Re: Re: [here string] uncompatible change on here string function
On Fri, Nov 24, 2017 at 11:22:56AM +0800, 尹剑虹 wrote:
> It's not the point. the problem is bash-4.4 auto add quote around $var or
> $(cmd) after '<<<'
>
> On bash-4.2 and before following command always works fine
> read ignore fstype <<<$(df --output=fstype $mountpoint)
>
> But after Update bash-4.4, our customer's script get fail, do you understand
> what I mean?
The script is broken. It happened to work in bash 4.2 because of a bug
in bash 4.2 that cancelled out the bug in the script.
wooledg:~$ df --output=fstype /
Type
ext4
Here is how the script should be written:
{ read; read -r fstype; } < <(df --output=fstype "$mountpoint")
And testing it:
wooledg:~$ mountpoint=/
wooledg:~$ { read; read -r fstype; } < <(df --output=fstype "$mountpoint")
wooledg:~$ declare -p fstype
declare -- fstype="ext4"
Bash: sleep execution issue with bash loadable builtins
Hi,
Found a 'sleep' execution issue with bash loadable builtins and has
performed
the below sleep test on bash-4.4-rc1.
Bash version: 4.4-rc1
Web link: https://ftp.gnu.org/gnu/bash/bash-4.4-rc1.tar.gz
Reproducible test case and Console logs:
bash-4.4-rc1#./bash
bash-4.4-rc1# cd examples/loadables/
bash-4.4-rc1/examples/loadables# enable -f ./sleep sleep
bash-4.4-rc1/examples/loadables# date; sleep 1 & date; sleep 4; date
Mon Nov 27 17:00:01 IST 2017
[1] 5355
Mon Nov 27 17:00:01 IST 2017
Mon Nov 27 17:00:02 IST 2017--> Took 1 Sec only instead of
4 Sec's.
[1]+ Donesleep 1
bash-4.4-rc1/examples/loadables#
bash-4.4-rc1/examples/loadables# date; sleep 4 & date; sleep 1; date
Mon Nov 27 17:00:16 IST 2017
[1] 5386
Mon Nov 27 17:00:16 IST 2017
Mon Nov 27 17:00:17 IST 2017
bash-4.4-rc1/examples/loadables#
bash-4.4-rc1/examples/loadables# date; sleep 1 & date; sleep 10; date
Mon Nov 27 17:00:29 IST 2017
[1]+ Donesleep 4
[1] 5390
Mon Nov 27 17:00:29 IST 2017
Mon Nov 27 17:00:30 IST 2017 -> Took 1 Sec only instead of
10 Sec's.
[1]+ Donesleep 1
root@cavium-Vostro-3446
:~/Desktop/bugzilla-CGX/77504/bash-4.4-rc1/examples/loadables#
Fix patch:
==
# Attached the fix patch -
Fix_for_bash_loadable_builtin_sleep_execution_issue.patch
Please kindly review and suggest your comments.
Thanks,
Thiruvadi Rajaraman
Fix for bash loadable builtins 'sleep' execution failure issue
The rootcause of the failure was found with select() execution
with time information in fsleep() function in lib/sh/ufunc.c file.
The fix patch content resolves the issue.
Test logs (After applying patch)
=
# date; sleep 1 & date; sleep 10; date
Mon Nov 27 17:25:31 IST 2017
[1] 13173
Mon Nov 27 17:25:31 IST 2017
Mon Nov 27 17:25:41 IST 2017-> Took 10 Sec's sleep
[1]+ Donesleep 1
#
Signed-off-by: Thiruvadi Rajaraman
Index: bash-4.4-rc1/lib/sh/ufuncs.c
===
--- bash-4.4-rc1.orig/lib/sh/ufuncs.c 2017-11-27 13:57:18.172796871 +0530
+++ bash-4.4-rc1/lib/sh/ufuncs.c 2017-11-27 17:21:13.622624417 +0530
@@ -90,11 +90,16 @@
unsigned int sec, usec;
{
struct timeval tv;
+ int ret = 0;
tv.tv_sec = sec;
tv.tv_usec = usec;
- return select(0, (fd_set *)0, (fd_set *)0, (fd_set *)0, &tv);
+ do {
+ ret = select(0, (fd_set *)0, (fd_set *)0, (fd_set *)0, &tv);
+ } while ( (ret == -1));
+
+ return ret;
}
#else /* !HAVE_TIMEVAL || !HAVE_SELECT */
int
Feature idea
Recently I wanted to put the content of the stdout and the stderr of a command into variables (without needing to write to a file), something like this: sterr=$(cmd >>>stout) Or: stout=$(cmd 2>>>sterr) Or even: cmd >>>stout 2>>>sterr Obviously the idea is to introduce an operator >>> that functions as a "reverse here-document" that stores the content of a file stream into a variable. There would be less need to store things in a file on a filesystem. I think that would be very useful extension, easy to comprehend in the light of current syntax, and not clashing with anything as far as I can see. Thanks for considering this, Peter
Re: [here string] uncompatible change on here string function
> > It's not the point. the problem is bash-4.4 auto add quote around $var or
> > $(cmd) after '<<<'
> >
> > On bash-4.2 and before following command always works fine
> > read ignore fstype <<<$(df --output=fstype $mountpoint)
> >
> > But after Update bash-4.4, our customer's script get fail, do you
> > understand what I mean?
