Prior to this fix, if you attempt to complete from inside a quote and the
completion contained an entity that cannot be represented inside quotes
(i.e. \n \r \t \b), the result would be a broken mess of quotes. Rewrite
the implementation so that it exits the quotes, emits the correct unquoted
escape, and then re-enters the quotes.
Properly escape literal tildes in tab completion results. Currently we
always escape tildes in unquoted arguments; in the future we may escape
only leading tildes.
Fixes#2274
Prior to this fix, autoloads like function and completion autoloads
would check their path variable (like fish_function_path) on every
autoload request. Switch to invalidating it in response to the variable
changing.
This improves time on a microbenchmark:
for i in (seq 50000)
setenv test_env val$i
end
from ~11 seconds to ~6.5 seconds.
The job control functions were a bit messy, in particular
`set_child_group`'s name would imply that all it does is set the child
group, but in reality it used to set the child group (via `setpgid`),
set the job's pgrp if it hasn't been set, and possibly assign control of
the terminal to the newly-created job.
These have been split into separate functions. Now `set_child_group`
does just (and only) that, `maybe_assign_terminal` might assign the
terminal to the new pgrp, and `on_process_created` is used to set the
job properties the first time an external process is created. This might
also speed things up (but probably not noticeably) as there are no more
repeated calls to `getpgrp()` if JOB_CONTROL is not set.
Additionally, this closes#4715 by no longer unconditionally calling
`setpgid` on all new processes, including those created by `posix_spawn`
which does not need this since the child's pgrep is set at in the
arguments to that API call.
This switches function execution from the function's source code to
its stored node and pstree. This means we no longer have to re-parse
the function every time we execute it.
The idea is that we can return the shared pointer directly, avoiding
lots of annoying little getter functions that each need to take locks.
It also helps to pull together the data structures used to initialize
functions versus store them.
This concerns block nodes with redirections, like
begin ... end | grep ...
Prior to this fix, we passed in a pointer to the node. Switch to passing
in the tnode and parsed source ref. This improves type safety and better
aligns with the function-node plans.
Prior to this fix, functions stored a string representation of their
contents. Switch them to storing a parsed source reference and the
tnode of the contents. This is part of an effort to avoid reparsing
a function's contents every time it executes.
Add a fish-specific wrapper around std::mutex that records whether it is
locked in a bool. This is to make ASSERT_IS_LOCKED() simpler (it can just
check the boolean instead of relying on try_lock) which will make Coverity
Scan happier.
Some details: Coverity Scan was complaining about an apparent double-unlock
because it's unaware of the semantics of try_lock(). Specifically fish
asserts that a lock is locked by asserting that try_lock fails; if it
succeeds fish prints an error and then unlocks the lock (so as not to leave
it locked). This unlock is of course correct, but it confused Coverity Scan.
Use wcstring/string instead of a character array. The variable
`term_env` was not being freed before the function exited.
Fixes defect 7520324 in coverity scan.
keepalive processes are typically killed by the main shell process.
However if the main shell exits the keepalive may linger. In WSL
keepalives are used more often, and the lingering keepalives are both
leaks and prevent the tests from finishing.
Have keepalives poll for their parent process ID and exit when it
changes, so they can clean themselves up. The polling frequency can be
low.
Have WSL use a keepalive whenever the first process is external.
This works around the fact that WSL prohibits setting an exited
process as the group leader.
When the pager wants to use the full screen to show many options, it reserves
space at the top to see the command. Previously it pretended the command was a
prompt and engaged the prompt layout mechanism to compute these lines. Instead
let's juts count newlines since escape sequences within commands are very rare.
