While moving completion resolve to the event system in #9668 we introduced what
is essentially a "DOS attack" on slow LSPs. Completion resolve requests were
made in the render loop and debounced with a timeout. Once the timeout expired
the resolve request was made. The problem is the next frame would immediately
request a new completion resolve request (and mark the old one as obsolete but
because LSP has no notion of cancelation the server would still process it). So
we were in essence sending one completion request to the server every 150ms and
only stopped if the server managed to respond before we rendered a new frame.
This caused overload on slower machines/with slower LS.
In this PR I revamped the resolve handler so that a request is only ever
resolved once. Both by checking if a request is already in-flight and by marking
failed resolve requests as resolved.
Make the popup positions more consistent.
Improvements:
1. if the signature popup content is bigger than the available space,
then the popup is always shown under the cursor, even if there more
space above the cursor than below
2. There is no mutation anymore inside required_size. Maybe in the future
we can update all widgets to have no mutations and change the trait
Signed-off-by: Ben Fekih, Hichem <hichem.f@live.de>
Currently the editor mode has no effect on the behavior of `search` and
`rsearch`. We can pass in the right movement for the editor mode to make
the behavior or `search` and `rsearch` match `search_next` and
`search_prev` in select mode.
implement handle_event to cycle through the function signatures.
To change the signature press alt+p/n .
Signed-off-by: Ben Fekih, Hichem <hichem.f@live.de>
when the available height for the popup is low/small, then it is not
possible to scroll until the end
Signed-off-by: Ben Fekih, Hichem <hichem.f@live.de>
Previously unnecessary/deprecated diagnostic tags replaced the highlight
for the severity of a diagnostic. This could cause either the severity
or unnecessary/deprecated scopes to disappear when diagnostic ranges
overlapped though. Plus the severity highlight can be interesting in
addition to the unnecessary/deprecated highlight.
So this change separates the unnecessary and deprecated highlights from
the severity highlights, so each is merged separately and when they
overlap, the highlights are combined.
Prior to this change, every integration test which wanted its line
endings to be handled transparently across platforms, i.e. test with
the same input that has its platform's line feed characters, converting
the line endings was up to each individual test by calling the
`platform_line` helper function. This significantly increases the amount
of boilerplate one has to copy between all the tests.
However, there are some test cases that need to exert strict control
over the exact input text without being manipulated behind the scenes by
the test framework.
So, with this change, the line feed conversions are factored into
the `TestCase` struct. By default, line endings of the input text
are converted to the platform's native line feed ending, but one can
explicitly specify in their test case when the input text should be left
alone and tested as is.
This reverts commit 0dc67ff885.
See the post-merge discussion in #9828. The old behavior was less
surprising and we have other ways to abort from a prompt, so let's
revert the behavior change.
This uses the new TreeCursor type from the parent commit to reimplement
the tree-sitter motions (`A-p/o/i/n`). Other tree-sitter related
features like textobjects are not touched with this change and will
need a different, unrelated approach to solve.
The refactor in bcf7b263 introduced a possible subtraction with overflow
when the statusline is layed out so that the left or right sides are
larger than the padding it would take to align the center area to the
middle.
When the left or right areas are too large, we can evenly space the
elements rather than trying to align the center area to the middle.
This prevents possible underflows and makes sense visually - it's
still easy to tell the areas apart at a glance.
The `shell_impl` and `shell_impl_async` functions no longer return
`success` because it was always `true`. If the command didn't succeed
both functions would return an `Err`.
This was also the reason, why `shell_keep_pipe` didn't work. It relied
upon the value of `success` and aborted in case of an `Err`.
It now removes any selection for which `shell_impl` returns `Err`.
If the command always fails, the selections are preserved and an error
message is displayed in the status bar.
In the `reload-all` command, we should not stop reloading the documents
if one error is found. Instead, we should report the error and continue
trying to reload the current open documents. This is useful in cases
where a backing file does not exist temporarily (e.g. when editing a git
patch in the outstanding chain that doesn't have a file just yet).
This change also remove the error messages in the cases where the
backing is `None`, like in new docs or `tutor`.
