This removes a handful of allocations for functions calling into the
function, which is nice because the prompt may call this function on
every keypress.
Pascal and I discussed this and we think it's generally better to
take a 'RopeSlice' rather than a '&Rope'. The code block rendering
function in the markdown component module is a good example for how
this can be useful: we can remove an allocation of a rope and instead
directly turn a '&str' into a 'RopeSlice' which is very cheap.
A change to prefer 'RopeSlice' to '&Rope' whenever the rope isn't
modified would be nice, but it would be a very large diff (around 500+
500-). Starting off with just the syntax functions seems like a nice
middle-ground, and we can remove a Rope allocation because of it.
Co-authored-by: Pascal Kuthe <pascal.kuthe@semimod.de>
Since regex is almost always injected into other languages,
`pattern_character`s will inherit the highlight for the structure that
injects them (for example `/foo/` in JavaScript or `~r/foo/` in Elixir).
This removes the string highlight when used in the prompt.
We also add `ERROR` node highlighting so that errors in regex syntax
appear in the prompt. This resolves a TODO in the `regex_prompt`
function about highlighting errors in the regex.
We can use tree-sitter-regex highlighting in prompts for entering
regexes, like `search` or `global_search`. The `highlighted_code_block`
function from the markdown component makes this a very small change.
This could be improved in the future by leaving the parsed syntax tree
on the prompt, allowing incremental updates. Prompt lines are usually so
short though and tree-sitter-regex is rather small and uncomplicated,
so that improvement probably wouldn't make a big difference.
* Add initial support for LSP DidChangeWatchedFiles
* Move file event Handler to helix-lsp
* Simplify file event handling
* Refactor file event handling
* Block on future within LSP file event handler
* Fully qualify uses of the file_event::Handler type
* Rename ops field to options
* Revert newline removal from helix-view/Cargo.toml
* Ensure file event Handler is cleaned up when lsp client is shutdown
* _apply_motion generalization where possible
API encourages users to not forget setting `editor.last_motion` when
applying a motion. But also not setting `last_motion` without applying a
motion first.
* (rename) will_find_char -> find_char
method name makes it sound like it would be returning a boolean.
* use _apply_motion in find_char
Feature that falls out from this is that repetitions of t,T,f,F are
saved with the context extention/move and count. (Not defaulting to extend
by 1 count).
* Finalize apply_motion API
last_motion is now a private field and can only be set by calling
Editor.apply_motion(). Removing need (and possibility) of writing:
`motion(editor); editor.last_motion = motion`
Now it's just: `editor.apply_motion(motion)`
* editor.last_message: rm Box wrap around Arc
* Use pre-existing `Direction` rather than custom `SearchDirection`.
* `LastMotion` type alias for `Option<Arc<dyn Fn(&mut Editor)>>`
* Take motion rather than cloning it.
Co-authored-by: Michael Davis <mcarsondavis@gmail.com>
* last_motion as Option<Motion>.
* Use `Box` over `Arc` for `last_motion`.
---------
Co-authored-by: Michael Davis <mcarsondavis@gmail.com>
The spec explicitly disallows publishDiagnostic to be sent before
the initialize response:
> ... the server is not allowed to send any requests or notifications to
> the client until it has responded with an InitializeResult ...
(https://microsoft.github.io/language-server-protocol/specifications/lsp/3.17/specification/#initialize)
But if a non-compliant server sends this we currently panic because we
'.expect()' the server capabilities to be known to fetch the position
encoding. Instead of panicking we can discard the notification and log
the non-compliant behavior.
Merges the code for the Picker and FilePicker into a single Picker that
can show a file preview if a preview callback is provided. This change
was mainly made to facilitate refactoring out a simple skeleton of a
picker that does not do any filtering to be reused in a normal Picker
and a DynamicPicker (see #5714; in particular [mikes-comment] and
[gokuls-comment]).
The crux of the issue is that a picker maintains a list of predefined
options (eg. list of files in the directory) and (re-)filters them every
time the picker prompt changes, while a dynamic picker (eg. interactive
global search, #4687) recalculates the full list of options on every
prompt change. Using a filtering picker to drive a dynamic picker hence
does duplicate work of filtering thousands of matches for no reason. It
could also cause problems like interfering with the regex pattern in the
global search.
