The state being restored here would be file browser state, which might confuse the user if they are opening a folder via “mate” and then get a file browser restored for that project folder that show another location than the actual folder they open.
The reason we ignore the state when loading, rather than not saving it in the first place, is that by saving the state we make the folder show up in File → Open Favorites…
Previously we would set a mark, mutate the buffer, and then restore selection based on how the mark was moved.
When a mutation was changing the buffer to its existing value (e.g. toggling case of a space character) then the buffer would effectively ignore the mutation, and the mark would not be properly updated, meaning for “toggle case of character” the caret would not be moved forward.
Additionally we previously always moved to the end of a mutation, even if it was done on a user selection. We now only move to the end if the user didn’t select anything (but ran the mutation on an implicitly selected unit, like a word). The “toggle case of word” action is affected by this.
Iteration was unintentionally changed (in c2397484b8) to iterate over const references, which would prevent the referenced types to be updated.
The reason the compiler did not complain about it is due to the catch-all template function, indirectly changing visitation from non-const to const, and providing no-op visitation functions for the const types.
For this reason, I have replaced the template (member) function with functions for all types.
This is mainly because the SOMAXCONN constant looks better (in the source) than some arbitrarily chosen number. The constant is presently defined as 128 so it doesn’t seem wasteful, and presumably the queue will only take up space if we are not actually responding to socket connections.
Previously we would only pass the repository’s root to scm::info, and thus the check on the path passed in was enough, but today we are passing in paths for which we do not yet know the repository root, so the check should (also) be done for the path that ends up containing version control meta data.
We still do the initial check since it also rejects non-local paths, thus avoiding stat’ing paths on network file systems.
ZWNJ is so common in Persian texts, and its unicode bi-direction class in neutral.
But the representation chunk contains latin characters and are so left-to-right.
So, when ZWNJ is replaced by the representation chunk, rtl texts are rendered incorrectly.
ZWJ and ZWSP are from the same class, but less common at least in Persian texts.
For example if multiple files with same base name are modified or untracked, then previously it wasn’t possible to tell them apart (other than using the tool tip which show full path).
This is incase the item’s display name is different than the file’s display name, for example if the display name contains part of the path’s parent (to disambiguate it).
Instead we manually call recalculatesKeyViewLoop on the window. This is because the auto-recalculation messes up. First time the window is brought up, it works fine, but the second time, nextValidKeyView will return ‘nil’ for the last view (causing the key view loop to not be a loop).
Currently for VoiceOver user when the text cursor is at the end of the search
field (which is 99% of the time) and the user wants to navigate the results
in the table view using arrow down, then after each arrow down, VoiceOver first
plays a "end-of-text" sound, then announces whole contents of the search field,
and only then announces the newly selected search result. (The same happens
when text cursor is at the beginning of the search field and user presses
arrow up.)
This is probably a bug in the way AppKit handles text field accessibility - see
http://lists.apple.com/archives/accessibility-dev/2014/Feb/msg00019.html
I also reported this as <rdar://problem/16271507>
I hereby present a hacky workaround for this - a subclass of NSSearchField
that tricks accessibility into thinking there is one extra space before and
after the text in the search field. Therefore VoiceOver will not think the user
is at the end of text when they actually are, and therefore will not play the
"end-of-text" sound and announce whole contents of the search field before
getting to the information user wants (the newly selected search result).
While this is a bit hacky, along with the previous commit it allows VoiceOver
to have the same instant great experience as a sighted user when filtering
and browsing results in OakChooser. So I think it's worth the hackiness until
there is a better alternative.
There is only last issue, as currently VoiceOver is very chatty when the
OakChooser window appears: first says "Go to file", if user interrupts it with
an action (i.e. moving VO cursor or typing a character), then VoiceOver starts
telling the whole title of the window, and only after user interrupts it again
does VoiceOver say or announce only the user actions (including search results
table selection change).
So the simpliest if the user wants to start navigating the table items
immediately after showing the OakChooser's window (i.e. without entering
a search term) is to, after showing the OakChooser (e.g. with ⌘T or ⇧⌘T),
quickly type any letter and immediately delete it with backspace and then
use arrow down/up.
This solves accessibility of situation of a user browsing results in
OakChooser (currently File Chooser and Symbol Chooser). Typically
the user is in search field and after typing the search string
wants to use arrow down and up to browse results.
The problem this presented to accessibility was that VoiceOver reads
only changes of selection in the current VoiceOver item. As the user
is on the search field but the selection changes in the results table
below, VoiceOver did not read the various search results when pressing
arrow up/down.
Alternative was to leave the search field, move to the results table,
interact with it and then navigate it with VoiceOver. This is however
not the desired user experience comparable to that of sighted users,
as the VoiceOver user still has to do quite a few steps after entering
the search string to browse the results, not to speak about the situation
when the user would like to change the search string - he/she would need
to leave the table and get back to the search field.
This solves the problem by making the search field (or more generally
any user interface element that triggers change of selection in the
results table, which should be the element the VoiceOver cursor is on)
announce to accessibility the contents of the selected row in search
results table.
See this thread on accessibility-dev mailing list where the options for
implementing such a user interface in an accessible way are discussed:
http://lists.apple.com/archives/accessibility-dev/2013/Dec/msg00000.html
and
http://lists.apple.com/archives/accessibility-dev/2014/Feb/msg00016.html
If the user did not move selection (it is on the first line) and they
want to hear it, they should use arrow up to hear it. Then they can
use arrow down to move through results.
For example selecting a column of numbers and selecting “Add Numbers in Selection” will now always insert the result at a logical position rather than on the defined end points of the selection.
