There are two code paths for Ctrl+Tab and for everything else:
Ctrl + Tab is working perfectly
* On the first tab navigation TerminalPage::_SelectNextTab resets the
tab commands, and since palette is not visible selects the relevant
tab index
* On the second navigation Ctrl+Tab is intercepted by
CommandPalette::_previewKeyDownHandler and everything works fine
But with custom binding things are screwed:
* On the first tab navigation TerminalPage::_SelectNextTab resets the
tab commands, and since palette is not visible selects the relevant
tab index
* On the second navigation keys are not intercepted and thus
TerminalPage::_SelectNextTab is called again. It resets the commands,
but now since the palette is visible we simply invoke
CommandPalette::SelectNextItem. Which in turn misbehaves because no
item is selected.
The approach for the solution is not to reset anything if the command
palette is already open.
* Manual testing of both custom and non-custom bindings with different
amount of tabs.
Closes#8247
(cherry picked from commit 0437fe9d8e)
Revert the tab switcher behavior for 1.4.
The default is once again `"useTabSwitcher":false`, since we
ninja-changed `true` to MRU switching. That was maybe a bad call.
This commit will teach CommandPalette to clamp the scroll page up and
scroll page down navigation so as to not wrap.
Closes#8189
(cherry picked from commit 624d07f283)
With this PR, the Terminal will check to make sure the "Touch, Keyboard and Handwriting Panel Service" is enabled at startup. If it isn't, then the Terminal won't be able to receive keyboard input (see #4448 and the 20 linked issues to that one).
* See #4448 for more details
* [x] Closes#7886
* [ ] Should this make #4448 not-open as well?
* [x] I work here
* [n/a] Tests added/passed
* [x] Docs: https://github.com/MicrosoftDocs/terminal/pull/168
I manually set the service to "Disabled", restarted the machine, verified the dialog opens (and that I'm unable to type in the Terminal), then re-set the service to automatic and rebooted, and the dialog doesn't appear.
(cherry picked from commit d5d2b7727f)
This pull request switches up the treatment we use for pattern-detected
links and OSC 8 hyperlinks:
* Links generated via OSC 8 have a sparse dotted underline instead of a
thick dashed one
* Links generated by pattern detection _are not underlined until they've
hovered_
* This papers over a visual glitch that is a result of us updating
the pattern matches every ~500ms (on change)
Closes#8123
(cherry picked from commit 26ca73b823)
<!-- Enter a brief description/summary of your PR here. What does it fix/what does it change/how was it tested (even manually, if necessary)? -->
## Summary of the Pull Request
In the focus mode the top border disappears upon resize. While this behavior is expected in the maximized / full screen mode, it should not happen in the focus mode.
<!-- Other than the issue solved, is this relevant to any other issues/existing PRs? -->
## References
<!-- Please review the items on the PR checklist before submitting-->
## PR Checklist
* [x] Closes https://github.com/microsoft/terminal/issues/7012
* [x] CLA signed
* [ ] Tests added/passed - nope, only manual testing
* [ ] Documentation updated - irrelevant
* [ ] Schema updated - irrelevant
* [ ] I've discussed this with core contributors already. If not checked, I'm ready to accept this work might be rejected in favor of a different grand plan. Issue number where discussion took place: #xxx
<!-- Provide a more detailed description of the PR, other things fixed or any additional comments/features here -->
## Detailed Description of the Pull Request / Additional comments
_GetTopBorderHeight method returns 0 when maximized or no title bar is visible. However the existence of top border has nothing to do with whether the title bar is visible. We want to leave the border as long as the window is not in some form of maximizing (maximized / full screen)
<!-- Describe how you validated the behavior. Add automated tests wherever possible, but list manual validation steps taken as well -->
## Validation Steps Performed
* Manual - dragging, resizing, maximizing both in focus and non focus modes + full screen testing
(cherry picked from commit 990628a78b)
<!-- Enter a brief description/summary of your PR here. What does it fix/what does it change/how was it tested (even manually, if necessary)? -->
## Summary of the Pull Request
<!-- Other than the issue solved, is this relevant to any other issues/existing PRs? -->
## References
<!-- Please review the items on the PR checklist before submitting-->
## PR Checklist
* [x] Closes https://github.com/microsoft/terminal/issues/7996
* [x] CLA signed.