I tried to directly extract a PickerBase to be reused in Picker and
FilePicker and DynamicPicker, but the problem is that DynamicPicker is
actually a DynamicFilePicker (i.e. it can preview file contents) which
means we would need PickerBase, Picker, FilePicker, DynamicPicker and
DynamicFilePicker and then another way of sharing the previewing code
between a FilePicker and a DynamicFilePicker. By merging Picker and
FilePicker into Picker, we only need PickerBase, Picker and
DynamicPicker.
[gokuls-comment]: https://github.com/helix-editor/helix/issues/5714#issuecomment-1410949578
[mikes-comment]: https://github.com/helix-editor/helix/issues/5714#issuecomment-1407451963
Resolves issue #6888 by adding a command to join all selections and yank
them to the specified register. The typed command takes an argument as
the separator to use when joining the selections.
Previously a count or register selection would be lost while opening
the command palette. This change allows using a register selection or
count in any command chosen from the command palette.
Previously the register selection (via `"`) would be lost in the middle
of any key sequence longer than one key. For example, `<space>f` would
clear the register selection after the `<space>` making it inaccessible
for the `file_picker` command.
This behavior does not currently have any effect in the default keymap
but might affect custom keymaps. This change aligns the behavior of the
register with count. Making this change allows propagating the register
to the `command_palette` (see the child commit) or other pickers should
we decide to use registers in those in the future. (Interactive global
search for example.)
Does not change any behavior other than making the tuple slightly
more idiomatic. Keymap infobox shows key events, then the respective
description. This commit makes sure that order is used from the get go,
rather than flipping it midway.
* chore: avoid format! call with argument when useless
* feat: also clear diagnostics for unopened documents when exiting an LSP
* feat: we already worked on `self.editor.diagnostics` no need to redo the checks
* Add command for merging non-consecutive ranges
* Add `merge_selections` command to book
* Simplify `merge_ranges`
Heeded the advice of @the-mikedavis to stop iterating over all ranges and simply merge the first and the last range, as the invariants of `Selection` guarantee that the list of ranges is always sorted and never empty.
* Clarify doc comment of `merge_ranges`
* Add `helix_lsp::client::Client::supports_feature(&self, LanguageServerFeature)`
* Extend `doc.language_servers_with_feature` to use this method as filter as well
* Add macro `language_server_with_feature!` to reduce boilerplate for non-mergeable language server requests (like goto-definition)
* Refactored most of the `find_map` code to use the either the macro or filter directly via `doc.language_servers_with_feature`
Language Servers are now configured in a separate table in `languages.toml`:
```toml
[langauge-server.mylang-lsp]
command = "mylang-lsp"
args = ["--stdio"]
config = { provideFormatter = true }
[language-server.efm-lsp-prettier]
command = "efm-langserver"
[language-server.efm-lsp-prettier.config]
documentFormatting = true
languages = { typescript = [ { formatCommand ="prettier --stdin-filepath ${INPUT}", formatStdin = true } ] }
```
The language server for a language is configured like this (`typescript-language-server` is configured by default):
```toml
[[language]]
name = "typescript"
language-servers = [ { name = "efm-lsp-prettier", only-features = [ "format" ] }, "typescript-language-server" ]
```
or equivalent:
```toml
[[language]]
name = "typescript"
language-servers = [ { name = "typescript-language-server", except-features = [ "format" ] }, "efm-lsp-prettier" ]
```
Each requested LSP feature is priorized in the order of the `language-servers` array.
For example the first `goto-definition` supported language server (in this case `typescript-language-server`) will be taken for the relevant LSP request (command `goto_definition`).
If no `except-features` or `only-features` is given all features for the language server are enabled, as long as the language server supports these. If it doesn't the next language server which supports the feature is tried.
The list of supported features are:
- `format`
- `goto-definition`
- `goto-declaration`
- `goto-type-definition`
- `goto-reference`
- `goto-implementation`
- `signature-help`
- `hover`
- `document-highlight`
- `completion`
- `code-action`
- `workspace-command`
- `document-symbols`
- `workspace-symbols`
- `diagnostics`
- `rename-symbol`
- `inlay-hints`
Another side-effect/difference that comes with this PR, is that only one language server instance is started if different languages use the same language server.
There was an issue with autocompletion of a path with a space in it.