>
> The script is broken. It happened to work in bash 4.2 because of a bug
> in bash 4.2 that cancelled out the bug in the script.
>
> wooledg:~$ df --output=fstype /
> Type
> ext4
>
> Here is how the script should be written:
>
> { read; read -r fstype; } < <(df --output=fstype "$mountpoint")
>
> And testing it:
>
> wooledg:~$ mountpoint=/
> wooledg:~$ { read; read -r fstype; } < <(df --output=fstype "$mountpoint")
> wooledg:~$ declare -p fstype
> declare -- fstype="ext4"
Or even
fstype=$( stat --file-system --print "%T\n" "$mountpoint" )
--
Vlad
Re: Bash: sleep execution issue with bash loadable builtins
On 11/27/17 4:17 AM, Thiruvadi Rajaraman wrote: > Hi, > > Found a 'sleep' execution issue with bash loadable builtins and has > performed > the below sleep test on bash-4.4-rc1. That's an interesting one. It looks like the SIGCHLD interrupts the select loop, even though bash supplies SA_RESTART when installing its SIGCHLD handler. It's probably too hard to restart it in general, since select doesn't necessarily modify its timeval argument when it returns early (Linux does; many other OSs do not). There is a problem with your fix in that, in most cases, you've just made everything that uses this function non-interruptible, especially in an interactive shell. I think a better fix would be to change fsleep() to cope with select(2) being interrupted using the bash primitives that deal with signal and trap handling. Chet -- ``The lyf so short, the craft so long to lerne.'' - Chaucer ``Ars longa, vita brevis'' - Hippocrates Chet Ramey, UTech, CWRUc...@case.eduhttp://cnswww.cns.cwru.edu/~chet/
Re: Bash: sleep execution issue with bash loadable builtins
On 2017-11-27 at 17:47 +0530, Thiruvadi Rajaraman wrote: > Reproducible test case and Console logs: > Simpler test case: bash-4.4-rc1# cd examples/loadables/ bash-4.4-rc1/examples/loadables# enable -f ./sleep sleep bash-4.4-rc1/examples/loadables# sleep 1 & time sleep 10 [1] 8892 real0m1.005s user0m0.001s sys 0m0.004s [1]+ Donesleep 1 > Fix patch: > == > # Attached the fix patch - > Fix_for_bash_loadable_builtin_sleep_execution_issue.patch > > > Please kindly review and suggest your comments. > > Thanks, > Thiruvadi Rajaraman I guess the line + } while ( (ret == -1)); should actually be while ( (ret == -1) && (errno == EINTR) ); ? Also there's the issue that select() _may_ modify the object pointed to by the timeout argument [POSIX]. But it may not, in which case this would end up oversleeping. On such system, doing eg. sleep 9 & time sleep 10 would end up sleeping 19 seconds. The solution is probably to change that select() into a pselect() that masks SIGCHLD (as well as _some_ other signals, but not SIGINT). Best regards
Re: [PATCH] Fix hang if $OLDPWD points to inaccessible directory
On Wed, 4 Oct 2017, Eduardo A. Bustamante López wrote: > On Tue, Oct 03, 2017 at 10:29:08PM +0200, Mikulas Patocka wrote: > > If $OLDPWD points to a non-existing directory, 'cd -' will fail. > > But if we clear $OLDPWD, 'cd -' will fail too (with just different message). > [...] > > I performed the following tests: > > > dualbus@ubuntu:~$ for sh in mksh ksh93 dash zsh posh bash; do echo $sh \| > $(OLDPWD=/invalid $sh -c 'echo "$OLDPWD | "; cd -; echo " | $?"' 2>&1); done > mksh | /invalid | mksh: cd: /invalid: No such file or directory | 2 > ksh93 | /invalid | ksh93: cd: /invalid: [No such file or directory] | 1 > dash | /invalid | dash: 1: cd: can't cd to /invalid | 2 > zsh | /home/dualbus | | 0 > posh | /invalid | posh: cd: /invalid - No such file or directory | 1 > bash | | bash: line 0: cd: OLDPWD not set | 1 > > dualbus@ubuntu:~$ for sh in mksh ksh93 dash zsh posh bash; do echo $sh \| > $(OLDPWD=/tmp $sh -c 'echo "$OLDPWD | "; cd -; echo " | $?"' 2>&1); done > mksh | /tmp | /tmp | 0 > ksh93 | /tmp | /tmp | 0 > dash | /tmp | /tmp | 0 > zsh | /home/dualbus | | 0 > posh | /tmp | /tmp | 0 > bash | /tmp | /tmp | 0 > > > So: > > - Bash is the only major shell that performs stat() on OLDPWD to see if it's > going to import it from the environment or initialize it to an empty > value. > > - Zsh doesn't import OLDPWD from the environment (bash's old behavior) > > - The rest of the shells (mksh, ATT ksh, dash, posh) import OLDPWD from > the environment unconditionally. > > - Mikulas already made the argument above of why this change wouldn't > break existing scripts that rely on bash performing the stat(). > > - I don't see a way around this behavior using existing bash features. > The value of OLDPWD can't be reset via BASH_ENV or other > initialization mechanisms. Sure, `unset OLDPWD; program' might work, > but I wouldn't call that a solution. > > > Given the above, I've changed my mind, and I think this patch should be > fine :-) Hi Will you commit my patch? - if you agree with it. Mikulas
