side. Sometimes it was compared to int, which was -1 on EOF, and on
unsigned char machines UPEOF was (unsigned char)-1. This worked
by chance because isalpha((unsigned char)-1) returns false usually,
but it does not when the locale is invalid!
- ansification
- format of output of jobs command (etc)
- job identiers %+, %- etc
- $? and $(...)
- correct quoting of output of set, export -p and readonly -p
- differentiation between nornal and 'posix special' builtins
- correct behaviour (posix) for errors on builtins and special builtins
- builtin printf and kill
- set -o debug (if compiled with DEBUG)
- cd src obj (as ksh - too useful to do without)
- unset -e name, remove non-readonly variable from export list.
(so I could unset -e PS1 before running the test shell...)
> The wrong process is aborting when variable assignment fails
> in the vfork path. So the following command fails to execute
> the second echo (shown here with the correct output).
>
> $ (readonly r; r= /bin/echo a; echo b)
> r: is read only
> b
>
> fix: defer the mklocal() to the child shell.
> Also 'jobs' fails in a non-interactive shell.
> In showjobs(), the code that puts the process back into its
> own process group should only be run if the shell is actually
> doing job control - eg if 'mflag' is set.
$ /fred # non existant command
$ ^C # stops working
He says:
Ok the extra INTOFF is the one in exverror().
In almost all cases this doesn't matter because the longjmp()s
all end up in main() and the FORCEINTON call sorts it out
for the next command.
(There are a significant number of INTON/OFF mismatches through
the error paths...)
In any case the above failure can be 'fixed' by changing 2 (I think
they are both needed) INTON calls to FORCEINTON within evalcommand.
The following patch seems to work:
We should really look in the code and fix the INTON->INTOFF pairs.
effects, and add double to it, so that it aligns doubles correctly too. This
is just a workaround to fix the sparc64 problem where ALIGN() is now defined
in some include file to be 16 instead of 8. Thanks to martin for debugging this.
the non-vfork case. Having said that, it would be nice if pipelines of
simple commands were vforked too. Right now they are not.
Explain that setpgid() might fail because we are doing it both in the
parent and the child case, because we don't know which one will come
first.
Suspending a pipeline prints %1 Suspended n times where n is the number
of processes, but that was there before. It is easy to fix, but I'll
leave the code alone for now.
Propagate isroot, throughout the eval process and maintain it properly.
Fixes sleep 10 | cat^C not exiting because sleep and cat ended up in
their own process groups, because wasroot was always true in the children.
Plus my changes:
- walking process group fix in foregrounding a job.
- reset of process group in parent shell if interrupted before the wait.
- move INTON lower in the dowait so that the job structure is
consistent.
- error check all setpgid(), tcsetpgrp() calls.
- eliminate unneeded strpgid() call.
- check that we don't belong in the process group before we try to
set it.
race conditions -- now we always synchronously wait for the job to finish.
In evalcommand(), add the same INTOFF/INTON locking as evalpipe(), to prevent
leaving internal state inconsistent, and also to insure that we synchronously
wait for the job.
for TARGET_CHAR when building mksyntax. This isn't perfect, but
it lets the host tool work on non-BSD systems without completely
redoing how sh is built.
point to a tty.
Unfortunately the shell assumed that it could do all process group
handling ioctls to fd=2, but this is not correct. Jobs that redirected
fd=2 would be unable to perform the ioctls and silently fail since
the error reporting channel is fd=2... Instead open /dev/tty set
it to close on exec, and use that instead (like all other shells
do). We don't handle the case where the OS does not provide FD_CLOEXEC
or FIOCLEX, because I am lazy.
While I am there:
- Simplify the code by defining functions for tc{g,s}pgrp when OLD_TTY_DRIVER
is defined.
- make sure that 'sh &' works by stopping itself. Don't kill the shell's
process group, kill the shell itself.
commit where pipeline commands didn't inherit the correct process group.
Reviewed by Christos.
Change a trace format string arg to use %p instead of %x and a long cast.
the syntax maps to determine the beginning and end quotes (kill
CENDQUOTE). Handle single quotes opening and closing via checking
the current syntax map. Keep a bitmap of doublequote state one bit
per variable nesting level. For the first 32 nested double quotes,
we don't need any additional memory, but for more we allocate
dynamically.
alias expansion inside the switch as appropriate. This is achieved by a
flag noalias which is turned on and off in as we parse. In the following
example [1] and [0] indicate the value of noalias.
[0] case <expr> in
[1] <lit> ) [0] <expr> ;;
[1] <lit> ) [0] <expr> ;;
...
[1] esac [0]
FreeBSD does:
[0] case <expr> in [1]
<lit> ) <expr> ;;
<lit> ) <expr> ;;
...
esac [0]
This handles the following shell script:
alias a=ls
case $1 in
a) echo a;
a;;
f) echo f;;
*) echo default;;
esac
Make sure that each va_start has one and only one matching va_end,
especially in error cases.
If the va_list is used multiple times, do multiple va_starts/va_ends.
If a function gets va_list as argument, don't let it use va_end (since
it's the callers responsibility).
Improved by comments from enami and christos -- thanks!
Heimdal/krb4/KAME changes already fed back, rest to follow.
Inspired by, but not not based on, OpenBSD.
(i.e. processes started from shell scripts). Fixes problem where kill -STOP'ing
a subprocess of a shell script would cause the shell to proceed to the next
command.
have separate man pages for them.
Xref passwd 5 instead of 4, environ 7 instead of 5, and comment out xref
to profile(4), which we don't have.
