This reverts commit 3a5585df95.
This reverts a change that removed a lock. It's indeed true that in master,
fish script is bound to the main thread. But I'm working to remove that
limitation and these locks are important in that future.
When switching to the new ast, commands that were not decorated
statements (like function declarations) would be rejected from
autosuggestion validation because we could not find a command. Stop
rejecting them.
The owning locks were added after the original code and decorated with
comments indicating they are thread-safe, even though they're only ever
used from the main thread. Presuming the intent was to make future
manipulation of the code safer rather than to actually make use of any
thread safety guarantees, these have been wrapped in a new
`thread_exclusive` type which always calls ASSERT_IS_MAIN_THREAD.
The benefit is that this does not perform a syscall to lock a mutex
each time the variables are accessed.
When executed interactively and not piped, `functions` adds a comma as a
separator between each result. This removes the separator after the last
item.
highlight.cpp was blindly calling path_get_path for each head command
typed at the prompt which triggers a lot of syscalls via waccess.
It's still going to do that while commands are being composed, but now
it won't if we can make a cheap lookup to the builtins/functions hash
table and can determine that it's a valid command before inspecting the
filesystem.
When fish receives a "cancellation inducing" signal (SIGINT in particular)
it has to unwind execution - for example while loops or whatever else that
is executing. There are two ways this may come about:
1. The fish process received the signal
2. A child process received the signal
An example of the second case is:
some_command | some_function
Here `some_command` is the tty owner and so will receive control-C, but
then fish has to cancel function execution.
Prior to this change, these were handled uniformly: both would just set a
cancellation signal inside the parser. However in the future we will have
multiple parsers and it may not be obvious which one to set the flag in.
So instead distinguish these cases: if a process receives SIGINT we mark
the signal in its job group, and if fish receives it we set a global
variable.
We were previously aborting the main event loop before calling fish_exit
in the event of a SIGHUP. This patch causes the SIGHUP to be stored in a
separate state variable from a regular "must exit" condition so the
associated event can be fired before we terminate the loop.
All streams are redirected before the event is called to prevent a
SIGTTIN/SIGTTOU due to the user script reading/writing from a disposed
tty.
Closes#7014
This error only happens on recent versions of gcc, see previous
commit e6bb7fc973 for more info.
Instead of using `ignore_result()` here, I've added a `system_assert`
function/macro that mimics the behavior of all the other `system()`
calls in the file.
Previously fish weak-linked wcsdup, wcsncasecmp, and wcscasecmp.
This enabled fish to be used on 10.6. However the minimum Mac version
is now 10.9, where these symbols are available.
This reworks the "a=" detection to be simpler.
If we detect a variable assignment that produces an error,
simply consume it.
We also take the opportunity to not highlight it as an error,
and add some tests.
Original commit is 1ca05d32d3.
Typing that command in an interactive prompt would make the highlighter thread
eat up CPU and memory. Probably not the right fix; I think the token should
already have been consumed when the error is detected, then there is no need
to consume it when unwinding.
When selectiong a large completion entry in the pager, it would clobber the
prompt. To reproduce, first run this command
complete -c : -xa '(
# completion entries that, when applied to the commandline
# need one, two, or three lines respectively
echo 1
echo 2(string repeat -n (math $COLUMNS - 5) x)
echo 3(string repeat -n $COLUMNS x)
printf %s\n n(seq $LINES)
)'
then type ": " and hit Tab repeatedly. When cycling through completion
entries, observe that fish always tries to render the pager with the same
size, even though the number of lines occupied by the command line buffer
changes due to soft wrapping.
Fix this by rendering the pager after the command line has been rendered, so
we know how many lines we have left.
This switches fish_indent from parsing with parse_tree
to the new ast.
This is the most difficult transition because the new ast retains less
lexical information than the old parse tree. The strategy is:
1. Use parse_util_compute_indents to compute indenting for each token.
2. Compute the "gap text" between the text of significant tokens. This
contains whitespace, comments, etc.
3. "Fix up" the gap text while leaving the significant tokens alone.
This switches parse_util_compute_indents from parsing with parse_tree to
the new ast.
It also reworks the parse_util_compute_indents tests, because
parse_util_compute_indents will be the backing for fish_indent.
This is the first commit of a series intended to replace the existing
"parse tree" machinery. It adds a new abstract syntax tree and uses a more
normal recursive descent parser.
Initially there are no users of the new ast. The following commits will
replace parse_tree -> ast for all usages.
Prior to this commit, fish used NUL ('\0') to disable control
functions (for example, the function that generates SIGTSTP).
However NUL may in fact be bindable and is on macOS via
control-space.
Use instead _POSIX_VDISABLE if defined and not -1.
This was always awkward as fish script, and had problems with
interrupting the autoloading.
Note that we still leave the old function intact to facilitate easier
upgrading for now.
