Now this was a cargo cult anti-pattern to write home about.
Doing painful save-and-restore bookkeeping around a separate
`magic_overruled` decoy global is just as messy as doing painful
save-and-restore logic around `p_magic` itself. only that now you need
to wrap every access to the effective value in a function call.
This replaces this with a marvellous new Clean Code technique™:
passing in the intended behavior as a function parameter to functions
where either the option or an explicit value might be used.
The added test shows the context, one expected a JSON field
to be a Object but it was a null value
pros: shows `vim.NIL` instead of `a userdata`
cons: the context `field 'foo'` is lost. I think this is generated
with internal magic which is hard to replicate.
This was originally used because we needed a lua interpreter to run the
standalone _busted_ lua application (except we used fuckery behind the
scenes to fake its binary dependencies with our own binary
dependencies). Now we longer do that, we instead run `test.harness` as a
standard nvim -l script. as this mode is no longer used in tests it
risks going to bitrot anyway.
NB: -ll mode still had some theoretical benefits like quicker
initialization and more "native" printing behavior, etc.
But we can incrementally work on "nvim -l" mode instead
of maintaining a separate special thing.
when running `zig build functionaltest`, it might
take a while until the test harness nvim instance checks
for interrupts, which causes a lot of tests to keep running after
CTRL+C. Check after each test instead
This is in preparation of zig 0.17 where b.install_path is no
longer known inside `pub fn build()`.
Builds which should hardcode a runtime path into the binary has to specify it.
We could use a nlua0 helper to encode the install path at build time but
I think this is cleaner anyway. (like currently the appimage still
encodes a temporary build path which is nonsensical, it is better to
just include none)
Similarily b.args will not be available anymore (there will just
be an opaque pass-through)
The new lua based runner replaces Makefile, runnvim.sh and runnvim.vim
As it happens, we run a `--headless` nvim inside a `:terminal` layer,
this is pointless.
Also there is still a lot remnants for oldesttests, but we don't
have any except for test1.in which just checks the environment
for following, nonexistant oldesttests. so just skip that.
For now, the actual vimscript code which runs in vim-under-test is
completely unchanged.
On macos, luajit is finally working with the latest ziglua master.
Also fix some minor bugs regarding locales, such as incorrect
HAVE_WORKING_LIBINTL checks
This fixes all clang 22 warnings of the form: "warning: diagnostic behavior
may be improved by adding the 'format(printf, 3, 4)' attribute to the
declaration of 'some_function'"
dialog_msg() was essentially a NULL-handling wrapper. but some callsites
already use their own NULL handling or NULL:s are impossible as the
value has already been assumed non-NULL. Non-locally assuming
a buffer size is also a code smell, especially if we can
just use "sizeof buff" without a wrapper.
append_redir() is... whatever it is.
CI currently uses clang-tidy 20, but this affects local builds
and CI is going to be upgraded sooner or later.
Some remaining systematic issues:
- clang-tidy warns agains any atoi() or atol() usage (because of no
error handling)
- functions which takes (char *fmt, char *only_string_arg) and expect
fmt to contain exactly one "%s" usage.
- error: initializing non-local variable with non-const expression depending on
uninitialized non-local variable (cppcoreguidelines-interfaces-global-init)
This is a much worse problem in C++ (hence C++ core guidelines) where
initialization is intermingled with arbitrary code execution. I
"think" in plain C, the linker will either resolve all these
deterministically or barf an error. But with some restructuring
we could make all static initialization actually static..
This feature might be a little silly and niche, but it is very useful
for _my_ workflow (and open source is about mee)
An issue which is never present on high quality RELEASE builds, but
might occur on Debug builds is that the Nvim server crashes
on some error in your unfinished PR code. If you compile your debug
builds with sanitizers enabled, as you should, the ASAN/UBSAN runtime
will print some useful info about your mistake to stderr or a log file,
such as a stack trace. This can be used to jump to the error in the
code.
This allows the nvim server to install a signal hander in the ui client,
which can load this log file in a good safe version of nvim and parse it
using 'errorformat'
This is inspired by the "press ENTER" free workflow of ui2 and applies
it beyond the lifetime cycle of the nvim instance.
example config:
```lua
local asan = vim.env.ASAN_OPTIONS
if asan ~= nil and string.match(asan, "log_path=/tmp/nvim_asan") then
local myname = "/tmp/nvim_asan."..vim.uv.getpid()
local args = {"--embed", "-n", "+set efm=%+A%*[^/]%f:%l:%c", "+silent cfile "..myname, "+silent cfirst", "+silent copen"}
vim.api.nvim__set_restart_on_crash("nvim", args)
end
```
and run your debug nvim like so
ASAN_OPTIONS=handle_abort=1,handle_sigill=1,log_path=/tmp/nvim_asan ./build/bin/nvim
Both for tests and for system wide install, $PREFIX/lib/nvim/parser
is a valid path for tree-sitter parsers. This also brings the build.zig
behavior in line with how we set up the paths in CMakeLists.txt
neovim always uses encoding=utf8. Thus a lot of utf8 tests are just a
lot of tests. Meanwhile test_regexp_latin.vim is a bit of a lone child,
it can just run as a separate test without a shell layer (running it in
the same process as its utf8 sibling is not allowed
the unix.vim file was probably accidentally ignored at some point.