Improve markup of SYNOPSIS.
Some whitespace fixes while I'm here.
POSIX recommendations.
- trap now accepts signal names and signal numbers
e.g. INT, SIGINT, 2
- added option -l that outputs a list of valid signals
- added signal EXIT to list of valid signals
- a `-' in the action part will reset specified signal to their
default behaviour
- changed standard output format to make it suitable as an input
to another shell that achieves the same trapping results
__CONCAT("PATH=",_PATH_STDPATH);
actually works to concantate strings, it's because the preprocessor expands
it into "PATH=""whatever _PATH_STDPATH is" as separate strings, and then
ANSI string concatenation is performed on that. It's more straightforward
to just use ANSI string concatenation directly, and newer GCCs complain
(rightly) about mis-use of token pasting.
Fix from FreeBSD:
growstackblock() sometimes relocates a stack_block considered empty
without properly relocating stack marks referencing that block.
The first call to popstackmark() with the unrelocated stack mark
as argument then causes sh to abort.
Relocating the relevant stack marks seems to solve this problem.
The patch changes the semantics of popstackmark() somewhat. It can
only be called once after a call to setstackmark(), thus cmdloop() in
main.c needs an extra call to setstackmark().
and so it shouldn't use __P. (this should probably be done better, by
not declaring the parser functions in headers used by host programs,
but this works well enough.)
function (which nobody was able to explain): it's critical to allowing a
complex command run from an interactive shell to be terminated. So, reinstate
it and fix it correctly. See the comment if you really want the gory details.
leading to unexpected behaviour. Disable the no-fork optimization for now.
We need to revisit this and keep enough state around to recover from such
changes.
even if the specified file is not in the PATH.
This change enforces security and makes it conform to POSIX.
Closes PR #6794.
I thought Christos committed this but not appeared yet. :)
The setstackmark()/popstackmark() pair in dotcmd(), used for freeing
stack storage possibly allocated by find_dot_file(), is redundant for now
since dotcmd() is surrounded by another pair in evalcommand().
This redundancy, however, may help future modifications
(suggested by Christos).
The problem was that system calls got restarted after a signal,
instead of returning EINTR. Thus the read builtin, had no way to
know that a signal occured that could change the course of execution.
Since the code has sprinkled checks for EINTR all over the place,
it is supposed to work properly with non restartable syscalls.
The fix is to use siginterrupt(signo, 1), before setting a signal
handler, to make sure that system calls don't get restarted.
Rewrite man page in mandoc format rather than nasty man format.
Fix a ton of parsing errors, and generate proper .Xr's.
document all known environment variables.
suggest ksh rather than bash.
The last two fix PR #1966. Wheee!
Somebody with access to the POSIX spec needs to go in here, and document
our adherence, or lack thereof.
such as sh -c 'echo `echo foo`' .
The memory allocated with ckmalloc() at
parser.c:1349:readtoken1() (search for "done:" label)
was never freed.
I changed this to use 'string stack' framework of Ash.
Note that a string on string stack is properly freed on
exception and end of command parsing, and no explicit free
or signal handlings required.
See TOUR for an overview, and memalloc.[ch] for details
of string stack.
and must be freed to avoid memory leaks if called repeatedly.
The leaks occured on symbolic umask command, such as "umask go-w",
which is undocumented.
- it was in the wrong place
- makefiles shouldn't override CFLAGS; only CPPFLAGS and COPTS
- christos fixed unsigned char stuff in 1.33 which should
remove the need for -w anyway
* move .include <bsd.prog.mk> to EOF
involved in the `for' list, the list was recorded twice, leading to incorrect
argument expansion.
Introduce ifsfree() function that free's the IFS region list, GC'ing duplicated
code.
sleep
cmd='set `type "sleep"`; eval echo \$$#'
which=`eval $cmd`
echo $which
because the region did not get recorded at all, and it was interpreted as
a single word. I modified the code to keep track when the result of a
backquote expansion has been recorded to avoid recording it twice. I still
feel that this is not the right fix... More to come.
assignment. E.g.
echo ${foo:=`echo 1 2 3 4`}
prints:
1 2 3 1 2 3 4
because when the arquments are not quoted, the backquote result
gets recorded twice. The fix right now is to comment out the
record_region() call in expbackq(). I hope that it does not break
anything else.
escape character (including line continuation), unless the `-r' option
is specified:
* adopt to this behaviour, add the `-r' option to disable it;
* remove the `-e' option, which was previously necessary to get this behaviour.
cmdenviron is pointing to varlist.list; varlist gets reset everytime
you enter evalcommand, but cmdenviron does not. The wonders of global
variables...
unsigned character machines. So that people don't have to reverse engineer
this again:
mksyntax detects if characters are signed or not and builts a syntax
table that has a base of 129 for signed characters or 1 for unsigned
characters. This is so the largest negative signed char [-128] + the
base == 1. 0 is special and means end of file in both cases. PEOF
is -1 for the unsigned character case and -129 for the signed
character case, so that syntax[PEOF + base] == syntax[0] == CEOF
So PEOF has to be -1, but it is explicitly compared with
unsigned characters on machines where characters are unsigned.
The quick fix is to define UPEOF the (unsigned char) version of PEOF
and use that. A better fix is to always use unsigned characters
when referencing symbol table entries, but that would require
extensive changes to the shell. So to summarize
syntax[0] == CEOF, base + PEOF == 0
unsigned signed
base 1 129
PEOF -1 -129
dsl