Depending on the system I am working on, there might be 2 different possible paths (mutually exclusive):
System1: /tmp/aword/foo
System2: /tmp/bword/foo
I am supposed to echo something into the foo file regardless of which system I encounter (through a shell script).
How do I include a regular expression within the path itself, to take the correct (existent) path?
somethings I have tried:
#doesn't work
echo Hello > /tmp/(a|b)word/foo
#doesn't work
echo Hello > /tmp/[a|b]word/foo
is there a way of doing this without having to include a test before this which tests for path existence?
If it literally is aword and bword and you know that only one of them exists, you can use
echo 'Hello' > /tmp/[ab]word/foo
This is a shell pattern and documented in the Bash manual or the POSIX sh spec.
If, however, both paths exist, Bash will complain with
-bash: [ab]word: ambiguous redirect
Background
I've a script. It's purpose is to generate config files for various system services from templates whenever my gateway acquires a new IP from my ISP. This process includes making successive edits with sed to replace $[template] strings in my custom templates with the correct information.
And to do that I've created a small function designed to take input from stdin, redirect it to a temporary file passed as an argument, and then move that file to replace the destination (and also, often, source) config file. The "edit-in-place dance", if you will.
I created a simple test script with the problematic function:
#!/bin/bash
inplace_dance() {
read -r -d '' data
printf '%s' "${data}" > "${1}~"
mv "${1}~" "${1}"
}
# ATTN: ls is only being used to generate input for testing. It is not being parsed.
ls -l ~/ | inplace_dance ~/test.out
Unfortunately, this works. So it's not the function itself. I also tried it with my custom logging utility (see "complications" below):
#!/bin/bash
. /usr/local/lib/logging.bash
log_identifier='test'
log_console='on'
inplace_dance() {
read -r -d '' data
printf '%s' "${data}" > "${1}~"
mv "${1}~" "${1}"
}
# ATTN: ls is only being used to generate input for testing. It is not being parsed.
bashlog 'notice' $(ls -l ~/ | inplace_dance '/home/wolferz/test.out')
This also works.
The Problem
In its original context, however, it does not work. I've confirmed that ${data} gets set just fine. And that ${1} contains the correct filename. What fails is the second line of the function. I've confirmed printf is being run (see, "Additional Info - Without The Redirect" below)... but the file its output is being redirected to is never created.
And I've been over the code a dozen-dozen times (not an exaggeration) and have yet to identify the cause. So, in desperation, I'm going to post it here and hope some kind soul will wade through my code and maybe spot the problem that I'm missing. I'd also happily take advice on improvements/replacements to my logging utility (in the hopes of tracking down the problem) or further troubleshooting steps.
Here is the original context. The important lines are 106-110, 136-140, 144-147, and 151-155
Additional Info
☛ PATH/Environment
The PATH is PATH=/usr/local/sbin:/usr/local/bin:/usr/bin. I believe this is being inherited from systemd (systemd=>dhcpcd.service=>dhcpcd=>dhcpcd-run-hooks=>dhcpcd.exit-hook).
dhcpcd-run-hooks (see "Complications" below) does clear the environment (keeping the above PATH) when it runs. Thus, I've added an example of the environment the script runs in to the "original context" gist. In this case, the environment when $reason == 'BOUND'. This is output by printenv | sort at the end of execution (and thus should show the final state of the environment).
NOTE: Be aware this is Arch Linux and the absence of /bin, /sbin, and /usr/sbin in the PATH is normal (they are just symlinks to /usr/bin anyway).
☛ Return Code
Inserting echo $? after the second line of the function gives me a return code of "0". This is true both with the redirect in line 2 and without (just the printf).
☛ Without The Redirect
Without the redirect, in the original context, the second line of the function prints the contents of ${data} to stdout (which is then captured by bashlog()) exactly as expected.
⚠️ Execute Instead of Source.
Turns out that $0 was /usr/lib/dhcpcd/dhcpcd-run-hooks rather than my script. Apparently dhcpcd-run-hooks doesn't execute the script... it sources it. I made some changes to line 196 to fix this.
♔ Aaaaaand that seems to have fixed all problems. ♔
I'm trying to confirm that was the silver bullet now... I didn't notice it was working till I had made several other changes as well. If I can confirm it I'll submit an answer.
Complications
What complicates matters quite a bit is that it's original context is a /etc/dhcpcd.exit-hook script. dhcpcd-run-hooks appears to eat all stderr and stdout which makes troubleshooting... unpleasant. I've implemented my own logging utility to capture the output of commands in the script and pass it to journald but it's not helping in this case. Either no error is being generated or, somehow, the error is not getting captured by my logging utility. The script is running as root and there is no mandatory access control installed so it shouldn't be a permissions issue.
This is really just out of curiosity.
A typo made me notice that in Bash, the following:
$ .anything
does not print any error ("anything" not to be interpreted literally, it can really be anything, and no space after the dot).
