MSE sourced elements should ignore the preload attribute.
See http://w3c.github.io/media-source/#mediasource-attach (First note in the draft at time of commit)
We're treating the attribute as if it was not defined (defaults to `auto`).
We're also ignoring the default action pref for MSE (was already special-cased).
This resolves#981
We try to derive the dictionary to use based on the following priorities:
1) Content preference: the language the user set for the site before.
2) The value of "spellchecker.dictionary.override" which reflects a
global choice of language explicitly set by the user.
3) Language set by the website, or any other dictionary that partly matches that.
Eg. if the website is "en-GB", a user who only has "en-US" will get that.
If the website is generic "en", the user will get one of the "en-*" installed,
pretty much at random.
4) The user's locale
5) Leave the current dictionary set.
6) The content of the "LANG" environment variable (if set)
7) The first spell check dictionary installed.
Stop using the (broken) spellchecker.dictionary "holding pref"
- Enable 128-bit RSA+AES+SHA256 suites
- Disable all static DHE suites because:
* There's no way to negotiate DH key sizes.
* It's barely in use and when it is, it can fall back to RSA
* We now have broader RSA/AES support with stronger SHA-HMAC
This works around a spec conflict in RFC7692, avoiding socket closure when allowed but unknown parameters are sent back by the server.
Added parameters:
- server_no_context_takeover
- server_max_window_bits
- client_max_window_bits
This adds the following suites for web compatibility despite the
deprecated RSA key exchange that makes little sense with a
very strong HMAC or GCM:
TLS_RSA_WITH_AES_256_GCM_SHA384
TLS_RSA_WITH_AES_256_CBC_SHA256
TLS_RSA_WITH_AES_128_GCM_SHA256
TLS_RSA_WITH_AES_128_CBC_SHA256
Only the 256-bit ones are enabled by default.
FinishInternal() won't get called if an error has been posted before,
so the libwebp cleanup calls need to be in a different place; like the
destructor.
Too-large images won't use HQ scaling and also won't display at 1:1 zoom
because we simply don't have the surface cache space available for that kind of out-of-scope image.
It will allow the user to save such images to disk and view in specialized software.
Because of the platform structure, we have to provide a way to add ranges to
Selection objects that aren't associated with a shell/nsFrameSelection.
This commit extends the original implementation with workarounds to add
these ranges by reaching into the internal implementation for adding ranges
and calling those functions directly.
Xidorn Quan
Pale Moon
trav90
janekptacijarabaci
JustOff
Agent Orange
trav90
Nicholas Hurley
rhinoduck