An actual invokation of nvim-under-test would in practice look like
["/path/to/neovim/build/bin/nvim", "-u", "unix.vim", "-U", "NONE", "-i", "NONE", "--noplugin", "--headless", "-u", "NONE", "--cmd", "set shortmess-=F", "-S", "runtest.vim", "test_arabic.vim"]
but -u NONE cancels out the earlier -u unix.vim
By now, too many tests rely on specific behavior from "NONE", so copy in
the useful parts of unix.vim to the cmdline again. also, some tests
conflict with `directory=.` (or even `directory=Xtempswapdir`) so don't use that.
`-U NONE` is dead code in Nvim, remove it.
using the GNU compiler we just get a bunch of const warnings we can fix.
clang, however, gets really upset that the standard library suddenly
starts using a lot of c11 features, despite us being in -std=gnu99 mode.
Basically, _GNU_SOURCE which we set is taken as a _carte blanche_ by the
glibc headers to do whatever they please, and thus we must inform clang
that everything is still OK.
Previous implementation used a full sort but this is unnecessary. Pairs
do not need to be processed in id-order, all that we did was make
"start" and "end" marks find each other. This can be done with a hash
table instead. The "grow only" khash_t mentioned in the TODO is
the since long merged #249855970157e refactor with dense value array,
to not regress on memory locality.
This makes little difference in practice except for extremely unlikely
ambiguities. But it looks nicer in compiler errors and stacktraces where
you see the fully expanded names
fix#37867
This bug happens when only one end is moved between different nodes,
but the other end is also moved but within the same node.
When this happens we need the correct previous position even for the
internal move. This code shall be refactored to make the intent clearer,
(and avoid some unnecessary processing) but this is a fix for the observable
bug.
Thanks to KevinGoodsell for providing a deterministic
reproduce using fuzzing techniques.
fixes#37586
when doing `packadd mypackage` up to two exact paths are added
to &rtp. Instead of recalculating runtime_search_path from scratch,
we can "just" splice these two paths in
This is simple in theory, but get complicated in practice as
"after" dirs do exist and need some wrangling.
Echasnovski did some benchmarking, to show that this reduces overhead
of a init.lua configuration style where separate `packadd!` calls are
used spread out during the config. In addition, "batched" addition
(either using "start" packages or packadd! a lot of opt packages at
once) does not regress.
A theoretical simplification could be to NEVER explicitly add "after"
dirs to &rtp, but implicitly add all existing "after" dirs in reverse
order when calculating the effective run time path. This might be tricky
to do without breaking 12 tpope plugins again tho.
We might also instead consider solutions where &rtp remains fully expanded but no longer is the main source of truth. But this is all post 0.12 work. This PR is an alright stopgap to make 0.12 fully support intended use cases of vim.pack.add() .
These were disabled as they failed when tested with cross-compiling,
but it works fine for windows natively on ci, so let's use that.
mac os builds after disabling debug info, which currently
triggers a bug in zig's MachO linker: natecraddock/ziglua#191
Tracking issue: https://codeberg.org/ziglang/zig/issues/30669
Unfortunately actually using the JIT of luajit still triggers
issues on mac os, so we cannot use it yet.
Also sync luajit version used via ziglua to a recent one
like we use in cmake, see natecraddock/ziglua#196
On linux /dev/stdin is defined as a symlink to /proc/self/fd/0
This in turn is defined as a "magic" symlink which is allowed to point
to internal kernel objects which really does not have a file
name. As a glaring inconsistency, fopen("/proc/self/fd/0", "r")
works if fd was originally opened using pipe() but not using
socketpair(). As it happens UV_CREATE_PIPE does not create pipes
but creates socket pairs. These two unfortunate conditions
means that using /dev/stdin and similar does not work in
shell commands in nvim on linux. as a work around, override
libuv's descicion and create an actual pipe pair.
This change is not needed on BSD:s but done unconditionally for simplicity,
except for on windows where it is not done for stdout because of windows
fixes#35984
Working towards making the build.zig build actually useful for
users..
For building a full runtime distribution in /usr or /opt/nvim or
whatever, use
sudo zig build install -p /usr
by default "zig build" will now work like "make" e.g. just build the binary
and the generated parts of the runtime. This will work for development,
but you need to use both parts of the runtime, just like with
an "uninstalled" cmake build:
zig build
VIM=. ./zig-out/bin/nvim --clean --cmd "set rtp+=./zig-out/runtime"
As a wrapper, `zig build run_dev` can be used
compile time features are hot again.
