ci: install nodejs 8 in Appveyor, Travis
provider: check node version for debug support
Resolve https://github.com/neovim/neovim/pull/7577#issuecomment-350590592 for Unix.
provider: test if nodejs in ci supports --inspect-brk
nodejs host for neovim requires nodejs 6+ to work properly.
nodejs 6.12+ or 7.6+ is required for debug support via `node --inspect-brk`.
provider: run cli.js of nodejs host directly
npm shims are useless because the user cannot set node to debug mode via
--inspect-brk. This is problematic on Windows which use batchfiles and
shell scripts to compensate for not supporting shebang.
The patch uses `npm root -g` to get the absolute path of the global npm
modules. If that fails, then the user did not install neovim npm package
globally. Use that absolute path to find `neovim/bin/cli.js`, which is
what the npm shim actually runs with node. glob() is for a simple file
check in case bin/ is removed because the npm shims are ignored now.
Since "builtin" terminfo definitions were implemented (7cbf52db1b),
the decisions made by tui.c and terminfo.c are more relevant. Exposing
that decision in the 'term' option helps with troubleshooting.
Also: remove code that allowed setting t_Co. `:set t_Co=…` has never
worked; the highlight_spec test asserting that nvim_set_option('t_Co')
_does_ work makes no sense, and should not have worked.
Asynchronous API functions are served immediately, which means pending
input could change the state of Nvim shortly after an async API function
result is returned.
nvim_get_mode() is different:
- If RPCs are known to be blocked, it responds immediately (without
flushing the input/event queue)
- else it is handled just-in-time before waiting for input, after
pending input was processed. This makes the result more reliable
(but not perfect).
Internally this is handled as a special case, but _semantically_ nothing
has changed: API users never know when input flushes, so this internal
special-case doesn't violate that. As far as API users are concerned,
nvim_get_mode() is just another asynchronous API function.
In all cases nvim_get_mode() never blocks for more than the time it
takes to flush the input/event queue (~µs).
Note: This doesn't address #6166; nvim_get_mode() will provoke #6166 if
e.g. `d` is operator-pending.
Closes#6159
Lua has too many pitfalls here:
- os.execute() requires shell-escaping
- os.execute() has breaking changes between Lua 5.1 and 5.2
- No native way in Lua to handle "readonly" etc. on Windows
Hope this will make people using feed_command less likely: this hides bugs.
Already found at least two:
1. msgpackparse() will show internal error: hash_add() in case of duplicate
keys, though it will still work correctly. Currently silenced.
2. ttimeoutlen was spelled incorrectly, resulting in option not being set when
expected. Test was still functioning somehow though. Currently fixed.
Escaping from a '-- More --' prompt in tests is awkward as it doesn't
take keys from the typebuffer, requiring a call to `feed()` in lua at
the correct time. Moreover, it's rarer that a test will want the
'-- More --' prompt to be activated than not.
- Vim "unix default" of 'noshowcmd' is serving few users. And it's
inconsistent.
- 'ruler' and 'belloff=all' improve the out-of-the-box experience.
- Continue to use 'noshowcmd' and 'noruler' by default in the functional
tests to keep them fast.
TODO: Add a "disable slow stuff" command or mapping to address the
use-case of a very slow terminal connection.
- Do not exclude any directories from `find` search, remove dumps before tests
instead.
- Install `apport` on travis so that linux tests should produce core dumps
(based on information from travis-ci/travis-ci#3754, not sure whether it still
applies).
- Check cores in lua so that one has an idea which test is failing exactly. Do
this only 10% of time on linux because traversing the file system is slow.
Unit tests are still not touched, though it is what `app` argument in
`check_cores` is for.
TODO? consider using `find`, it may be faster. Consider retiring `os.execute`,
dealing with escaping is bad.
- Add support for TEST_FILE to the `oldtest` target, for consistency
with the busted/lua tests.
Caveat: with the busted/lua tests TEST_FILE takes a full path, whereas
for `oldtest` it must be "test_foo.res".
- Add support for NVIM_PRG, again so that all test-related targets are
consistent.