I am curious about how this is interpreted in bash.
Note that echo $? after such command returns 127. This usually means "command not found". It does make sense in this case, however I find it odd that no error message is printed.
Why would $ anything actually print bash:anything: command not found... (assuming that no anything cmd is in the PATH), while $ .anything slips through silently?
System: Fedora Core 22
Bash version: GNU bash, version 4.3.39(1)-release (x86_64-redhat-linux-gnu)
EDIT:
Some comments below indicated the problem as non-reproducible at first.
The answer of #hek2mgl below summarises the many contributions to this issue, which was eventually found (by #n.m.) as reproducible in FC22 and submitted as a bug report in https://bugzilla.redhat.com/show_bug.cgi?id=1292531
bash supports a handler for situations when a command can't be found. You can define the following function:
function command_not_found_handle() {
command=$1
# do something
}
Using that function it is possible to suppress the error message. Search for that function in your bash startup files.
Another way to find that out is to unset the function. Like this:
$ unset -f command_not_found_handle
$ .anything # Should display the error message
After some research, #n.m. found out that the described behaviour is by intention. FC22 implements command_not_found_handle and calls the program /etc/libexec/pk-command-not-found. This program is part of the PackageKit project and will try to suggest installable packages if you type a command name that can't be found.
In it's main() function the program explicitly checks if the command name starts with a dot and silently returns in that case. This behaviour was introduced in this commit:
https://github.com/hughsie/PackageKit/commit/0e85001b
as a response to this bug report:
https://bugzilla.redhat.com/show_bug.cgi?id=1151185
IMHO this behaviour is questionable. At least other distros are not doing so. But now you know that the behaviour is 100% reproducible and you may follow up on that bug report.
I wrote hook for command line:
# Transforms command 'ls?' to 'man ls'
function question_to_man() {
if [[ $2 =~ '^\w+\?$' ]]; then
man ${2[0,-2]}
fi
}
autoload -Uz add-zsh-hook
add-zsh-hook preexec question_to_man
But when I do:
> ls?
After exiting from man I get:
> zsh: no matches found: ls?
How can I get rid of from message about wrong command?
? is special to zsh and is the wildcard for a single character. That means that if you type ls? zsh tries find matching file names in the current directory (any three letter name starting with "ls").
There are two ways to work around that:
You can make "?" "unspecial" by quoting it: ls\?, 'ls?' or "ls?".
You make zsh handle the cases where it does not match better:
The default behaviour if no match can be found is to print an error. This can be changed by disabling the NOMATCH option (also NULL_GLOB must not be set):
setopt NO_NOMATCH
setopt NO_NULL_GLOB
This will leave the word untouched, if there is no matching file.
Caution: In the (maybe unlikely) case that there is a file with a matching name, zsh will try to execute a command with the name of the first matching file. That is if there is a file named "lsx", then ls? will be replaced by lsx and zsh will try to run it. This may or may not fail, but will most likely not be the desired effect.
Both methods have their pro and cons. 1. is probably not exactly what you are looking for and 2. does not work every time as well as changes your shells behaviour.
Also (as #chepner noted in his comment) preexec runs additionally to not instead of a command. That means you may get the help for ls but zsh will still try to run ls? or even lsx (or another matching name).
To avoid that, I would suggest defining a command_not_found_handler function instead of preexec. From the zsh manual:
If no external command is found but a function command_not_found_handler exists the shell executes this function with all command line arguments. The function should return status zero if it successfully handled the command, or non-zero status if it failed. In the latter case the standard handling is applied: ‘command not found’ is printed to standard error and the shell exits with status 127. Note that the handler is executed in a subshell forked to execute an external command, hence changes to directories, shell parameters, etc. have no effect on the main shell.
So this should do the trick:
command_not_found_handler () {
if [[ $1 =~ '\?$' ]]; then
man ${1%\?}
return 0
else
return 1
fi
}
If you have a lot of matching file names but seldomly misstype commands (the usual reason for "Command not found" errors) you might want to consider using this instead:
command_not_found_handler () {
man ${1%?}
}
This does not check for "?" at the end, but just cuts away any last character (note the missing "\" in ${1%?}) and tries to run man on the rest. So even if a file name matches, man will be run unless there is indeed a command with the same name as the matched file.
Note: This will interfere with other tools using command_not_found_handler for example the command-not-found tool from Ubuntu (if enabled for zsh).
That all being said, zsh has a widget called run-help which can be bound to a key (in Emacs mode it is by default bound to Alt+H) and than runs man for the current command.
The main advantages of using run-help over the above are:
You can call it any time while typing a longer command, as long as the command name is complete.
After you leave the manpage, the command is still there unchanged, so you can continue writing on it.
You can even bind it to Alt+? to make it more similar: bindkey '^[?' run-help
I'm creating a project and using GNU Autoconf tools to do the configuring and making. I've set up all my library checking and header file checking but can't seem to figure out how to check if an executable exists on the system and fail if it doesn't exist.