Before:
:o test\ dir -> <TAB> -> test\ dirfile1
After:
:o test\ dir -> <TAB> -> test\ dir\file1
Currently, when forward deleting (`delete_char_forward` bound to `del`,
`delete_word_forward`, `kill_to_line_end`) the cursor is moved to the
left in append mode (or generally when the cursor is at the end of the
selection). For example in a document `|abc|def` (|indicates selection)
if enter append mode the cursor is moved to `c` and the selection
becomes: `|abcd|ef`. When deleting forward (`del`) `d` is deleted. The
expectation would be that the selection doesn't shrink so that `del`
again deletes `e` and then `f`. This would look as follows:
`|abcd|ef`
`|abce|f`
`|abcf|`
`|abc |`
This is inline with how other editors like kakoune work.
However, helix currently moves the selection backwards leading to the
following behavior:
`|abcd|ef`
`|abc|ef`
`|ab|ef`
`ef`
This means that `delete_char_forward` essentially acts like
`delete_char_backward` after deleting the first character in append
mode.
To fix the problem the cursor must be moved to the right while deleting
forward (first fix in this commit). Furthermore, when the EOF char is
reached a newline char must be inserted (just like when entering
appendmode) to prevent the cursor from moving to the right
Some deletion operations (especially those that use indentation)
can generate overlapping deletion ranges when using multiple cursors.
To fix that problem a new `Transaction::delete` and
`Transaction:delete_by_selection` function were added. These functions
merge overlapping deletion ranges instead of generating an invalid
transaction. This merging of changes is only possible for deletions
and not for other changes and therefore require its own function.
The function has been used in all commands that currently delete
text by using `Transaction::change_by_selection`.
When re requesting a completion that already has a selected item we
reuse that selections savepoint. However, the selection has likely
changed since that savepoint which requires us to use the selection
from that savepoint
* inject language based on file extension
Nodes can now be captured with "injection.filename". If this capture
contains a valid file extension known to Helix, then the content will
be highlighted as that language.
* inject language by shebang
Nodes can now be captured with "injection.shebang". If this capture
contains a valid shebang line known to Helix, then the content will
be highlighted as the language the shebang calls for.
* add documentation for language injection
* nix: fix highlights
The `@` is now highlighted properly on either side of the function arg.
Also, extending the phases with `buildPhase = prev.buildPhase + ''''`
is now highlighted properly.
Fix highlighting of `''$` style escapes (requires tree-sitter-nix bump)
Fix `inherit` highlighting.
* simplify injection_for_match
Split out injection pair logic into its own method to make the overall
flow easier to follow.
Also transform the top-level function into a method on a
HighlightConfiguration.
* markdown: add shebang injection query
This picks up changes to the `editor.mouse` option at runtime - either
through `:set-option` or `:config-reload`. When the value changes, we
tell the terminal to enable or disable mouse capture sequences.
* Fix crash on opening from suspend state (#6725)
* Fix code style
* revert using of the imperative code style. Add panic if couldn't set terminal raw mode
* remove redundant import of core::panic macros
* small refactoring
* Fix#6669: Theme preview doesn't return theme to normal when delete name with Alt-Backspace
* Fix#6669: Return theme preview to normal theme for all remaining keybinds that change the promt text
The current implementation didn't reload the theme if no no theme was
explicitly configured (so the default theme was used). This commit
brings `refresh_theme` in line with the initialization code.
Add new theme highlight keys, for setting the colour of the breakpoint
character and the current line at which execution has been paused at.
The two new keys are `ui.highlight.frameline` and `ui.debug.breakpoint`.
Highlight according to those keys, both the line at which debugging
is paused at and the breakpoint indicator.
Add an indicator for the current line at which execution is paused
at, themed by the `ui.debug.active` theme scope. Update various themes
to showcase how the new functionality works.
Better icons are dependent on #2869, and as such will be handled in the
future, once it lands.
Closes: #5952
Signed-off-by: Filip Dutescu <filip.dutescu@gmail.com>
While scrolling (with the `scroll`) command scrolloff was calculated
slightly differently than in `ensure_cursor_in_view` which could cause
the cursor to get stuck while scrolling
Virtual text lines (either caused by softwrapped inlay hints that take
multiple or line annotations) currently block scrolling downwards.
if the visual offset passed to char_idx_at_visual_offset or
visual_offset_from_block is within a virtual text line then the char
position before the virtual text and a visual offset are returned.
We previously ignored that visual offset and as a result the cursor
would be stuck at the start of the virtual text. This commit fixes
that by simply moving the cursor to the next char (so past the virtual
text) if this visual offset is non-zero
The current test DSL currently has no way to express being at the end of
a line, save for putting an explicit LF or CRLF inside the `#[|]#`. The
problem with this approach is that it can add unintended extra new lines
if used in conjunction with raw strings, which insert newlines for you.
