Problem: nvim_open_win blocking all win_set_buf autocommands when !enter &&
!noautocmd is too aggressive.
Solution: temporarily block WinEnter/Leave and BufEnter/Leave events when
setting the buffer. Delegate the firing of BufWinEnter back to win_set_buf,
which also has the advantage of keeping the timing consistent (e.g: before the
epilogue in enter_buffer, which also handles restoring the cursor position if
autocommands didn't change it, among other things). Reword the documentation for
noautocmd a bit.
I pondered modifying do_buffer and callees to allow for BufEnter/Leave being
conditionally disabled, but it seems too invasive (and potentially error-prone,
especially if new code paths to BufEnter/Leave are added in the future).
Unfortunately, doing this has the drawback of blocking ALL such events for the
duration, which also means blocking unrelated such events; like if window
switching occurs in a ++nested autocmd fired by win_set_buf. If this turns out
to be a problem in practice, a different solution specialized for nvim_open_win
could be considered. :-)
Problem: currently, for splits, nvim_win_set_config accepts win without any of
split or vertical set, which has little effect and seems error-prone.
Solution: require at least one of split or vertical to also be set for splits.
Also, update nvim_win_set_config docs, as it's no longer limited to just
floating and external windows.
Problem: splitting is disallowed in some cases to prevent the window layout
changes while a window is closing, but it's not checked for.
Solution: check for this, and set the API error message directly.
(Also sneak in a change to tui.c that got lost from #27352; it's a char* buf,
and the memset is assuming one byte each anyway)
Problem: WinNew and win_enter autocommands can delete the target buffer to
switch to, causing a heap-use-after-free.
Solution: store a bufref to the buffer, check it before attempting to switch.
Problem: if switch_win{_noblock} fails to restore the old curwin after WinNew
(e.g: it was closed), wp will become the new curwin, but win_set_buf enter
events would still be blocked if !enter && !noautocmd.
Solution: fire them, as we've actually entered the new window.
Note: there's a problem of switch_win{_noblock} failing to restore the old
curwin, leaving us in wp without triggering WinEnter/WinLeave, but this affects
all callers of switch_win{_noblock} anyways. (It's also not clear how WinLeave
can be called if the old curwin was closed already).
Problem: BufWinEnter is not fired when not entering a new window, even when a
different buffer is specified and buffer-related autocommands are unblocked
(!noautocmd).
Solution: fire it in the context of the new window and buffer. Do not do it if
the buffer is unchanged, like :{s}buffer.
Be wary of autocommands! For example, it's possible for nvim_win_set_config to
be used in an autocommand to move a window to a different tabpage (in contrast,
things like wincmd T actually create a *new* window, so it may not have been
possible before, meaning other parts of Nvim could assume windows can't do
this... I'd be especially cautious of logic that restores curwin and curtab
without checking if curwin is still valid in curtab, if any such logic exists).
Also, bail early from win_set_buf if setting the temp curwin fails; this
shouldn't be possible, as the callers check that wp is valid, but in case that's
not true, win_set_buf will no longer continue setting a buffer for the wrong
window.
Note that pum_create_float_preview also uses win_set_buf, but from a glance,
doesn't look like it properly checks for autocmds screwing things up (win_enter,
nvim_create_buf...). I haven't addressed that here.
Also adds some test coverage for nvim_open_win autocommands.
Closes#27121.
Problem: win_set_config should have the observable effect of moving an existing
window to another place, but instead fires autocommands as if a new window was
created and entered (and does not fire autocommands reflecting a "return" to the
original window).
Solution: do not fire win_enter-related autocommands when splitting the window,
but continue to fire them when entering the window that fills the new space when
moving a window to a different tabpage, as the new curwin changes.
Also, remove "++once" from the WinEnter autocmd in the other test, as omitting
it also crashed Nvim before this fix.
Problem: win_enter autocommands can close new_curwin, crashing if it was the
last window in its tabpage after removing win, or can close parent, crashing
when attempting to split it later.
