Problem:
vim.uv.os_setenv gets "stuck" per-key. #32550
Caused by the internal `envmap` cache. #7920
:echo $FOO <-- prints nothing
:lua vim.uv.os_setenv("FOO", "bar")
:echo $FOO <-- prints bar (as expected)
:lua vim.uv.os_setenv("FOO", "fizz")
:echo $FOO <-- prints bar, still (not expected. Should be "fizz")
:lua vim.uv.os_unsetenv("FOO")
:echo $FOO <-- prints bar, still (not expected. Should be nothing)
:lua vim.uv.os_setenv("FOO", "buzz")
:echo $FOO <-- prints bar, still (not expected. Should be "buzz")
Solution:
- Remove the `envmap` cache.
- Callers to `os_getenv` must free the result.
- Update all call-sites.
- Introduce `os_getenv_noalloc`.
- Extend `os_env_exists()` the `nonempty` parameter.
Same idea as a7be4b7bf8, but that only showed the context if the
length of the string differed. Since these tests check both string
length and string content, the ctx should be provided for both.
ERROR test/unit/testutil.lua @ 797: vim_snprintf() positional arguments
test/unit/testutil.lua:769: test/unit/testutil.lua:753: (string) '
test/unit/strings_spec.lua:159: snprintf(buf, 4, "%1$0.*2$b", 12ULL, cdata<int>: 0xf78c8ed8) = 001100
Expected objects to be the same.
Passed in:
(string) '000'
Expected:
(string) '001''
Problem: On 32-bit architectures, musl libc makes heavy use of
__typeof__ as part of its __REDIR macro for optional backwards
compatibility with 32-bit time_t values. Unfortunately, the
__typeof__ keyword is not supported by the LuaJIT C parser.
Solution: Filter out the keyword in filter_complex_blocks.
Problem: This keyword is used by GCC and Clang to prevent -Wpedantic
(and other options) from emitting warnings for many GNU C extensions.
This is used heavily in Alpine Linux through musl libc and
foritfy-headers. Without filtering the __extension__ keyword some type
definitions are duplicated. For example, timeval is defined once as
struct timeval { time_t tv_sec; suseconds_t tv_usec; };
and once as:
__extension__ struct timeval { time_t tv_sec; suseconds_t tv_usec; };
Without this patch, the LuaJIT C parser doesn't recognize that these
definitions are equivalent, causing unit test to fail on Alpine Linux.
Solution: Filter out the keyword in filter_complex_blocks.
Since these assertions all use a common function to perform the test
assertions, it's difficult to figure out which test failed:
ERROR test/unit/testutil.lua @ 785: vim_snprintf() positional arguments
test/unit/testutil.lua:757: test/unit/testutil.lua:741: (string) '
test/unit/strings_spec.lua:143: Expected objects to be the same.
Passed in:
(number) 6400
Expected:
(number) 6'
exit code: 256
Adding context to the assertion makes it clearer what the problem is:
ERROR test/unit/testutil.lua @ 785: vim_snprintf() positional arguments
test/unit/testutil.lua:757: test/unit/testutil.lua:741: (string) '
test/unit/strings_spec.lua:149: snprintf(buf, 0, "%1$0.*2$b", cdata<unsigned int>: 0xf78d0f38, cdata<int>: 0xf78dc4e0) = 001100
Expected objects to be the same.
Passed in:
(number) 6400
Expected:
(number) 6'
exit code: 256
This commit adds basic support for the kitty keyboard protocol to
Neovim's builtin terminal. For now only the first mode ("Disambiguate
escape codes") is supported.
Problem:
Whether an option is allowed to be empty isn't well defined and
isn't properly checked.
Solution:
- For non-list string options, explicitly check the option value
if it is empty.
- Annotate non-list string options that can accept an empty value.
- Adjust command completion to ignore the empty value.
- Render values in Lua meta files
It's probably not worth adding the C test files to regular formatting as
they're pretty much never touched, but ensuring the files are formatted
according to our standards and getting rid of warnings is a cheap
one-time fix.
In order to run unittests with a release build, we need the test
functions to be accessible when NDEBUG is defined. Moving the functions
into the test fixture ensures they are available and only available for
use by the unit tests.
These were imported from the v0.3.3 git tag
https://github.com/neovim/libvterm/tree/v0.3.3 and not the latest
commit. This is for compatibility reasons as the libvterm code was
imported from v0.3.3.
In the api_info() output:
:new|put =map(filter(api_info().functions, '!has_key(v:val,''deprecated_since'')'), 'v:val')
...
{'return_type': 'ArrayOf(Integer, 2)', 'name': 'nvim_win_get_position', 'method': v:true, 'parameters': [['Window', 'window']], 'since': 1}
The `ArrayOf(Integer, 2)` return type didn't break clients when we added
it, which is evidence that clients don't use the `return_type` field,
thus renaming Dictionary => Dict in api_info() is not (in practice)
a breaking change.