I've tried:
AC_CHECK_PROG(TEST,testprogram,testprogram,AC_MSG_ERROR(Cannot find testprogram.))
When I configure it runs and outputs:
Checking for testprogram... find: `testprogram. 15426 5 ': No such file or directory
but does not fail.
I found this to be the shortest approach.
AC_CHECK_PROG(FFMPEG_CHECK,ffmpeg,yes)
AS_IF([test x"$FFMPEG_CHECK" != x"yes"], [AC_MSG_ERROR([Please install ffmpeg before configuring.])])
Try this which is what I just lifted from a project of mine, it looks for something called quantlib-config in the path:
# borrowed from a check for gnome in GNU gretl: def. a check for quantlib-config
AC_DEFUN(AC_PROG_QUANTLIB, [AC_CHECK_PROG(QUANTLIB,quantlib-config,yes)])
AC_PROG_QUANTLIB
if test x"${QUANTLIB}" == x"yes" ; then
# use quantlib-config for QL settings
[.... more stuff omitted here ...]
else
AC_MSG_ERROR([Please install QuantLib before trying to build RQuantLib.])
fi
Similar to the above, but has the advantage of also being able to interact with automake by exporting the condition variable
AC_CHECK_PROG([ffmpeg],[ffmpeg],[yes],[no])
AM_CONDITIONAL([FOUND_FFMPEG], [test "x$ffmpeg" = xyes])
AM_COND_IF([FOUND_FFMPEG],,[AC_MSG_ERROR([required program 'ffmpeg' not found.])])
When using AC_CHECK_PROG, this is the most concise version that I've run across is:
AC_CHECK_PROG(BOGUS,[bogus],[bogus],[no])
test "$BOGUS" == "no" && AC_MSG_ERROR([Required program 'bogus' not found.])
When the program is missing, this output will be generated:
./configure
...cut...
checking for bogus... no
configure: error: Required program 'bogus' not found.
Or when coupled with the built-in autoconf program checks, use this instead:
AC_PROG_YACC
AC_PROG_LEX
test "$YACC" == ":" && AC_MSG_ERROR([Required program 'bison' not found.])
test "$LEX" == ":" && AC_MSG_ERROR([Required program 'flex' not found.])
Stumbled here while looking for this issue, I should note that if you want to have your program just looked in pathm a runtime test is enough:
if ! which programname >/dev/null ; then
AC_MSG_ERROR([Missing programname]
fi
This is not exactly a short approach, it's rather a general purporse approach (although when there are dozens of programs to check it might be also the shortest approach). It's taken from a project of mine (the prefix NA_ stands for “Not Autotools”).
A general purpose macro
dnl ***************************************************************************
dnl NA_REQ_PROGS(prog1, [descr1][, prog2, [descr2][, etc., [...]]])
dnl
dnl Checks whether one or more programs have been provided by the user or can
dnl be retrieved automatically. For each program `progx` an uppercase variable
dnl named `PROGX` containing the path where `progx` is located will be created.
dnl If a program is not reachable and the user has not provided any path for it
dnl an error will be generated. The program names given to this function will
dnl be advertised among the `influential environment variables` visible when
dnl launching `./configure --help`.
dnl ***************************************************************************
AC_DEFUN([NA_REQ_PROGS], [
m4_if([$#], [0], [], [
AC_ARG_VAR(m4_translit([$1], [a-z], [A-Z]), [$2])
AS_IF([test "x#S|#{]m4_translit([$1], [a-z], [A-Z])[}" = x], [
AC_PATH_PROG(m4_translit([$1], [a-z], [A-Z]), [$1])
AS_IF([test "x#S|#{]m4_translit([$1], [a-z], [A-Z])[}" = x], [
AC_MSG_ERROR([$1 utility not found])
])
])
m4_if(m4_eval([$# + 1 >> 1]), [1], [], [NA_REQ_PROGS(m4_shift2($*))])
])
])
Sample usage
NA_REQ_PROGS(
[find], [Unix find utility],
[xargs], [Unix xargs utility],
[customprogram], [Some custom program],
[etcetera], [Et cetera]
)
So that within Makefile.am you can do
$(XARGS)
or
$(CUSTOMPROGRAM)
and so on.
Features
It advertises the programs among the “influential environment variables” visible when the final user launches ./configure --help, so that an alternative path to the program can be provided
A bash variable named with the same name of the program, but upper case, containing the path where the program is located, is created
En error is thrown if any of the programs given have not been found and the user has not provided any alternative path for them
The macro can take infinite (couples of) arguments
When you should use it
When the programs to be tested are vital for compiling your project, so that the user must be able to provide an alternative path for them and an error must be thrown if at least one program is not available at all
When condition #1 applies to more than one single program, in which case there is no need to write a general purpose macro and you should just use your own customized code