- Use consistent name for NVIM_PRG. But still need to support NVIM_PROG
for QuickBuild CI.
Note: The `oldtest` target is driven by the top-level Makefile, because
it requires a TTY. CMake 3.2 added a USES_TERMINAL flag to
add_custom_target(). But we support CMake 2.8...
add_custom_target(oldtest
COMMAND make clean
COMMAND make NVIM_PRG=$<TARGET_FILE:nvim> $ENV{MAKEOVERRIDES}
DEPENDS nvim
WORKING_DIRECTORY "${CMAKE_CURRENT_SOURCE_DIR}/src/nvim/testdir"
USES_TERMINAL true
)
Previously alternate branches were not accounted for properly, with this
change g- after an undo to a branch point works.
The current sequence number b_u_seq_cur is used in undo_time(), in
u_doit() this was calculated by subtracting one from the curhead
sequence number.
The curhead header entry represents the change that was just undone, so
the sequence number we want is that of the change we have moved to. This
is the sequence number of the undo head that is the uh_next element of
this curhead. That sequence number is not always one less than the
curhead sequence number -- there may have been an alternate branch at
this point.
Instead of subtracting one, we now directly find the sequence number of
curhead->uh_next.
- Default to powershell.
- Avoid hardcoded "-c".
- Remove ^M character from received lines.
- pending_win32(): clear() is unnecessary and it pollutes the tests.
Closes#3973
Helped-by: Rui Abreu Ferreira <raf-ep@gmx.com>
Note some bugs were judged to have too ugly a fix to solve, tests to
demonstrate these problems, and the explanation behind not fixing them
are below.
describe('register . problems', function()
before_each(reset)
-- The difficulty here is: The basic requirement is that the text
-- inserted is treated as if it were typed in insert mode. This is why
-- the paste method is to enter insert mode and enter the ". register
-- into readbuf1.
-- We can't add a count into the readbuf here because the insert mode
-- count is implemented with readbuf2 which is checked for characters
-- after readbuf1.
-- Hence, the ".gp command (which adds extra characters into readbuf1
-- to emulate leaving the cursor after the text by moving the cursor
-- after inserting the text) would insert the motion characters into
-- the buffer instead of using them to move after the insert has been
-- done.
-- I could probably get this working properly with a special flag put
-- into start_redo_ins() and set in do_put(), but I think this adds
-- much more complexity than fixing this bug justifies.
pending('should not change the ". register with ".2p', function()
local orig_register = funcs.getreg('.')
feed('2".p')
eq(orig_register, funcs.getreg('.'))
end)
describe("cursor positioning after undo and redo with '.'", function()
before_each(reset)
local function make_cursor_test(macro_string)
return function()
feed(macro_string)
local afterpos = funcs.getcurpos()
local orig_string = curbuf_contents()
feed('u.')
eq(afterpos, funcs.getcurpos())
expect(orig_string)
end
end
-- The difficulty here is: setting the cursor after the end of the
-- pasted text is done by adding a motion command to the
-- stuffbuffer after the insert.
-- Modifying 'redobuff' is done in the code that handles inserting
-- text and moving around.
-- I could add a special case in ins_esc() that checks for a flag
-- set in do_put() to add the motion character to the redo buffer,
-- but I think that is starting to get way too convoluted for the
-- benefit.
pending('should be the same after ".gp and ".gpu.',
make_cursor_test('".gp'))
-- The difficulty here is: putting forwards is implemented by using
-- 'a' instead of 'i' to start insert.
-- Undoing with 'u' an insert that began with 'a' leaves the cursor
-- where the first character was inserted, not where the cursor was
-- when the 'a' was pressed.
-- We account for this the first time by saving the cursor position
-- in do_put(), but this isn't stored in redobuff for a second time
-- around.
-- We can't change how such a fundamental action as undo after
-- inserting with 'a' behaves, we could add in a special case
-- whereby we set a flag in do_put() and read it when entering
-- insert mode but this seems like way too much to fix such a minor
-- bug.
pending('should be the same after ".pu. and ".pu.u.',
make_cursor_test('".pu.'))
end)
end)
Jan Edmund Lazo