This is a simple attempt to mitigate this problem. If there is an
explicit newline character at the end of the selection, and then it
is immediately followed by the same newline character at the right end
of the selection, this following newline is removed. This way, one can
express a cursor at the end of a line explicitly.
* helix-term: send the STOP signal to all processes in the process group
From kill(3p):
If pid is 0, sig shall be sent to all processes (excluding an unspecified set
of system processes) whose process group ID is equal to the process group ID
of the sender, and for which the process has permission to send a signal.
This fixes the issue of running `git commit`, attempting to suspend
helix with ^Z, and then not regaining control over the terminal and
having to press ^Z again.
* helix-term: use libc directly to send STOP signal
* helix-term: document safety of libc::kill
* helix-term: properly handle libc::kill's failure
I misread the manpage for POSIX `kill` -- it returns `-1` in
the failure case, and sets `errno`, which is retrieved via
`std::io::Error::last_os_error()`, has its string representation printed
out, and then exits with the matching status code (or 1 if, for whatever
reason, there is no matching status code).
* helix-term: expand upon why we need to SIGSTOP the entire process group
Also add a link back to one of the upstream issues.
* misc: missing inline, outdated link
* doc: Add new theme keys and config option to book
* fix: don't panic in Tree::try_get(view_id)
Necessary for later, where we could be receiving an LSP response
for a closed window, in which case we don't want to crash while
checking for its existence
* fix: reset idle timer on all mouse events
* refacto: Introduce Overlay::new and InlineAnnotation::new
* refacto: extract make_job_callback from Context::callback
* feat: add LSP display_inlay_hint option to config
* feat: communicate inlay hints support capabilities of helix to LSP server
* feat: Add function to request range of inlay hint from LSP
* feat: Save inlay hints in document, per view
* feat: Update inlay hints on document changes
* feat: Compute inlay hints on idle timeout
* nit: Add todo's about inlay hints for later
* fix: compute text annotations for current view in view.rs, not document.rs
* doc: Improve Document::text_annotations() description
* nit: getters don't use 'get_' in front
* fix: Drop inlay hints annotations on config refresh if necessary
* fix: padding theming for LSP inlay hints
* fix: tracking of outdated inlay hints should not be dependant on document revision (because of undos and such)
* fix: follow LSP spec and don't highlight padding as virtual text
* config: add some LSP inlay hint configs
Multicursor completions may overlap and therefore overlapping
completions must be dropped to avoid crashes. Furthermore, multicursor
edits might simply be out of range if the word before/after the cursor
is shorter. This currently leads to crashes, instead these selections
are now also removed for completions.
This commit also significantly refactors snippet transaction generation
so that tabstops behave correctly with the above rules. Furthermore,
snippet tabstops need to be carefully mapped to ensure their position
is correct and consistent with our selection semantics. Finally,
we now keep a partially updated Rope while creating snippet
transactions so that we can fill information into snippets that
depends on the position in the document.
Most LSPs will complete case-insensitive matches, particularly from
lowercase to uppercase. In some cases, notably Pyright, this is given
as a simple insert text instead of TextEdit. When this happens, the
prefix text was left unedited.
* Generalised to multiple runtime directories with priorities
This is an implementation for #3346.
Previously, one of the following runtime directories were used:
1. `$HELIX_RUNTIME`
2. sibling directory to `$CARGO_MANIFEST_DIR`
3. subdirectory of user config directory
4. subdirectory of path to helix executable
The first directory provided / found to exist in this order was used as a
root for all runtime file searches (grammars, themes, queries).
This change lowers the priority of `$HELIX_RUNTIME` so that the user
config runtime has higher priority. More significantly, all of these
directories are now searched for runtime files, enabling a user to override
default or system-level runtime files. If the same file name appears
in multiple runtime directories, the following priority is now used:
1. sibling directory to `$CARGO_MANIFEST_DIR`
2. subdirectory of user config directory
3. `$HELIX_RUNTIME`
4. subdirectory of path to helix executable
One exception to this rule is that a user can have a `themes`
directory directly in the user config directory that has higher piority
to `themes` directories in runtime directories. That behaviour has been
preserved.
As part of implementing this feature `theme::Loader` was simplified
and the cycle detection logic of the theme inheritance was improved to
cover more cases and to be more explicit.