There were several issues with the way that the include tests for curses.h
were being done that were ultimately causing fish to use the headers from
ncurses but link against curses on platforms that provide an actual
libcurses.so that isn't just a symlink to libncurses.so
In particular, the old code was first testing for curses's cureses.h and then
falling back to libncurses's implementation of the same - but that logic was
reversed when it came to including term.h, in which case it was testing for
the ncurses term.h and falling back to the curses.h header. Long story short,
while cmake will link against libcurses.so if both libcurses.so and
libncurses.so are present (unless CURSES_NEED_NCURSES evaluates to TRUE, but
that makes ncurses a hard requirement), but we were brining in some of the
defines from the ncurses headers, causing SIGSEGV panics when fish ultimately
tried to access variables that weren't exported or were mapped to undefined
areas of memory in the other library.
Additionally it is an error to include termios.h prior to including the plain
Jane curses.h (not ncurses/curses.h), causing errors about unimplemented types
SGTTY/chtype. So far as I can tell, both curses.h and ncurses/curses.h pull in
termios.h themselves so it shouldn't even be necessary to manually include it,
but I have just moved its #include below that of curses.h
This special cases expansion of $history variables, so that slicing
history no longer needs to construct the entire history array. Speedup
is around 100x in my test.
Fixes#4650
Prior to this fix, if the user typed normal characters while the
completion pager was shown, it would begin searching. This feature was
not well liked, so we are going to instead just append the characters as
normal and disable paging. Control-S can be used to toggle the search
field.
Fixes#2249
try_get_child() was taking the address of a reference; clang was thereby
assuming it could not be null and so was dropping the null check. Ensure
we do not dereference a null pointer.
Fixes#4678
This was a symbol that represented either an argument or a redirection.
This was only used as part of argument_or_redirection_list.
It's simpler to just have these types be alternatives in the list type.
Some dotfile users like to add directories to PATH that point at
non-existent directories (because those directories exist on other
machines). Stop warning in that case, unless those directories contain
a colon, in which case it's probably a user error.
Changed cd completion to differentiate between cd autosuggest and cd tab
completion. cd autosuggest will find deepest unique hierarchy and cd tab
completion will not.
Issue #4402
This untangles the CMake versioning issues (I hope) as discussed in #4626.
Note most of the advice found on the Internet about how to inject git
versions into CMake is just wrong.
The behavior we want is to unconditionally run the script
build_tools/git_version_gen.sh at build time (i.e. when you invoke ninja or
make, and not when you invoke cmake, which is build system generation time).
This script is careful to only update the FISH-BUILD-VERSION-FILE if the
contents have changed, to avoid spurious rebuilding dependencies of
FISH-BUILD-VERSION-FILE. Assuming the git version hasn't changed, the script
will run, but not update FISH-BUILD-VERSION-FILE, and therefore
fish_version.o will not have to be rebuilt.
This might normally rebuild more than is necessary even if the timestamp is
not updated, because ninja computes the dependency chain ahead of time. But
Ninja also supports the 'restat' option for just this case, and CMake is rad
and exposes this via BYPRODUCTS. So mark FISH-BUILD-VERSION-FILE as a
byproduct and make the script always update a dummy file
(fish-build-version-witness.txt). Note this is the use case for which
BYPRODUCTS is designed.
We also have fish_version.cpp #include "FISH-BUILD-VERSION-FILE", and do a
semi-silly thing and make FISH-BUILD-VERSION-FILE valid C++ (so there's just
one version file). This means we have to filter out the quotes in other
cases..
This reverts commit 25839b8c36.
This was an attempt to simplify the version generation, but it
computed the version at build sytem generation time rather than
at build time, requiring another run of CMake to update it.
Correctly generate FISH_BUILD_VERSION for use in fish_version.h/cpp and
fish.pc to allow `fish --version` and `echo $version` to work again.
Not needing the same convoluted measures used by Makefile builds to
prevent the regeneration of the fish version file when it hasn't
changed.
Purposely created a new `cmake_git_version_gen.sh` file so that the old
`git_version_gen.sh` remains compatible with the existing Makefile build
script. Same reason why `fish.pc.in` was not modified to use a lowercase
variable name to match the CMAKE variable of the same name.
