS 1.3 connection to a server. The resulting error is ssl_error_rx_malformed_server_hello with an inability to connect to the server. To enable users to still connect to the servers or devices in question, we've added an option to switch off the downgrade sentinel. To switch it off as a temporary workaround, set security.tls.hello_downgrade_check to false.
• If copy and paste operations to/from the browser are performed, Pale Moon writes clipboard contents to disk in a temporary cache file if the copy/paste amount is particularly large, to avoid using large amounts of memory to hold this data. The average paste/clipboard size doesn't tend to hit this limit in which case it is just held in memory.
  Previously, these cache files, while in the O.S. temporary file location (%TEMP% or /tmp), would not be consistently cleaned up, potentially causing privacy issues if persisted. This was changed to using auto-cleaning anonymous temp files, improving user privacy and relying less on the O.S. or user performing cleanup of temporary file storage. Thanks to Sandra for pointing this out and providing the patch.

v32.1.0 (2023-03-21)

• Shadow DOM and CustomElements, collectively making up WebComponents, have been enabled by default which should bring much broader web compatibility to the browser for many a site that uses web 2.0+ frameworks. See implementation notes.
• Tab titles in the browser now fade if they are too long instead of using ellipses, to provide a little more readable space to page titles. Note that this may require some updates to tab extensions or themes.
• A number of site-specific overrides have been updated or removed because they are no longer necessary or current with the platform developments in terms of web compatibility. We could use your help evaluating the ones that are still there; see the issue on our repo.
• Updated our promises and async function implementation to the current spec.
• Implemented Promise.any()
  
• Fixed several crashes related to regular expression code.
• Improved regular expression object handling so it can be properly garbage collected.
• Fixed some VP8 video playback.
• Fixed an issue where the caret (text cursor) would sometimes not be properly visible.
• Updated the embedded emoji font.
• Implemented the :is() and :where() CSS pseudo-classes.
• Implemented complex selectors for the :not() CSS pseudo-class.
• Implemented the inset CSS shorthand property.
• Implemented the env() environment variable CSS function. See implementation notes.
  
• Implemented handling for RGB encoded video playback (instead of just YUV).
• Implemented handling for full-range videos (0-255 luminance levels) giving better video playback quality.
• Removed the WebP image decoder pref. See implementation notes.
• Enabled the Web text-to-speech API by default (only supported on some operating systems).
• Updated NSPR to 4.35 and NSS to 3.79.4
• Cleaned up unused "tracking protection" plumbing. See implementation notes.
• Cleaned up URI Classifier plumbing (Google SafeBrowsing leftover).
• Fixed several intermittent and difficult-to-trace crashes.
• Improved content type security of jar: channels. DiD
• Improved JavaScript JIT code generation safety. DiD
• Fixed potential crash scenarios in the graphics subsystem. DiD
• Improved filename safety when saving files to prevent potential environment leaks.
• Security issues addressed: CVE-2023-25751, CVE-2023-28163 and several others that do not have a CVE.
  
• UXP Mozilla security patch summary: 1 fixed, 4 DiD, 14 not applicable.

• Google WebComponents has been long-running major feature work in UXP. We're finally at a level with this (after several setbacks and brick-walling) that it can be enabled by default. Please note that while this greatly improves web compatibility with many Chrome-focused websites using these controversial technologies, our implementation is not yet complete and more work is necessary. As a result, this change to enable it by default may actually break some previously-working websites as well, but it's expected the majority will work at our current state of implementation. Please visit the forum if you need help with web compatibility issues.
• The env() CSS function was implemented for compatibility with websites that rely on this without fallback. Note that this function actually has no real use for desktops as it is primarily used to indicate environmental restrictions of mobile screens, e.g. extra space needed to avoid a camera notch or folding screen margin. However, due to the way certain sites implement their styling in a mobile-first approach, it is assumed that this function is available on all systems and in all browsers by these sites. Note that Pale Moon simply hard-codes queried values here.
• WebP images have had a stable and complete implementation in Pale Moon for a long time now, so the preference to disable support for it has been removed, as it's considered by now to be one of the "staple" image formats supported by web browsers. This was done to reduce complexity for content negotiation, especially since we're adding more support for JPEG-XL that still isn't as-complete. From here on out, we simply always support WebP decoding.
• While we've had a preference for "tracking protection" in our browser implementation (in about:config), this marketed feature of Firefox was never adopted by us, because it is for the most part a service-based feature, and the non-service parts were undesirable as they were crippling useful APIs. Our effective protection against tracking has not changed, we have simply removed the preference and plumbing for a non-functional service feature that would potentially give the false impression it would do anything.
  As a reminder, if you are concerned about tracking, use a competent adblocker extension, and enable "Tell sites not to share or sell my data" in Preferences -> Privacy under "Data Privacy". You may also want to enable "canvas poisoning" by setting canvas.poisondata to true in about:config to reduce the risk of fingerprinting through canvases.
  

v32.0.1 (2023-02-21)

• Fixed a crash in the new regular expression code.
• Added {Extended_Pictographic} unicode property escape to regular expressions.
• Fixed a regression in regular expressions for literal parsing of invalid ranges.
• Updated NSS to pick up fixes.
• Security issues addressed: CVE-2023-25733 DiD, CVE-2023-25739 DiD and CVE-2023-0767.
• UXP Mozilla security patch summary: 1 fixed, 2 DiD, 14 not applicable.

v32.0.0 (2023-01-24)

• Implemented Regular Expression named capture groups.
• Implemented Regular Expression unicode property escapes.
• Re-implemented Regular Expression lookaround/lookbehind (without crashing this time ;) ).
• Implemented progressive decoding for JPEG-XL.
• Implemented animation for JPEG-XL.
• Implemented a compatibility mode for <button> elements. See implementation notes.
  
• Renamed CSS offset-* properties to inset-* to align with the latest spec and the web.
• Fixed CSS inheritance and padding issues in some cases.
• Aligned parsing of incorrectly duplicated HSTS headers with expected behavior (discard all but the first one).
• Implemented a method to avoid memory exhaustion in case of (very) large resolution animated images.
  
• Updated the JPEG-XL and Highway libraries to a recent, stable version.
• Cleaned up some unused CSS prefixing code.
  
• Improved the ability to link on *nix operating systems with other linkers than gcc's default.
• Stability improvements (potential crash fixes).
• Security issues addressed: CVE-2023-23598, CVE-2023-23599 and several others that do not have a CVE number.
• UXP Mozilla security patch summary: 4 fixed, 2 DiD, 19 not applicable.

• We're working on finalizing official builds for Mac OS and FreeBSD. These are currently in beta and can be downloaded from the Contributed Builds page. Please note that you may run into some system compatibility issues with these builds. If you do, please go to the forum and report it in the appropriate board!

• To provide users with a temporary work-around for non-compliant websites, a compatibility mode for <button> elements was implemented, which will treat <button> elements as generic containers instead of actual form button elements. This has been necessary because Chrome is not standards compliant in this respect and website developers regularly make the mistake of trying to use active content on button faces and expecting pointer events to end up being sent to the active content and not the button (which is not what the standard prescribes! See "content model" on the standards page stating there "must be no interactive content descendant"). Webmasters should be alerted to this compliance issue, but it can (temporarily) be worked around in the browser from this point for forward by setting the preference dom.forms.button.standards_compliant to false and restarting the browser. Note that this is a workaround and the only actual solution is advocacy for the standard and more browsers becoming standards compliant.