* Removed AsRef usage to avoid binary growth
* Health displaying ;-separated runtime dirs
* Changed HELIX_RUNTIME build from src instructions
* Updated doc for more detail on runtime directories
* Improved health symlink printing and theme cycle errors
The health display of runtime symlinks now prints both ends of the
link.
Separate errors are given when theme file is not found and when the
only theme file found would form an inheritence cycle.
* Satisfied clippy on passing Path
* Clarified highest priority runtime directory purpose
* Further clarified multiple runtime details in book
Also gave markdown headings to subsections.
Fixed a error with table indentation not building
table that also appears present on master.
---------
Co-authored-by: Paul Scott <paul.scott@anu.edu.au>
Co-authored-by: Blaž Hrastnik <blaz@mxxn.io>
* Fix#6092
Cause were some incorrect assumptions that missed an edge case in the
`Selection.contains()` calculation. Tests were added accordingly.
* Fix Selection.contains() edge-case handling.
Removing the len check short-circuit was the only thing needed as
pointed out by @dead10ck.
Repeating completions currently crates a savepoint when a completion
popup was triggered (so after the request completed). Just like for
normal completions the savepoint must be created at the request.
The occurrence of the completion request was previously not saved in
`last_insert`. To that end a new `InsertEvent::RequestCompletion`
variant has been added. When replayed this event creates a snapshot
that is "actived" by the `TriggerCompletion` event and subsequently
used during any `InsertEvent::CompletiuonApply` events.
Completion requests are computed asynchronously to avoid common micro
freezes while editing. This means that once a completion request
completes, the state of the editor might have changed. Currently,
there is a check to ensure we are still in insert mode. However,
we also need to ensure that the view and document hasn't changed
to avoid accidentally using a savepoint with the wrong view/document.
Furthermore, the editor might request a new completion while the
previous completion request hasn't complemented yet. This can
lead to weird flickering or an outdated completion request replacing
a newer completion that has already completed (the LSP server
is not required to process completion requests in order). This change
also needed to ensure determinism/linear ordering so that completion
popup always correspond to the last completion request.
Fixing autocomplete required moving the document savepoint before the
asynchronous completion request. However, this in turn causes new bugs:
If the completion popup is open, the savepoint is restored when the
popup closes (or another entry is selected). However, at that point
a new completion request might already have been created which
would have replaced the new savepoint (therefore leading to incorrectly
applied complies).
This commit fixes that bug by allowing in arbitrary number of
savepoints to be tracked on the document. The savepoints are reference
counted and therefore remain valid as long as any reference to them
remains. Weak reference are stored on the document and any reference
that can not be upgraded anymore (hence no strong reference remain)
are automatically discarded.
Currently, the selection is not saved/restored when completion
checkpoints are applied. This is usually fine because undoing changes
usually restores maps selections back in insert mode. But this is not
always the case and especially problematic in the presence of
multi-cursor completions (since completions are applied relative to
the selection/cursor) and snippets (which can change the selection)
* LSP: Support textDocument/prepareRename
'textDocument/prepareRename' can be used by the client to ask the
server the range of the symbol under the cursor which would be changed
by a subsequent call to 'textDocument/rename' with that position.
We can use this information to fill the prompt with an accurate prefill
which can improve the UX for renaming symbols when the symbol doesn't
align with the "word" textobject. (We currently use the "word"
textobject as a default value for the prompt.)
Co-authored-by: Michael Davis <mcarsondavis@gmail.com>
* clippy fixes
* rustfmt
* Update helix-term/src/commands/lsp.rs
Co-authored-by: Michael Davis <mcarsondavis@gmail.com>
* Update helix-term/src/commands/lsp.rs
Co-authored-by: Michael Davis <mcarsondavis@gmail.com>
* fix clippy from suggestions
* Update helix-term/src/commands/lsp.rs
Co-authored-by: Michael Davis <mcarsondavis@gmail.com>
---------
Co-authored-by: Michael Davis <mcarsondavis@gmail.com>
* use max_line_width + 1 during softwrap to account for newline char
Helix softwrap implementation always wraps lines so that the newline
character doesn't get cut off so he line wraps one chars earlier then
in other editors. This is necessary, because newline chars are always
selecatble in helix and must never be hidden.
However That means that `max_line_width` currently wraps one char
earlier than expected. The typical definition of line width does not
include the newline character and other helix commands like `:reflow`
also don't count the newline character here.