Closes#4626
A large portion of time was spent constructing strings and passing
them to debug(). Turn debug into a macro so that the strings are only
constructed if they're going to be printed.
This adds a new class arg_iterator_t which encapsulates decisions about
whether to read arguments from stdin or argv. It also migrates the
unread bytes buffer from a static variable to an instance variable.
Profiling with callgrind revealed that about 60% of the time in a `something | string match` call
was actually spent in `string_get_arg_stdin()`,
because it was calling `read` one byte at a time.
This makes it read in chunks similar to builtin read.
This increases performance for `getent hosts | string match -v '0.0.0.0*'` from about 300ms to about 30ms (i.e. 90%).
At that point it's _actually_ quicker than `grep`.
To improve performance even more, we'd have to cut down on str2wcstring.
Fixes#4604.
Prior to this fix, a "bare variable" in math like 'x + 1' would be
looked up in the environment, i.e. equivalent to '$x + 1'. This appears
to have been done for performance. However this breaks the orthogonality
of fish; performance is not a sufficient justification to give math this
level of built-in power, especially because the performance of math is
not a bottleneck. The implementation is also ugly.
Remove this feature so that variables must be prefixed with the dollar
sign and undergo normal variable expansion. Reading 'git grep' output
does not show any uses of this in fish functions or completions.
Also added to changelog.
Fixes#4393
For some reason on Solaris the previous code was refusing to compile
with an error (regarding the declaration of stdout in the opts struct)
error: declaration of ‘__iob’ as array of references
The obvious guess that it had something to do with the name of the
variable in question proved true; renaming it from `stdout` to
`opts.stdout` allows the build to go through.
macOS 10.5 and earlier do not support the convention of returning
a dynamically allocated string, plus this seems like an unnecessary
malloc. Always allocate a buffer for realpath() to write into.
It seems that `parse_cmd_opts` does not correctly handle no arguments,
and so argc was being decremented to -1 causing uninitialized memory
access when argv[0] was dereferenced at a later point.
No longer using `-` to indicate reading to stdout. Use lack of arguments
as stdout indicator. This prevents mixing of variables with stdout
reading and makes it clear that stdout may not be mixed with delimiters
or array mode.
Added an option to read to stdout via `read -`. While it may seem
useless at first blush, it lets you do things like include
mysql -p(read --silent) ...
Without needing to save to a local variable and then echo it back.
Kicks in when `-` is provided as the variable name to read to. This is
in keeping with the de facto syntax for reading/writing from/to
stdin/stdout instead of a file in, e.g., tar, cat, and other standard
unix utilities.
This silences binding errors due to keys not found in the current
termcap config in the default fish bindings.
Closes#4188, #4431, and obviates the original fix for #1155
It was necessary to re-implement builtin_bind as a class in order to
avoid passing around the options array from function to function and
as adding an opts parameter to `get_terminfo_sequence` would require
otps to be passed to all other builtin_bind_ functions so they could, in
turn, pass it to `get_terminfo_sequence`.
On powerpc64 (big-endian platform) one test failed as:
Testing file printf.in ... fail
Output differs for file printf.in. Diff follows:
--- printf.tmp.out 2017-10-02 18:14:17.740000000 -0700
+++ printf.out 2017-10-02 18:11:59.370000000 -0700
@@ -1,5 +1,5 @@
Hello 1 2 3.000000 4.000000 5 6
-a B 0 18446744073709551615
+a B 10 18446744073709551615
It happens due to roughly the following code:
swprintf(..., L"%o", (long long)8);
Here mismatch happens between "%o" (requires 32-bit value)
and 'long long' (requires 64-bit value).
The fix turns it effectively to:
swprintf(..., L"%llo", (long long)8);
as it was previously done for 'x', 'd' and other int-like types.
Makes tests pass on powerpc64.
