Replace GTK_RESPONSE_ACCEPT (-3) with a non-negative application-defined
response ID (kFilePickerAccept = 0) for the accept button in the GTK file
chooser dialog, and remove the gtk_dialog_set_default_response() call.
GTK treats buttons with negative built-in response IDs as candidates for
the dialog's default widget, which activates on Enter key press. By using
a non-negative ID and not calling set_default_response(), the accept button
is no longer pre-selected, preventing a page from tricking the user into
confirming a file upload by holding Enter before the dialog appears.
GTK's file chooser internally emits GTK_RESPONSE_ACCEPT via file_activated_cb
when the user double-clicks a file, so we keep that case in DoneNonPortal()
alongside kFilePickerAccept.
Instruction sinking was made possible by the introduction of recover
instructions, which are moving instructions to be executed on bailout, but this
is only possible for instructions which are not used.
This feature, which is now removed, was made to sink instructions which are
live, but only in a small set of branches, by moving the computation of these
instructions to bailout paths, and to the concerned if branches.
The current implementation never delivered and remained as dead-code for
multiple years, without having a firm benefit on performance oriented
benchmarks.
Thus, this change removes the part of the Sink optimization which deals with
live instructions but keeps the part of the Sink optimization which deals with
instructions which are only used on bailout paths.
This code made two assumptions:
- That the `SrcChar` type should determine the output type, as used in `CanUpperCaseSpecialCasing`.
- That `AppendUpperCaseSpecialCasing` would always return a value larger than the previous value.
The new code uses `DestChar`'s type for `CanUpperCaseSpecialCasing` and for a more accurate End of Buffer check.
Cache overflow textruns in a better place and stop using a lazy class.
This avoids doing caching in the rendering back-end, and does so where
necessary in TextOverflow (for ellipses)
Removed autodetection of this charset due to unreliability.
There are known security issues around charset auto-detection for ISO-2022-JP.
Given the usage is around 0.000002% of page loads, and Safari does not
support auto-detection of ISO-2022-JP, Chrome also planned to remove
support for it to eliminate the security issues.
See also https://www.sonarsource.com/blog/encoding-differentials-why-charset-matters/
Removal from the menu to avoid social engineering hazards and it being
literally unused on the web.
Moonchild
Kui-Feng Lee
John Schanck
Nicolas B. Pierron
Basilisk-Dev
Andy
Jens Stutte