* [ ] Documentation updated - irrelevant
* [ ] Schema updated - irrelevant
* [ ] I've discussed this with core contributors already. If not checked, I'm ready to accept this work might be rejected in favor of a different grand plan. Issue number where discussion took place: #xxx
<!-- Provide a more detailed description of the PR, other things fixed or any additional comments/features here -->
## Detailed Description of the Pull Request / Additional comments
Currently the value of AlwaysOnTop is read by the AppHost from AppLogic that takes this value from the root TerminalPage. However at this stage neither AppLogic nor TerminalPage are initialized, and thus the return value is always false.
This PR introduces a "GetInitialAlwaysOnTop" method to AppLogic that returns a value that is configured in the settings.
In addition, the TerminalPage creation was fixed to read the configuration value upon creation (and not just after settings reload).
<!-- Describe how you validated the behavior. Add automated tests wherever possible, but list manual validation steps taken as well -->
## Validation Steps Performed
* Only manual testing
* Starting the system with both initial value set to true and false
* Verifying that dynamic toggling on / off is not affected
(cherry picked from commit 5b2fd70940)
<!-- Enter a brief description/summary of your PR here. What does it fix/what does it change/how was it tested (even manually, if necessary)? -->
## Summary of the Pull Request
A second close command (middle click on taskbar preview) overrides the warning dialog and closes the application.
<!-- Other than the issue solved, is this relevant to any other issues/existing PRs? -->
## References
<!-- Please review the items on the PR checklist before submitting-->
## PR Checklist
* [x] Closes#7451
* [x] CLA signed. If not, go over [here](https://cla.opensource.microsoft.com/microsoft/Terminal) and sign the CLA
* [ ] Tests added/passed
* [ ] Documentation updated. If checked, please file a pull request on [our docs repo](https://github.com/MicrosoftDocs/terminal) and link it here: #xxx
* [ ] Schema updated.
* [ ] I've discussed this with core contributors already. If not checked, I'm ready to accept this work might be rejected in favor of a different grand plan. Issue number where discussion took place: #xxx
<!-- Provide a more detailed description of the PR, other things fixed or any additional comments/features here -->
## Detailed Description of the Pull Request / Additional comments
When a close command is invoked (middle click on taskbar preview or 'X' button), a new flag is set. When the user wants to close again (this time only via the taskbar preview, as the 'X' button is disabled), the application is closed. If the user cancels the dialog, the flag is reset to prevent accidental closing on a subsequent close command.
<!-- Describe how you validated the behavior. Add automated tests wherever possible, but list manual validation steps taken as well -->
## Validation Steps Performed
I am developing with a [Windows 10 virtual machine](https://developer.microsoft.com/en-us/windows/downloads/virtual-machines/) provided by Microsoft. I tested manually. I considered the 'X' button, middle click on taskbar preview, and Alt+F4. Only a middle click on the taskbar preview does override the dialog.
(cherry picked from commit d1e58bd71e)
Fire and forget on the hyperlink handler inside the TermControl.
## PR Checklist
* [x] Closes#7994
* [x] Tested manually
* [x] Hi, I work here.
## Detailed Description of the Pull Request / Additional comments
In `TermControl`, `_HyperlinkHandler` is called by
`_PointerPressedHandler` which has taken a write lock for all its
friends. However, `_HyperlinkHandler` downstreams to `ShellExecute`
which can pump the message queue looking for something. That pumping of
the queue can trigger messages that also want the write lock to update
state. They get stuck. Everything hangs.
`_HyperlinkHandler` really only needs read lock and really only for as
long as it takes to fill up its parameters before it's invoked... but
the simpler and more contained solution is to just fire and forget the
rest of the method that causes the deadlock to a continuation at the
tail of the dispatcher queue so `_PointerPressedHandler` can complete
and naturally drop the write lock.
## Validation Steps Performed
- Launched `main` manually on my box and clicked the hyperlink that is
detected when Powershell starts and it froze.
- Launched this change manually on my box and clicked the hyperlink that
is detected when Powershell starts and it did not freeze.