Signed-off-by: Sergei Trofimovich <slyfox@gentoo.org>
The POSIX standard specifies that a buffer should be supplied to
getcwd(), not doing so is undefined (or rather, platform-defined)
behavior. This was causing the getcwd errors on illumos (though not seen
on Solaris 11) reported in #3340Closes#3340
Took care of remaining issues preventing fish from building on Solaris.
Mainly caused by some assumptions that certain defines are POSIX when
they are not (`NAME_MAX`).
Moved `NAME_MAX` defines to common.h - for some reason, it was being
defined in a cpp file (`env_universal_common.cpp`) even though it is used
in multiple source files.
Now compiles on Solaris 11 with GNU Make. Still some warnings because
fish was written with GNU getopt in mind and the Solaris version doesn't
use `const char *` but rather just `char *` for getopt values, but it
builds nevertheless.
Assuming this closes#3340
We had pid_status defined as a pid_t instance, which was fine since on
most platforms pid_t is an alias for int. However, that is not
universally the case and waitpid takes an int *, not a pid_t *.
A completion may have zero length; in this case the length of the
prefix was omitted and the completion was not visible. Correct the
calculation to account for zero-width completions.
Fixes#4424
A recent discussion involving whether `can_be_encoded()` was broken
caused me to notice that we are inconsistent about whether Unicode code
points are specified using `\xXXXX` or `\uXXXX` notation. Which is
harmless but silly and potentially confusing.
This flag was only documented for a few weeks before being renamed
`--show-time` and has been deprecated for a long time. Fish 3.0 is a good
opportunity to remove it.
Instead of treating the search term as a literal string to be matched
treat it as a glob. This allows the user to get a more useful set of
results by using the `*` glob character in the search term.
Partial fix for #3136
* Implement `history search --reverse`
It should be possible to have `history search` output ordered oldest to
newest like nearly every other shell including bash, ksh, zsh, and csh.
We can't make this the default because too many people expect the
current behavior. This simply makes it possible for people to define
their own abbreviations or functions that provide behavior they are
likely used to if they are transitioning to fish from another shell.
This also fixes a bug in the `history` function with respect to how it
handles the `-n` / `--max` flag.
Fixes#4354
* Fix comment for format_history_record()
As discussed in #3805, this patch disables assigning fish to its own
process group at startup. This was trialled in #4349 alongside other
pgrp fixes which introduced additional problems, but this particular fix
seems to be OK.
Fixes#3805 and works around Microsoft/BashOnWindows#1653
* Hoist `for` loop control var to enclosing scope
It should be possible to reference the last value assigned to a `for`
loop control var when the loop terminates. This makes it easier to detect
if we broke out of the loop among other things. This change makes fish
`for` loops behave like most other shells.
Fixes#1935
* Remove redundant line
The type cached_esc_sequences_t caches escape sequences, and is tasked
with finding an escape sequence that prefixes a given string. Before
this fix, it did so by storing the lengths of cached escape sequences,
and searching for substrings of that length. The new implementation
instead stores all cached escape sequences in a sorted vector, and uses
binary search to find the shortest escape sequence that is a prefix of
the input. This is a substantial simplification that also reduces
allocations.
This eliminates the "missing" notion of env_var_t. Instead
env_get returns a maybe_t<env_var_t>, which forces callers to
handle the possibility that the variable is missing.
maybe_t is an implementation of the Maybe/Optional type, allowing
for an optional value to be stored. This will enable a more
principled approach for functions that return values or failure,
such as env_get.
This commit backs out certain optimizations around setting environment
variables, and replaces them with move semantics. env_set accepts a
list, by value, permitting callers to use std::move to transfer
ownership.
Commit f872f25f introduced a freed memory access regression on line 460
of env.cpp, where an environment variable was converted to a temporary
string, the .c_str() address of which was stored while the string
temporary was destroyed.
This commit keeps a reference to the original string lying around so
that the c_str() pointer does not point to freed memory.
