POSIX does not require the shell to fork for a subshell environment, and we
use that possibility in various ways (command substitutions with a single
command and most subshells that are the final command of a shell process).
Therefore do not tie subshells to forking in the man page.
Command substitutions with expansions are a bit strange, causing a fork for
$(...$(($x))...) because $x might expand to y=2; they will probably be
changed later but this is how they work now.
These already worked: $# ${#} ${##} ${#-} ${#?}
These now work as well: ${#+word} ${#-word} ${##word} ${#%word}
There is an ambiguity in the standard with ${#?}: it could be the length of
$? or it could be $# giving an error in the (impossible) case that it is not
set. We continue to use the former interpretation as it seems more useful.
New features:
* proper lazy evaluation of || and &&
* ?: ternary operator
* executable is considerably smaller (8K on i386) because lex and yacc are
no longer used
Differences from dash:
* arith_t instead of intmax_t
* imaxdiv() not used
* unset or null variables default to 0
* let/exp builtin (undocumented, will probably be removed later)
Obtained from: dash
* In {(...) <redir1;} <redir2, do not drop redir1.
* Maintain the difference between (...) <redir and {(...)} <redir:
In (...) <redir, the redirection is performed in the child, while in
{(...)} <redir it should be performed in the parent (like {(...); :;}
<redir)
POSIX requires this and it is simpler than the previous code that remembered
command locations when appending directories to PATH.
In particular,
PATH=$PATH
is no longer a no-op but discards all cached command locations.
If execve() returns an [ENOEXEC] error, check if the file is binary before
trying to execute it using sh. A file is considered binary if at least one
of the first 256 bytes is '\0'.
In particular, trying to execute ELF binaries for the wrong architecture now
fails with an "Exec format error" message instead of syntax errors and
potentially strange results.
These are called "shell procedures" in the source.
If execve() failed with [ENOEXEC], the shell would reinitialize itself
and execute the program as a script. This requires a fair amount of code
which is not frequently used (most scripts have a #! magic number).
Therefore just execute a new instance of sh (_PATH_BSHELL) to run the
script.
This matches the constants from <signal.h> with 'SIG' removed, which POSIX
requires kill and trap to accept and 'kill -l' to write.
'kill -l', 'trap', 'trap -l' output is now upper case.
In Turkish locales, signal names with an upper case 'I' are now accepted,
while signal names with a lower case 'i' are no longer accepted, and the
output of 'killall -l' now contains proper capital 'I' without dot instead
of a dotted capital 'I'.
* There is no plan for an alternative to the command "set".
* Attempting to unset a readonly variable has not raised an error for quite
a while, so the order of unsetting a variable and a function with the same
name does not matter.
MFC after: 1 week
When a foreground job exits on a signal, a message is printed to stdout
about this. The buffer was not flushed after this which could result in the
message being written to the wrong file if the next command was a builtin
and had stdout redirected.
Example:
sh -c 'kill -9 $$'; : > foo; echo FOO:; cat foo
Reported by: gcooper
MFC after: 1 week
This is useful so that it is easier to exit on a signal than to reset the
trap to default and resend the signal. It matches ksh93. POSIX says that
'exit' without args from a trap action uses the exit status from the last
command before the trap, which is different from 'exit $?' and matches this
if the previous command is assumed to have exited on the signal.
If the signal is SIGSTOP, SIGTSTP, SIGTTIN or SIGTTOU, or if the default
action for the signal is to ignore it, a normal _exit(2) is done with exit
status 128+signal_number.
* Make 'trap --' do the same as 'trap' instead of nothing.
* Make '--' stop option processing (note that '-' action is not an option).
Side effect: The error message for an unknown option is different.
All builtins are now always found before a PATH search.
Most ash derivatives have an undocumented feature where the presence of an
entry "%builtin" in $PATH will cause builtins to be checked at that point of
the PATH search, rather than before looking at any directories as documented
in the man page (very old versions do document this feature).
I am removing this feature from sh, as it complicates the code, may violate
expectations (for example, /usr/bin/alias is very close to a forkbomb with
PATH=/usr/bin:%builtin, only /usr/bin/builtin not being another link saves
it) and appears to be unused (all the %builtin google code search finds is
in some sort of ash source code).
Note that aliases and functions took and take precedence above builtins.
Because aliases work on a lexical level they can only ever be overridden on
a lexical level (quoting or preceding 'builtin' or 'command'). Allowing
override of functions via PATH does not really fit in the model of sh and it
would work differently from %builtin if implemented.
Note: POSIX says special builtins are found before functions. We comply to
this because we do not allow functions with the same name as a special
builtin.
Silence from: freebsd-hackers@ (message sent 20101225)
Discussed with: dougb
It should use the original exit status, just like falling off the
end of the trap handler.
Outside an EXIT trap, 'exit' is still equivalent to 'exit $?'.
An error message is written, the builtin is not executed, nonzero exit
status is returned but the shell does not abort.
This was already checked for special builtins and external commands, with
the same consequences except that the shell aborts for special builtins.
Obtained from: NetBSD
Change the criterion for builtins to be safe to execute in the same process
in optimized command substitution from a blacklist of only cd, . and eval to
a whitelist.
This avoids clobbering the main shell environment such as by $(exit 4) and
$(set -x).
The builtins jobid, jobs, times and trap can still show information not
available in a child process; this is deliberately permitted. (Changing
traps is not.)
For some builtins, whether they are safe depends on the arguments passed to
them. Some of these are always considered unsafe to keep things simple; this
only harms efficiency a little in the rare case they are used alone in a
command substitution.
If SIGINT arrived at exactly the right moment (unlikely), an exception
handler in a no longer active stack frame would be called.
Because the old handler was not used in the normal path, clang thought it
was a dead value and if an exception happened it would longjmp() to garbage.
This caused builtins/fc1.0 to fail if histedit.c was compiled with clang.
MFC after: 1 week
Before considering to execute a command substitution in the same process,
check if any of the expansions may have a side effect; if so, execute it in
a new process just like happens if it is not a single simple command.
Although the check happens at run time, it is a static check that does not
depend on current state. It is triggered by:
- expanding $! (which may cause the job to be remembered)
- ${var=value} default value assignment
- assignment operators in arithmetic
- parameter substitutions in arithmetic except ${#param}, $$, $# and $?
- command substitutions in arithmetic
This means that $((v+1)) does not prevent optimized command substitution,
whereas $(($v+1)) does, because $v might expand to something containing
assignment operators.
Scripts should not depend on these exact details for correctness. It is also
imaginable to have the shell fork if and when a side effect is encountered
or to create a new temporary namespace for variables.
Due to the $! change, the construct $(jobs $!) no longer works. The value of
$! should be stored in a variable outside command substitution first.
Command substitutions consisting of a single simple command are executed in
the main shell process but this should be invisible apart from performance
and very few exceptions such as $(trap).