(cherry picked from commit 2ea4742f07)
<!-- Enter a brief description/summary of your PR here. What does it fix/what does it change/how was it tested (even manually, if necessary)? -->
## Summary of the Pull Request
Realized that we don't copy the current hyperlink id when we copy buffers, quick fix for that
<!-- Please review the items on the PR checklist before submitting-->
## PR Checklist
* [ ] Closes #xxx
* [x] CLA signed. If not, go over [here](https://cla.opensource.microsoft.com/microsoft/Terminal) and sign the CLA
* [ ] Tests added/passed
* [ ] Documentation updated. If checked, please file a pull request on [our docs repo](https://github.com/MicrosoftDocs/terminal) and link it here: #xxx
* [ ] Schema updated.
* [x] I work here
(cherry picked from commit ce4fd2970a)
While not explicitly permitted, a wide range of software (including
Windows' own touch keyboard) sets the `wScan` member of the `KEYBDINPUT`
structure to 0, resulting in `scanCode` being 0 as well. In these
situations we'll now use the `vkey` to get a `scanCode`.
Validation
----------
* AutoHotkey
* Use a keyboard layout with `AltGr` key
* Execute the following script:
```ahk
#NoEnv
#Warn
SendMode Input
SetWorkingDir %A_ScriptDir%
<^>!8::SendInput {Raw}»
```
* Press `AltGr+8` while the Terminal is in the foreground
* Ensure » is being echoed ✔️
* PowerToys
* Add a `Ctrl+I -> ↑ (up arrow)` keyboard shortcut
* Press `Ctrl+I` while the Terminal is in the foreground
* Ensure the shell history is being navigated backwards ✔️
* Windows Touch Keyboard
* Right-click or tap and hold the taskbar and select "Show touch
keyboard" button
* Open touch keyboard
* Ensure keyboard works like a regular keyboard ✔️
* Ensure unicode characters are echoed on the Terminal as well (except
for Emojis) ✔️Closes#7438Closes#7495Closes#7843
(cherry picked from commit d51d8dc768)
Close the tab context menu when clicking on the title bar
## Detailed Description of the Pull Request / Additional comments
Following #2438, hide the tabs context menu on `TerminalPage::TitlebarClicked()`.
We don't know which of the tabs is showing the context menu, do it on all tabs.
## Validation Steps Performed
Open the context menu from any tab, click on title bar and see the context menu disappear.
Closes#7988
(cherry picked from commit 7e8600147e)
We are getting some watson crash reports that the terminal is attempting
to resize to `(0, 0)`. This change makes it so that we prevent such
resizing and if so, throw an exception before we reach native code.
This commit adds resizing checks that prevent resizing the terminal WPF
control to a size of `(0, 0)`
(cherry picked from commit 5a518e5e58)
This PR resolves an issue I observed in
Microsoft.Terminal.Wpf.TerminalControl.CalculateMargins(). Specifically,
on line 194 in the project. In this example, the line: `height =
controlSize.Height - (this.TerminalRendererSize.Height /
dpiScale.DpiScaleX);` is associating the height margin with
dpiScale.DpiScaleX instead of dpiScale.DpiScaleY. This PR changes the
association to DpiScaleY.
Closes#8038
(cherry picked from commit c095a678a5)
This PR changes the ATS display order to _always_ be in most recently
used (MRU) order. I chose not to give ATS the option to be displayed
in-order because that order is better served through the traditional
left-right TabRow switching.
_Note_: `TabSearch` will stay in-order.
This means that users can only choose one order or another in their
`nextTab/prevTab` bindings. Setting `useTabSwitcher` to true will make
nT/pT open the ATS in MRU order. If it's set to false, the ATS won't
open and nT/pT will simply go left and right on the TabRow.
I'm open to getting rid of the global and making ATS its own keybinding,
but for now I figured I would keep the current behavior and open the PR
to get eyes on the code that doesn't have anything to do with the
settings.
Closes#973
(cherry picked from commit 00f5fbaf3d)
Currently, `CommandPalette` creates and maintains the `SwitchToTab`
commands used for the ATS. When `Command` goes into the
TerminalSettingsModel, the palette won't be able to access `Command`'s
implementation type, making it difficult for `CommandPalette` to tell
`Command` to listen to `Tab` for changes.
This PR changes the relationship up so `Tab` now manages its
`SwitchToTab` command, and `CommandPalette` just plops the command from
`Tab` into its list.