Valgrind warns that the sometimes uninitialized sigaction.sa_flags field
is sometimes used when passed to the signal handler.
This patch explicitly zeros out the sigaction.sa_flags field at creation
time.
cherry-picked from krader1961/fish-shell commit b69df4fe72
Fixes#4353 (regression in indexing of history contents) and introduces
new unit tests to catch bad $history indexing in the future.
This implements an LRU cache of recently seen math expressions. When
executing math inside loops and the like this can provide a 33% decrease
in the time to execute the `math` command.
We need our `math` builtin to behave like `bc` with respect to rounding
floating point values to integer to avoid breaking to many existing
uses. So when scale is zero round down to the nearest integer.
Another change for #3157.
The MuParser supports the concept of multiple expressions separated by
commas. This implements support for that so that you can do things like
this:
set results (math '1+1, 4*2, 9^2')
This is the second baby step in resolving #3157. Implement a bare minimum
builtin `math` command. This is solely to ensure that fish can be built
and run in the Travis build environments. This is okay since anyone running
`builtin math` today is already getting an error response.
Also, more work is needed to support bare var references, multiple result
values, etc.
Using a read-only variable like `status` as a for loop control variable
has never worked. But without this change you simply get non-sensical
behavior that leaves you scratching your head in puzzlement. This change
replaces the non-sensical behavior with an explicit error message.
Fixes#4342
Recent changes to switch to unordered sets/maps can cause the order in
which items are returned to be non-deterministic. This change ensures
that the argparse "Mutually exclusive flags" error message to be
deterministic with respect to the order of the interpolated values.
Make setting fish vars more efficient by avoiding creating a
wcstring_list_t for the case where we're setting one value. For the case
where we're passing a list of values swap it with the list in the var
rather than copying it. This makes the benchmark in #4200 approximately
6% faster.
Since we are including XXHash32/64 anyway for the wchar_t* hashing,
we might as well use it.
Use arch-specific hash size and xxhash for all wcstring hashing
Instead of using XXHash64 for all platforms, use the 32-bit version
when running on 32-bit platforms where XXHash64 is significantly slower
than XXHash32 (and the additional precision will not be used).
Additionally, manually specify wcstring_hash as hashing method for
non-const wcstring unordered_set/map instances (the const varieties
don't have an in-library hash and so already use our xxhash-based
specialization when calling std::hash<const wcstring>).
commit 50f414a45d58fcab664ff662dd27befcfa0fdd95
Author: Mahmoud Al-Qudsi <mqudsi@neosmart.net>
Date: Sat Aug 19 13:43:35 2017 -0500
Converted file_id_t set to unordered_set with custom hash
commit 83ef2dd7cc1bc3e4fdf0b2d3546d6811326cc3c9
Author: Mahmoud Al-Qudsi <mqudsi@neosmart.net>
Date: Sat Aug 19 13:43:14 2017 -0500
Converted remaining set<wcstring> to unordered_set<wcstring>
commit 053da88f933f27505b3cf4810402e2a2be070203
Author: Mahmoud Al-Qudsi <mqudsi@neosmart.net>
Date: Sat Aug 19 13:29:21 2017 -0500
Switched function sets to unordered_set
commit d469742a14ac99599022a9258cda8255178826b5
Author: Mahmoud Al-Qudsi <mqudsi@neosmart.net>
Date: Sat Aug 19 13:21:32 2017 -0500
Converted list of modified variables to an unordered set
commit 5c06f866beeafb23878b1a932c7cd2558412c283
Author: Mahmoud Al-Qudsi <mqudsi@neosmart.net>
Date: Sat Aug 19 13:15:20 2017 -0500
Convert const_string_set_t to std::unordered_set
As it is a readonly-list of raw character pointer strings (not
wcstring), this necessitated the addition of a hashing function since
the C++ standard library does not come with a char pointer hash
function.