Solution: remove win first, check that parent is valid after win_enter.
NOTE: This isn't actually quite right, as this means win is not in the window
list or even has a frame when triggering enter autocommands (so it's not
considered valid in the tabpage). This is addressed in later commits.
Getting current channel info was kind of annoying via RPC. Two
functions had to be called:
1. `nvim_get_api_info` which returns `[channel_id, meta_data]`.
- This results in `channel_id = api.nvim_get_api_info()[0]`.
- Here the meta_data is sent but never used.
2. Finally call `nvim_get_chan_info(channel_id)`.
This commit reduces the need for `nvim_get_api_info` as passing 0
returns current channel info.
Problem:
When nvim_input is followed immediately by non-fast events on RPC, both
events and input are available after the polling done by the os_inchar()
in state_enter(), but state_enter() then chooses to process events even
if input is available, which is inconsistent with state_handle_k_event()
that stops processing events once input is available.
Solution:
Also check for available input after the os_inchar() in state_enter().
The "priority" field of extmarks can be used to set priorities of
extmarks which dictates which highlight group a range will actually have
when there are multiple extmarks applied. However, when multiple
extmarks have the same priority, the only way to enforce an actual
priority is through the order in which the extmarks are set.
It is not always possible or desirable to set extmarks in a specific
order, however, so we add a new "subpriority" field that explicitly
enforces the ordering of extmarks that have the same priority.
For now this will be used only to enforce priority of treesitter
highlights. A single node in a treesitter tree may match multiple
captures, in which case that node will have multiple extmarks set. The
order in which captures are returned from the treesitter API is not
_necessarily_ in the same order they are defined in a query file, so we
use the new subpriority field to force that ordering.
For now subpriorites are not documented and are not meant to be used by
external code, and it only applies to ephemeral extmarks. We indicate
the "private" nature of subpriorities by prefixing the field name with
an "_".
Problem: Things that temporarily change/restore curwin/buf (e.g:
win_execute, some autocmds) may break assumptions that
curwin/buf is the cmdwin when "cmdwin_type != 0", causing
issues.
Solution: Expose the cmdwin's real win/buf and check that instead. Also
try to ensure these variables are NULL if "cmdwin_type == 0",
allowing them to be used directly in most cases without
checking cmdwin_type. (Sean Dewar)
Reset and save `cmdwin_old_curwin` in a similar fashion.
Apply suitable changes for API functions and add Lua tests.
988f74311c
Extmarks can contain URLs which can then be drawn in any supporting UI.
In the TUI, for example, URLs are "drawn" by emitting the OSC 8 control
sequence to the TTY. On terminals which support the OSC 8 sequence this
will create clickable hyperlinks.
URLs are treated as inline highlights in the decoration subsystem, so
are included in the `DecorSignHighlight` structure. However, unlike
other inline highlights they use allocated memory which must be freed,
so they set the `ext` flag in `DecorInline` so that their lifetimes are
managed along with other allocated memory like virtual text.
The decoration subsystem then adds the URLs as a new highlight
attribute. The highlight subsystem maintains a set of unique URLs to
avoid duplicating allocations for the same string. To attach a URL to an
existing highlight attribute we call `hl_add_url` which finds the URL in
the set (allocating and adding it if it does not exist) and sets the
`url` highlight attribute to the index of the URL in the set (using an
index helps keep the size of the `HlAttrs` struct small).
This has the potential to lead to an increase in highlight attributes
if a URL is used over a range that contains many different highlight
attributes, because now each existing attribute must be combined with
the URL. In practice, however, URLs typically span a range containing a
single highlight (e.g. link text in Markdown), so this is likely just a
pathological edge case.
When a new highlight attribute is defined with a URL it is copied to all
attached UIs with the `hl_attr_define` UI event. The TUI manages its own
set of URLs (just like the highlight subsystem) to minimize allocations.