(cherry picked from commit 468c8c6728)
<!-- Enter a brief description/summary of your PR here. What does it fix/what does it change/how was it tested (even manually, if necessary)? -->
Adds the color slider to the tab color picker
<!-- Please review the items on the PR checklist before submitting-->
## PR Checklist
* [x] Closes#7948
* [x] CLA signed. If not, go over [here](https://cla.opensource.microsoft.com/microsoft/Terminal) and sign the CLA
* [ ] ~Tests added/passed~
* [ ] ~Documentation updated. If checked, please file a pull request on [our docs repo](https://github.com/MicrosoftDocs/terminal) and link it here: #xxx~
* [ ] ~Schema updated.~
* [x] I've discussed this with core contributors already. If not checked, I'm ready to accept this work might be rejected in favor of a different grand plan. Issue number where discussion took place: #7948
<!-- Provide a more detailed description of the PR, other things fixed or any additional comments/features here -->
<!-- Describe how you validated the behavior. Add automated tests wherever possible, but list manual validation steps taken as well -->
## Validation Steps Performed
*Not required*
(cherry picked from commit 4a95d94c55)
We wrap the call to `_WriteSettings` in
`CascadiaSettingsSerialization.cpp` in a try/catch block, and if we
catch an error we append a warning telling the user to check the
permissions on their settings file.
Closes#7727
(cherry picked from commit 16b8ea14d6)
Fixes the bug where `exit`ing inside a closed pane would leave the Terminal blank.
Additionally, removes `Tab::GetRootElement` and replaces it with the _observable_ `Tab::Content`. This should be more resilient in the future.
Also adds some tests, though admittedly not for this exact scenario. This scenario requires a cooperating TerminalConnection that I can drive for the sake of testing, and _ain't nobody got time for that_.
* Introduced in #6989
* [x] Closes#7252
* [x] I work here
* [x] Tests added/passed 🎉
* [n/a] Requires documentation to be updated
From notes I had left in `Tab.cpp` while I was working on this:
```
OKAY I see what's happening here the ActivePaneChanged Handler in TerminalPage
doesn't re-attach the tab content to the tree, it just updates the title of the
window.
So when the pane is `exit`ed, the pane's control is removed and re-attached to
the parent grid, which _isn't in the XAML tree_. And no one can go tell the
TerminalPage that it needs to re set up the tab content again.
The Page _manually_ does this in a few places, when various pane actions are
about to take place, it'll unzoom. It would be way easier if the Tab could just
manage the content of the page.
Or if the Tab just had a Content that was observable, that when that changed,
the page would auto readjust. That does sound like a LOT of work though.
```
Opened panes, closed panes, exited panes, zoomed panes, moved focus between panes, panes, panes, panes
(cherry picked from commit ccf9f03ed3)
## Summary of the Pull Request
Just deleting an unnecessary call to `_UpdateCommandsForPalette`
**Note:** This only fixes slowdown when opening/closing a tab, but not upon first startup (we still need to call `_UpdateCommandsForPalette` there
## References
Fixes the slowdown described in #7820 for opening and closing tabs, but doesn't improve startup time dramatically.
## Validation Steps Performed
Tested with ~100 profiles in my settings file
(cherry picked from commit 895ac06dbd)
Let's assume the user has bound the dead key ^ to a sendInput command
that sends "b". If the user presses the two keys ^a it'll produce "bâ",
despite us marking the key event as handled. We can use `ToUnicodeEx`
to clear such dead keys from the keyboard state and should make use of
that for keybindings. Unfortunately `SetKeyboardState` cannot be used
for this purpose as it doesn't clear the dead key state.
Validation
* Enabled a German keyboard layout
* Added the following two keybindings:
{ "command": { "action": "sendInput", "input": "x" }, "keys": "q" },
{ "command": { "action": "sendInput", "input": "b" }, "keys": "^" }
* Pressed the following keys → ensured that the given text is printed:
* q → x
* ´ → nothing
* a → á
* ^ → b
* a → a (previously this would print: â)
* ´ → nothing
* ^ → b
* a → a (unfortunately we cannot specifically clear only ^)
Closes#5784
(cherry picked from commit 4099aacacb)
It turns out that we missed part of the OSC 8 spec which indicated that
_hyperlinks with the same ID but different URIs are logically distinct._
> Character cells that have the same target URI and the same nonempty id
> are always underlined together on mouseover.
> The same id is only used for connecting character cells whose URIs is
> also the same. Character cells pointing to different URIs should never
> be underlined together when hovering over.