To that end, a zlib-licensed [0] port of the excellent, lightweight
XXHash family of 32- and 64-bit hashing algorithms in the form of a C++
header-only include library has been included. XXHash32/64 is pretty
much universally the fastest hashing library for general purpose
applications, and has been thoroughly vetted and is used in countless
open source projects. The single-header version of this library makes it
a lot simpler to include in the fish project, and the license
compatibility with fish' GPLv2 and the zero-lib nature should make it an
easy decision.
std::unordered_set brings a massive speedup as compared to the default
std::set, and the further use of the fast XXHash library to provide the
string hashing should make all forms of string lookups in fish
significantly faster (to a user-noticeable extent).
0: http://create.stephan-brumme.com/about.html
commit 30d7710be8f0c23a4d42f7e713fcb7850f99036e
Author: Mahmoud Al-Qudsi <mqudsi@neosmart.net>
Date: Sat Aug 19 12:29:39 2017 -0500
Using std::unordered_set for completions backing store
While the completions shown to the user are sorted, their storage in
memory does not need to be since they are re-sorted before they are
shown in completions.cpp.
commit 695e83331d7a60ba188e57f6ea0d9b6da54860c6
Author: Mahmoud Al-Qudsi <mqudsi@neosmart.net>
Date: Sat Aug 19 12:06:53 2017 -0500
Updated is_loading to use unordered_set
No longer using RAII wrappers around pthread_mutex_t and pthread_cond_t
in favor of the C++11 std::mutex, std::recursive_mutex, and
std::condition_variable data types.
The `react_to_variable_change()` function is called whenever a fish var
is set. Even as a consequence of statements like `for x in a b c`. It is
therefore critical that that function be as fast as possible. Especially
when setting the var doesn't have any side-effects which is true something
like 99.9999% of the time.
This change reduces the overhead of `react_to_variable_change()` to
unmeasurable levels. Making the synthetic benchmark in issue #4341
36% faster.
Fixes#4341
Internally fish should store vars as a vector of elements. The current
flat string representation is a holdover from when the code was written
in C.
Fixes#4200
A semi-empty var is one with a single empty string element. The
`env_var_t::empty()` method returns true for such vars but we want
`set --show` to report that it has a single empty element.
Newer versions of GCC and Clang are not satisfied by a cast to void,
this fix is adapted from glibc's solution.
New wrapper function ignore_result should be used when a function with
explicit _unused_attribute_ wrapper is called whose result will not be
handled.
Make the `env_var_t::missing_var()` object a singleton rather than a
dynamically constructed object. This requires some discipline in its use
since C++ doesn't directly support immutable objects. But it is slightly
more efficient and helps identify code that incorrectly mutates `env_var_t`
objects that should not be modified.
It's bugged me forever that the scope is the second arg to `env_get()`
but not `env_set()`. And since I'll be introducing some helper functions
that wrap `env_set()` now is a good time to change the order of its
arguments.
My previous change to eliminate `class var_entry_t` caused me to notice
that `env_get()` turned a set but empty var into a missing var. Which
is wrong. Fixing that brought to light several other pieces of code that
were wrong as a consequence of the aforementioned bug.
Another step to fixing issue #4200.
* Fix clearing abandoned line with VTE
With VTE-based terminals, resizing currently causes multi-line prompts
to go weird.
This changes the sequence we use to clear the line to one suggested by
a VTE
developer (https://bugzilla.gnome.org/show_bug.cgi?id=763390#c4).
It changes nothing in konsole 17.04.3 and urxvt 9.22, but they already
work.
Note that this does not fix the case where output did not end in a
newline, but that doesn't seem to be up to us. Also, it only affects
those lines.
Fixes#2320.
* Use terminfo definition instead of hardcoding
Thanks to @ixjlyons.
Doing `set -U var` when a global named `var` exists can result in
confusing behavior. Try to limit the confusion by improving the warning
we write. Also, only write the warning if interactive.
Fixes#4267