The TUI keeps track of which URL is "active" for the cell it is
printing. If no URL is active and a cell containing a URL is printed,
the opening OSC 8 sequence is emitted and that URL becomes the actively
tracked URL. If the cursor is moved while in the middle of a URL span,
we emit the terminating OSC sequence to prevent the hyperlink from
spanning multiple lines.
This does not support nested hyperlinks, but that is a rare (and,
frankly, bizarre) use case. If a valid use case for nested hyperlinks
ever presents itself we can address that issue then.
Problem:
Not all Lua code is checked by stylua. Automating code-style is an
important mechanism for reducing time spent on accidental
(non-essential) complexity.
Solution:
- Enable stylua for entire `test/` directory.
- Exclude these high-churn files until this issue is resolved: https://github.com/JohnnyMorganz/StyLua/issues/829
```
test/functional/ui/decorations_spec.lua | 3560 ++++++++++++++++++++++++++++++++++++----------------
test/functional/ui/float_spec.lua | 5826 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++-------------------------
test/functional/ui/multigrid_spec.lua | 1349 ++++++++++++++------
```
- Make surgical changes to these files (or add `stylua: ignore` in some
small scopes) to improve the result:
```
test/functional/vimscript/msgpack_functions_spec.lua | 1414 +++++++++++++++------
test/functional/api/buffer_spec.lua | 1389 +++++++++++----------
test/functional/api/vim_spec.lua | 2740 +++++++++++++++++++++++-----------------
```
- These "high churn" files are NOT excluded because the changes are
largely an improvement:
```
test/functional/plugin/lsp_spec.lua | 2198 ++++++++++++++++++---------------
test/functional/plugin/shada_spec.lua | 4078 +++++++++++++++++++++++++++++++++++-------------------------
test/functional/ui/cmdline_spec.lua | 1199 +++++++++++-------
test/functional/ui/popupmenu_spec.lua | 1267 +++++++++++--------
test/functional/ui/messages_spec.lua | 1643 +++++++++++++++---------
```
- TODO: how to check "all directories"? With `GLOB_DIRS *` and `/.deps/` (or
`.deps/`) in `.styluaignore`, Lua code in `.deps/` is still checked...
Problem: Unable to predict which byte-offset to place virtual text to
make it repeat visually in the wrapped part of a line.
Solution: Add a flag to nvim_buf_set_extmark() that causes virtual
text to repeat in wrapped lines.
Problem: We have `P_(BOOL|NUM|STRING)` macros to represent an option's type, which is redundant because `OptValType` can already do that. The current implementation of option type flags is also too limited to allow adding multitype options in the future.
Solution: Remove `P_(BOOL|NUM|STRING)` and replace it with a new `type_flags` attribute in `vimoption_T`. Also do some groundwork for adding multitype options in the future.
Side-effects: Attempting to set an invalid keycode option (e.g. `set t_foo=123`) no longer gives an error.
Problem: line2byte behavior is changed after commit b051b13. It no
longer return `-1` on empty buffer.
Solution: use `nof_ff` instead of `!ff_dos` as condition. Then
compatible behavior of line2byte() is restored.
This is the command invoked repeatedly to make the changes:
:%s/^\(.*\)|\%(\*\(\d\+\)\)\?$\n\1|\%(\*\(\d\+\)\)\?$/\=submatch(1)..'|*'..(max([str2nr(submatch(2)),1])+max([str2nr(submatch(3)),1]))/g
Problem:
Since e057b38e70#20757 we support empty key in JSON encode/decode,
but we don't allow it in RPC object => Vim dict conversion. But empty
string is a valid key in Vim dicts and the msgpack spec.
Empty string key was disallowed in 7c01d5ff92 (2014) but that
commit/PR doesn't explicitly discuss it, so presumably it was a "seems
reasonable" decision (or Vimscript didn't allow empty keys until later).
Solution:
Remove the check in `object_to_vim()`. Note that
`tv_dict_item_alloc_len` will invoke `memcpy(…, 0)` but that's allowed
by the C spec: https://stackoverflow.com/a/3751937/152142