This pull request fixes that oversight by appending the (hashed) URI to
the generated ID.
When Terminal receives one of these links over ConPTY, it will hash the
URL a second time and therefore append a second hashed ID. This is taken
as an acceptable cost.
Fixes#7698
(cherry picked from commit df7c3ccc3b)
This commit adds a missing conversion utf8 to utf16 in decoding base64
for handling multibyte text in copying via OSC 52.
## Validation Steps Performed
* automatically
* Tests w/ multibyte characters
* manually
* case1
* Executed `printf "\x1b]52;;%s\x1b\\" "$(printf '👍👍🏻👍🏼👍🏽👍🏾👍🏿' | base64)"`
* Verified `👍👍🏻👍🏼👍🏽👍🏾👍🏿` in my clipboard
* case2
* Copied `👍👍🏻👍🏼👍🏽👍🏾👍🏿` by tmux 2.6 default copy function (OSC 52)
* Verified `👍👍🏻👍🏼👍🏽👍🏾👍🏿` in my clipboard
Closes#7819
(cherry picked from commit 743283e434)
This PR adds support for the `DECREQTPARM` (Request Terminal Parameters)
escape sequence, which was originally used on the VT100 terminal to
report the serial communication parameters. Modern terminal emulators
simply hardcode the reported values for backward compatibility.
The `DECREQTPARM` sequence has one parameter, which was originally used
to tell the terminal whether it was permitted to send unsolicited
reports or not. However, since we have no reason to send an unsolicited
report, we don't need to keep track of that state, but the permission
parameter does still determine the value of the first parameter in the
response.
The response parameters are as follows:
| Parameter | Value | Meaning |
| ---------------- | ------ | ------------------------ |
| response type | 2 or 3 | unsolicited or solicited |
| parity | 1 | no parity |
| data bits | 1 | 8 bits per character |
| transmit speed | 128 | 38400 baud |
| receive speed | 128 | 38400 baud |
| clock multiplier | 1 | |
| flags | 0 | |
There is some variation in the baud rate reported by modern terminal
emulators, and 9600 baud seems to be a little more common than 38400
baud, but I thought the higher speed was probably more appropriate,
especially since that's also the value reported by XTerm.
## Validation Steps Performed
I've added a couple of adapter and output engine tests to verify that
the sequence is dispatched correctly, and the expected responses are
generated. I've also manually tested in Vttest and confirmed that we now
pass the `DECREQTPARM` test in the _Test of terminal reports_.
Closes#7852
(cherry picked from commit 30e363e7ac)
Took this as an easy starter. The method IslandWindow::SetAlwaysOnTop is
triggered once terminal settings are reloaded (in
TerminalPage::_RefreshUIForSettingsReload flow). This method calls
SetWindowPos without SWP_NOACTIVATE. As a result the window gets
activated, the focus is set and the cursor starts blinking.
Added SWP_NOACTIVATE in all SetWindowPos calls from IslandWindow and
NoClientIslandWindow (where it was missing). Please let me know if this
is an overkill - it is not required to fix the issue, however seems a
good practice, that might help if we decide to apply more settings
immediately.
## Validation Steps Performed
* Only manual testing - please guide me to the relevant UT framework, if
exists.
* Trying to reproduce this with VS attached doesn't work - the window
gets the focus in any case.
* Tested as a standalone application, by modifying different settings
(and comparing the results before and after the fix).
* Checked with Spy++ that no WM_ACTIVATE / WM_SETFOCUS is thrown upon
settings modification
* Applied terminal resizing, toggling full screen and focus mode to
check no regression was introduced.
Closes#7571
(cherry picked from commit cb732a4bcc)
Adds a tooltip to the new tab button and menu to let the user know
that holding alt will open a new pane instead.
Fixes#7851
Co-authored-by: Pankaj Bhojwani <pabhojwa@microsoft.com>
(cherry picked from commit 8d12388915)
This pull request introduces (a very, very stripped-down copy of) the
WIL fallback error reporter.
It emits error records, usually immediately before the application
implodes, into the event stream.
This should improve diagnosability of issues that take Terminal down,
and allow us to give out a .wprp file to gather traces from users.
(cherry picked from commit cd768934be)
`ScrollIntoView` is responsible for scrolling the viewport to include
the UTR's start endpoint. The crash was caused by `start` being at the
exclusive end, and attempting to scroll to it. This is now fixed by
clamping the result to the bottom of the buffer.