Fixes#7145.
Page-Down seems to deactivate history search, so trying to undo
would leave the command line in an inconsistent state.
Fixes#7162 which was introduced in
12a9cb29 Fix assertion failure on page up / page down
It is used exclusively as vector at the moment since we only ever append
at the end. Making it a deque would be useful when allowing to edit the
search string and subsequently resume the search at an arbitrary position
in the history.
When editing a multiline command line and pressing "up" with the cursor at the
first line, fish attempts a hsitory search. If the search fails, don't move
the cursor to the end of the multiline command because this can be annoying
when the user does not actually want to perform a history search.
After profiling bottlenecks in job execution, the calls to `tcgetpgrp`
were identified to take a good amount of the execution time. Collecting
metrics on which branches were taken revealed that in all "normal"
cases, there is no benefit to calling `tcgetpgrp` before calling
`tcsetpgrp` as it can instead be called only in the error case to
determine what sort of error handling behavior should be applied.
This makes the best-case scenario of a single syscall much more likely
than in the previous situation.
Profiling revealed string comparison in variable lookups to be a
significant hotspot. This change causes `make test` to complete ~4.5%
faster per `hyperfine`.
This was originally comparing two pointers for equality but after the
refactor to wcstring it ended up comparing a const string pointer to the
_contents_ of the wcstring.
I really kinda hate how insistent clang-format is to have line
breaks *IFF THE LINE IS TOO LONG*.
Like... lemme just add a break if it looks better, will you?
But it is the style at this time, so we shall tie an onion to our
belt.
This makes binding \cz possible.
We already ignore the SIGTSTP signal it sends, so until now it was useless.
(also STOP and START for good measure, but since we disable flow
control in fish anyway these already shouldn't have been sent)
Fixes#7152
Profiling shows that parsing color names still took much longer than it
should.
wcscasecmp is so slow that using it directly causes `try_parse_special`
to consume up to 3% of all of fish's cpu time due to extremely
inefficient invariant case lookups for wide characters (tested: Fedora
Server 32 w/ glibc 2.31 with -O2).
These control the disambiguation between ctrl-j and ctrl-m.
This can cause the enter key to send a ctrl-m, which programs might be
unprepared for.
(This is why you need to do `stty sane<ctrl-j>`)
The `reserve` here can, under certain circumstances, reserve more than
strictly needed.
The simple workaround is to just never look at more than we feed in.
(really what we'd *want* is to look at the length of the *color
names*, but those are wchar, so length lookup is crappy NULL-lookup)
This was profiled to be a hotspot during startup. The usage of
wcscasecmp in a loop caused repeated transforms to lowercase, which is
incredibly slow for wide characters.
This makes it possible to expand the signals checked by the type. I can't merge
the sigttin fixes for #7060 yet because they introduce new breakage, but this
will make merging any future fix easier.
In #7133, neovim crashing caused "OPOST" to be turned off, which
caused a weird staircase display.
So we just force a set of settings that don't seem useful to change to
avoid breaking the terminal with something like that.
Fixes#7133.
A broken/missing optspec or `--` is a bug in the script using
argparse, an unknown option or invalid argument is a bug in using that script.
So in the former case print a stacktrace, because the person writing
the `argparse` call is at fault, in the latter don't.
Fixes#6703.
Prior to this change, the posix_spawn code paths used a fair amount of
manual management around its allocated structures (attrs and file actions).
Encapsulate this into a new class that manages memory management and error
handling.
Finish the transition to termsize.h. Remove the scary termsize bits
from common.cpp, which can throw off events at arbitrary calls and are
dangerously reentrant. Migrate everyone to the new termsize.h.
fish's handling of terminal sizes is currently rather twisted. The
essential problem is that the terminal size may change at any point from a
SIGWINCH, and common_get_{width,height} may modify it and post variable
change events from arbitrary locations.
Tighten up the semantics. Assign responsibility for managing the tty size
to a new class, `termsize_container_t`. Rationalize locking and reentrancy.
Explicitly nail down the relationship between $COLUMNS/$LINES and the tty
size. The new semantics are: whatever changed most recently takes
precendence.
Prior to this fix, fish would attempt to resize the terminal via
TIOCSWINSZ, which was added as part of #3740. In practice this probably
never did anything useful since generally only the tty master can use
this. Remove the support and note it in the changelog.
Prior to this fix, s_reset would attempt to reset the screen, optionally
using the PROMPT_SP hack to go to the next line. This in turn required
passing in the screen width even if it wasn't needed (because we were
not going to abandon the line). Factor this into two functions:
- s_reset_line which does not apply the hack
- s_reset_abandoning_line which applies the PROMPT_SP hack
common_get_width will "lazily" decide the screen width, which means
changing the environment variable stack. This is a surprising thing
to do from the middle of screen rendering.
Switch to passing in widths explicitly to screen.