Note: this changes the &term value for builtin definition from
'builtin_xterm' to just 'xterm'. It's an xterm regardless of we use an
external definition or an internal. Prior to this commit the vast
majority of POSIX users will have used external terminfo, so plugins and
scripts are only going to have checked for &term == 'xterm' or 'tmux' or
whatever.
The status of external loading is still available in "nvim -V3" output.
This was part of an attempt to add "git describe" to zig build
without re-building too much. This doesn't yet work as zig upstream
changes are are also needed, but I think this was a sensible refactor
even in isolation, so breaking it out.
"API dispatch" doesn't depend on any ui event stuff nor on
version, that was just an accident of the metadata collection
being crammed into the same file.
Remove "manual invocation" instruction. this is going to bitrot anyway.
Both cmake and build.zig allows you to extract the command line of a
step, so you can debug it separately.
problem: termkey/driver-ti.c had its internal dependency upon unibilium
which would completely skip builtin terminfo defs.
solution: add termkey info to TerminfoEntry struct
NOTE: this disables a lot of named function keys which are present as
terminfo "keys" both are mostly unset in terminfo entries for modern
terminals, and also not recognized by nvim as mappable keys
anyway, except a few ones like `<undo>` which this still will keep.
We probably don't want to encode up to F63 keys forever. While only 12
physical keys are available on modern keybords, instead Chords using
F-keys are usually encoded as high key numbers, like <C-S-F3>
becoming <F39> . But reconsideirg that has implications for configuration
that is best done as a separate (breaking) change.
The processing of terminfo can be separated into two steps:
1. The initialization of terminfo, which includes trying to find $TERM
in a terminfo database file. As a fallback, common terminfo
definitions are compiled in. After this, we apply a lot of ad-hoc
patching to cover over limitations of terminfo.
2. While processing updates from nvim, actually using terminfo strings
and formatting them with runtime values. for this part, terminfo
essentially is a hyper-enhanced version of snprintf(), including
a sm0l stack based virtual machine which can manipulate the runtime
parameters.
This PR completely replaces libuniblium for step 2, with code
vendored from NetBSD's libtermkey which has been adapted to use typesafe
input parameters and to write into an output buffer in place.
The most immedatiate effects is a performance enhancement of
update_attrs() which is a very hot function when profiling the
TUI-process part of screen updates. In a stupid microbenchmark
(essentially calling nvim__screenshot over and over in a loop) this
leads to a speedup of ca 1.5x for redrawing the screen on the TUI-side.
What this means in practise when using nvim as a text editor is probably
no noticible effect at all, and when reabusing nvim as idk a full screen
RGB ASCII art rendrer maybe an increase from 72 to 75 FPS LMAO.
As nice side-effect, reduce the usage of unibilium to initialization only..
which will make it easier to remove, replace or make unibilium optional,
adressing #31989. Specifically, the builtin fallback doesn't use
unibilium at all, so a unibilium-free build is in principle possible
if the builtin definitions are good enough.
As a caveat, this PR doesn't touch libtermkey at all, which still has a
conditional dependency on unibilium. This will be investigated in a
follow-up PR
Note: the check of $TERMCOLOR was moved from tui/tui.c to
_defaults.lua in d7651b27d5 as we want to
skip the logic in _defaults.lua if the env var was set, but there
is no harm in TUI getting the right value when the TUI is trying to
initialize its terminfo shenanigans. Also this check is needed when
a TUI connects to a `--headless` server later, which will observe
a different $TERMCOLOR value than the nvim core process itself.
- Bump zig version to 0.15.1 and workaround zig fetch hang (ziglang/zig#24916)
- add mac os zig build (currently without luajit, linker failure)
- Add windows zig build, currently with very limited testing
Notably, the existence of a swap file like .buffer.c.swp might cause
mayhem. As a reasonable filter, only process *.c and *.h files using the
c grammar.
This partially reverts 0b8a72b739,
that is unreverts 15e77a56b7
"priority" is an internal neovim concept which does not occur in shared
queries. Ideally a single priority space should eventually be enough
for our needs. But as we don't want to poke at the usages of
priorities right now in the wider ecosystem,
introduce the "subpriorities" so that treesitter code can distinguish
highlights of the same priorities with different tree nesting depth.
This mainly affects `injection.combined` as parent-tree nodes might appear
in the middle of child-tree nodes which otherwise is not possible.
Continuing the work of #31400
That PR allowed the provider to be invoked multiple times per line.
We want only to do that when there actually is more data later on the
line. Additionally, we want to skip over lines which contain no new
highlight items. The TS query cursor already tells us what the next
position with more data is, so there is no need to reinvoke the range
callback before that.
NB: this removes the double buffering introduced in #32619 which
is funtamentally incompatible with this (nvim core is supposed to keep
track of long ranges by itself, without requiring a callback reinvoke
blitz). Need to adjust the priorities some other way to fix the same issue.
addDirectoryArg with a source dir is not enough to depend on changes
to invidiual files. as a workaround, mark this step as having
side-effects (always being run) for now.
bfredl