Skipped importing the following unit tests from libtermkey as they'd
require introducing a lot of unused code or require more effort to port
than is probably worth:
- 05read
- 12strpkey
- 20canon
- 40ti-override
Use the grapheme break algorithm from utf8proc to support grapheme
clusters from recent unicode versions.
Handle variant selector VS16 turning some codepoints into double-width
emoji. This means we need to use ptr2cells rather than char2cells when
possible.
This also makes shada reading slightly faster due to avoiding
some copying and allocation.
Use keysets to drive decoding of msgpack maps for shada entries.
Before this change, "static inline" functions in headers needed to have
their function attributes specified in a completely different way. The
prototype had to be duplicated, and REAL_FATTR_ had to be used instead
of the public FUNC_ATTR_ names.
TODO: need a check that a "header.h.inline.generated.h" file is not
forgotten when the first "static inline" function with attributes
is added to a header (they would just be silently missing).
Problem: fnamemodify() treats ".." and "../" differently.
Solution: Expand ".." properly like how "/.." is treated in 8.2.3388.
(zeertzjq)
closes: vim/vim#152181ee7420460
Problem: the recursive parameter in the *_equal functions can be removed
Solution: Remove the recursive parameter in dict_equal(), list_equal()
object_equal and tv_equal(). Use a comparison of the static
var recursive_cnt == 0 to determine whether or not tv_equal()
has been called recursively (Yinzuo Jiang).
closes: vim/vim#150707ccd1a2e85
Co-authored-by: Yinzuo Jiang <jiangyinzuo@foxmail.com>
If you like it you shouldn't put a ring on it.
This is what _every_ consumer of RStream used anyway, either by calling
rbuffer_reset, or rbuffer_consumed_compact (same as rbuffer_reset
without needing a scratch buffer), or by consuming everything in
each stream_read_cb call directly.
Problem: too many strlen() calls in search.c
Solution: refactor code and remove more strlen() calls,
use explicit variable to remember strlen
(John Marriott)
closes: vim/vim#147968c85a2a49a
Co-authored-by: John Marriott <basilisk@internode.on.net>
Use uv_fs_realpath() instead.
It seems that uv_fs_realpath() has some problems on non-Linux platforms:
- macOS and other BSDs: this function will fail with UV_ELOOP if more
than 32 symlinks are found while resolving the given path. This limit
is hardcoded and cannot be sidestepped.
- Windows: while this function works in the common case, there are a
number of corner cases where it doesn't:
- Paths in ramdisk volumes created by tools which sidestep the Volume
Manager (such as ImDisk) cannot be resolved.
- Inconsistent casing when using drive letters.
- Resolved path bypasses subst'd drives.
Ref: https://docs.libuv.org/en/v1.x/fs.html#c.uv_fs_realpath
I don't know if the old implementation that uses uv_chdir() and uv_cwd()
also suffers from the same problems.
- For the ELOOP case, chdir() seems to have the same limitations.
- On Windows, Vim doesn't use anything like chdir() either. It uses
_wfullpath(), while libuv uses GetFinalPathNameByHandleW().
Add more filters so that LuaJIT can parse headers on macOS 14.
The system headers use a style of enum introduced in C++11 (and allowed
as an extension in C by clang) of the form:
enum Name : Type {
The system headers also use bitfields in the mach_vm_range_recipe* types:
struct Foo { int bar : 32; }
Neither of these constructs can be parsed by LuaJIT, so filter the lines
out. Neither of these declarations are used by neovim's unittests.
There is a (now closed) issue about bitfields for LuaJIT:
https://github.com/LuaJIT/LuaJIT/issues/951Fixes#26145.
If the filename passed to vim_FullName() is a relative directory, and
does not exist, it is appended to the current working directory. Since
the return value of append_path() was ignored, and if the buffer length
was too small to fit getcwd() + dirname(filename), it would still try to
append the basename(filename).
This was manifesting as a failure in test/unit/path_spec.lua in:
itp('fails and uses filename if given filename contains non-existing directory', ..
This failure occurs when running the tests from directory with a short
path such as: /work/src/nv
test/unit/path_spec.lua:420: Expected objects to be the same.
Passed in:
(string) '/work/src/nv/test.file'
Expected:
(string) 'non_existing_dir/test.file'
This return value for the second call to append_path() to append
basename(filename) was checked, and this is where it would fail for
normal / longer getcwd()s.
LuaJIT does not handle -0.0 correctly in 'dual number mode' (which is
the default, and only supported mode for LuaJIT arm64). If LuaJIT is
forced to use 'dual number mode' on X64 (where the default is single),
this test will fail in the same manner.
Fix this by using tonumber("-0.0") instead of a -0.0 literal.
See: https://github.com/LuaJIT/LuaJIT/issues/858
Then we can just load metadata in C as a single msgpack blob. Which also
can be used directly as binarly data, instead of first unpacking all the
functions and ui_events metadata to immediately pack it again, which was
a bit of a silly walk (and one extra usecase of `msgpack_rpc_from_object`
which will get yak shaved in the next PR)