Most of the work here is to allow a test for this. `ScrollIntoView`
relied on a virtual `ChangeViewport` function. By making that
non-virtual, the `DummyElementProvider` in the tests can now be a
`ScreenInfoUiaProviderBase`. This opens up the possibility of more
UiaTextRange tests in the future too.
Closes#7839
(cherry picked from commit 7a1932c556)
This is not going to be our plan of record for Universal going forward.
This updates the Universal configuration to 1) match non-universal and 2) switch to local applications
(cherry picked from commit d33ca7e8eb)
## Summary of the Pull Request
This PR enables the ATS to display the active tab as the user navigates the tab switcher. We do this by dispatching the tab switch actions as the user navigates the menu, and manually _not_ focusing the new tab when the tab switcher is open.
## References
* #6732 - original tab switcher PR
* #6689 - That's a more involved, generic version of this, but this PR will be enough to stop most of the complaints hopefully
## PR Checklist
* [x] Closes#7409
* [x] I work here
* [ ] Tests added/passed
* [n/a] Requires documentation to be updated
## Validation Steps Performed
Opened tabs, tabbed through the menu, verified that it did what I'd expect
(cherry picked from commit 22887d721f)
The `MovementAtExclusiveEnd` test was improperly authored for the
following reasons:
- it should have used `TEST_METHOD_PROPERTY` to cover all of the
TextUnits
- TextUnit::Document (arguably one of the most important) was ommitted
accidentally (`!= TextUnit_Document` was used instead of `<=`)
- The created range was not `EndExclusive`, but rather, the last cell in
the buffer (`EndInclusive`)
The first half of this PR fixes the test.
The second half of this PR expands the test and fixes any related issues
to make the test pass (i.e. #7771):
- `TEST_METHOD_PROPERTY` was added for it to be degenerate (start/end at
`EndExclusive`) or not (last cell of buffer)
- `utr->_start` is now also validated after moving backwards
NOTE: `utr->_start` was not validated when moving forwards because
moving forwards should always fail when at/past the last chell in the
buffer.
Closes#7771
(cherry picked from commit e401edf9ef)
This performs a minor refactor on `TextBuffer::MoveToNextWord` that
relies more heavily on `TextBuffer::GetWordEnd`. Now, the logic is
simplified and looks more like `MoveToPreviousWord`.
This refactor required me to move the `lastCharPos` optimization down to
`GetWordEnd`. So word expansion gets this optimization for free now.
### WPR Traces
The percentages below represent the weight that a function call had. The
test scenario included moving by word on the CMD welcome message until
the last word was reached. Inspect.exe was used to limit any additional
calls that are generally performed by a screen reader.
| function | current | branch |
| -- | -- | -- |
| `UIA:Move` | 34.55% | 29.52% |
There is an improvement of about 5% in a release build of ConHost.
NOTE: `UIA::Move` already calls `Expand` after a move operation is
performed. I'm using this data to represent a performance improvement
across both functions.
Contributes to #5243
(cherry picked from commit 386ae04edf)
This fixes a bug when moving backwards by word that resulted in #7742.
This also includes...
- a minor refactor that leverages `GetWordStart` in `MoveToPreviousWord`
- additional unit tests for movement by word
- a feature test comprised of the referenced bug report
`MoveToPreviousWord()` would...
- move backwards for each whitespace character
- then, move backwards for each regular character
This would actually result in moving to the beginning of the current "word" (as defined by a11y).
We actually need to do this process twice:
- the first time gets you to the beginning of the current word
- attempt to move back by one character
- the second time gets you to the beginning of the previous word
Rather than implementing 4 while loops, we leverage `GetWordStart()` to
attempt to move to the beginning of the previous word. We call it twice
(as described above). The logic is unchanged, but we instead reuse a
function that has already undergone more testing.
To make sure this works as expected, additional unit tests were
introduced covering "MoveByWord" in the TextBuffer.
## Validation Steps Performed
Added test for repro steps.
Added unit tests for movement by word.
Closes#7742
(cherry picked from commit 9ec57a7d3c)
DestListLogoUri cannot handle paths that are separated with / unless
they're actually URLs. We have to guess somewhat whether something is a
file path and if it appears to be one, normalize it.
Fixes#7706
(cherry picked from commit f28ec65843)