This commit makes softwrap use `max_line_width + 1` instead of
`max_line_width` to correct the impedance missmatch.
* fix typos
Co-authored-by: Jonathan Lebon <jonathan@jlebon.com>
* Add text-width to config.toml
* text-width: update setting documentation
* rename leftover config item
* remove leftover max-line-length occurrences
* Make `text-width` optional in editor config
When it was only used for `:reflow` it made sense to have a default
value set to `80`, but now that soft-wrapping uses this setting, keeping
a default set to `80` would make soft-wrapping behave more aggressively.
* Allow softwrapping to ignore `text-width`
Softwrapping wraps by default to the viewport width or a configured
`text-width` (whichever's smaller). In some cases we only want to set
`text-width` to use for hard-wrapping and let longer lines flow if they
have enough space. This setting allows that.
* Revert "Make `text-width` optional in editor config"
This reverts commit b247d526d69adf41434b6fd9c4983369c785aa22.
* soft-wrap: allow per-language overrides
* Update book/src/configuration.md
Co-authored-by: Pascal Kuthe <pascal.kuthe@semimod.de>
* Update book/src/languages.md
Co-authored-by: Pascal Kuthe <pascal.kuthe@semimod.de>
* Update book/src/configuration.md
Co-authored-by: Pascal Kuthe <pascal.kuthe@semimod.de>
---------
Co-authored-by: Pascal Kuthe <pascal.kuthe@semimod.de>
Co-authored-by: Jonathan Lebon <jonathan@jlebon.com>
Co-authored-by: Alex Boehm <alexb@ozrunways.com>
Co-authored-by: Blaž Hrastnik <blaz@mxxn.io>
* Do not add intermediate lines to jumplist with :<linenum> command.
* Revert jumplist index changes.
* Reduce calculations during update cycle.
* Use jumplist for undo, set jumplist before preview.
* remove some debug logging
* Revert "remove some debug logging"
This reverts commit 5772c4327e7121c53ea0726a4d7333ae1c413ffb.
* Revert "Use jumplist for undo, set jumplist before preview."
This reverts commit f73a1b29824feaf16477b9df547fb28d9db81923.
* Add last_selection, update implementation.
* @pascalkuthe initial feedback
* Ensure ":goto 123" keybinding works as expected.
* fix clippies, prefer expect() for expect last_selection state
This fixes blank row text in a DynamicPicker which is initially given
no options. This can happen for language servers which respond to
the workspace symbol request for an empty query with an empty list
of symbols, and that behavior is somewhat common since returning all
symbols as the spec suggests is very expensive.
For empty options, `Picker::new` calculated the widths of each column
as 0. We can recalculate the column widths when the new options are
set to fix this. This refactor is also a good opportunity to formalize
setting new options on a picker: besides setting the new options and
calculating column widths we also want to reset the cursor and rescore
the options.
Previously we did not respond to malformed or unhandled LSP requests.
The JSONRPC spec says that all non-notification requests must have
responses:
> When a rpc call is made, the Server MUST reply with a Response,
> except for in the case of Notifications
(Note that Helix is the "Server" in this case. Also from the spec:
"The Server is defined as the origin of Response objects and the
handler of Request objects.")
So this change sends error replies for requests which can't be parsed
or handled. Request IDs are also now added to the log messages for
unhandled requests.
This moves the `Application::claim_term` and
`helix-term::application::restore_term` functions into the helix-tui
crate. How the terminal should be claimed and restored is a TUI concern
and is implemented differently through different TUI backends.
This cleans out a lot of crossterm and TUI code in Application and
makes it easier to modify claim/restore based on information we query
from the terminal host. The child commit will take advantage of this
to cache the check for whether the host terminal supports the keyboard
enhancement protocol. Without this change, caching that information
takes much more code which is not easily reusable for anything else.
The code to restore the terminal is somewhat duplicated by this patch:
we want to restore the terminal in cases of panics. Panic handler hooks
must live for `'static` and the Application's terminal does not.
This refactors the snippet logic to be largely unaware of the rest of
the document. The completion application logic is moved into
generate_transaction_from_snippet which is extended to support
dynamically computing replacement text.
When accepting a snippet completion we automatically delete the
placeholders for now as doing so manual is quite cumbersome. In the
future we should keep these as a mark + virtual text that is
automatically removed once the cursor moves there.
