/usr/bin/env questions regarding shebang line pecularities - linux

Questions:
What does the kernel do if you stick a shell-script into the shebang line?
How does the Kernel know which interpreter to launch?
Explanation:
I recently wanted to write a wrapper around /usr/bin/env because my CGI Environment does not allow me to set the PATH variable, except globally (which of course sucks!).
So I thought, "OK. Let's set PREPENDPATH and set PATH in a wrapper around env.". The resulting script (here called env.1) looked like this:
#!/bin/bash
/usr/bin/env PATH=$PREPENDPATH:$PATH $*
which looks like it should work. I checked how they both react, after setting PREPENDPATH:
$ which /usr/bin/env python
/usr/bin/env
/usr/bin/python
$ which /usr/bin/env.1 python
/usr/bin/env
/home/pi/prepend/bin/python
Look absolutely perfect! So far, so good. But look what happens to "Hello World!".
# Shebang is #!/usr/bin/env python
$ test-env.py
Hello World!
# Shebang is #!/usr/bin/env.1 python
$ test-env.1.py
Warning: unknown mime-type for "Hello World!" -- using "application/*"
Error: no such file "Hello World!"
I guess I am missing something pretty fundamental about UNIX.
I'm pretty lost, even after looking at the source code of the original env. It sets the environment and launches the program (or so it seems to me...).

First of all, you should very seldom use $* and you should almost always use "$#" instead. There are a number of questions here on SO which explain the ins and outs of why.
Second - the env command has two main uses. One is to print the current environment; the other is to completely control the environment of a command when it is run. The third use, which you are demonstrating, is to modify the environment, but frankly there's no need for that - the shells are quite capable of handling that for you.
Mode 1:
env
Mode 2:
env -i HOME=$HOME PATH=$PREPENDPATH:$PATH ... command args
This version cancels all inherited environment variables and runs command with precisely the environment set by the ENVVAR=value options.
The third mode - amending the environment - is less important because you can do that fine with regular (civilized) shells. (That means "not C shell" - again, there are other questions on SO with answers that explain that.) For example, you could perfectly well do:
#!/bin/bash
export PATH=${PREPENDPATH:?}:$PATH
exec python "$#"
This insists that $PREPENDPATH is set to a non-empty string in the environment, and then prepends it to $PATH, and exports the new PATH setting. Then, using that new PATH, it executes the python program with the relevant arguments. The exec replaces the shell script with python. Note that this is quite different from:
#!/bin/bash
PATH=${PREPENDPATH:?}:$PATH exec python "$#"
Superficially, this is the same. However, this will execute the python found on the pre-existing PATH, albeit with the new value of PATH in the process's environment. So, in the example, you'd end up executing Python from /usr/bin and not the one from /home/pi/prepend/bin.
In your situation, I would probably not use env and would just use an appropriate variant of the script with the explicit export.
The env command is unusual because it does not recognize the double-dash to separate options from the rest of the command. This is in part because it does not take many options, and in part because it is not clear whether the ENVVAR=value options should come before or after the double dash.
I actually have a series of scripts for running (different versions of) a database server. These scripts really use env (and a bunch of home-grown programs) to control the environment of the server:
#!/bin/ksh
#
# #(#)$Id: boot.black_19.sh,v 1.3 2008/06/25 15:44:44 jleffler Exp $
#
# Boot server black_19 - IDS 11.50.FC1
IXD=/usr/informix/11.50.FC1
IXS=black_19
cd $IXD || exit 1
IXF=$IXD/do.not.start.$IXS
if [ -f $IXF ]
then
echo "$0: will not start server $IXS because file $IXF exists" 1>&2
exit 1
fi
ONINIT=$IXD/bin/oninit.$IXS
if [ ! -f $ONINIT ]
then ONINIT=$IXD/bin/oninit
fi
tmpdir=$IXD/tmp
DAEMONIZE=/work1/jleffler/bin/daemonize
stdout=$tmpdir/$IXS.stdout
stderr=$tmpdir/$IXS.stderr
if [ ! -d $tmpdir ]
then asroot -u informix -g informix -C -- mkdir -p $tmpdir
fi
# Specialized programs carried to extremes:
# * asroot sets UID and GID values and then executes
# * env, which sets the environment precisely and then executes
# * daemonize, which makes the process into a daemon and then executes
# * oninit, which is what we really wanted to run in the first place!
# NB: daemonize defaults stdin to /dev/null and could set umask but
# oninit dinks with it all the time so there is no real point.
# NB: daemonize should not be necessary, but oninit doesn't close its
# controlling terminal and therefore causes cron-jobs that restart
# it to hang, and interactive shells that started it to hang, and
# tracing programs.
# ??? Anyone want to integrate truss into this sequence?
asroot -u informix -g informix -C -a dbaao -a dbsso -- \
env -i HOME=$IXD \
INFORMIXDIR=$IXD \
INFORMIXSERVER=$IXS \
INFORMIXCONCSMCFG=$IXD/etc/concsm.$IXS \
IFX_LISTEN_TIMEOUT=3 \
ONCONFIG=onconfig.$IXS \
PATH=/usr/bin:$IXD/bin \
SHELL=/usr/bin/ksh \
TZ=UTC0 \
$DAEMONIZE -act -d $IXD -o $stdout -e $stderr -- \
$ONINIT "$#"
case "$*" in
(*v*) track-oninit-v $stdout;;
esac

You should carefully read the wikipedia article about shebang.
When your system sees the magic number corresponding to the shebang, it does an execve on the given path after the shebang and gives the script itself as an argument.
Your script fails because the file you give (/usr/bin/env.1) is not an executable, but begins itself by a shebang....
Ideally, you could resolve it using... env on your script with this line as a shebang:
#!/usr/bin/env /usr/bin/env.1 python
It won't work though on linux as it treats "/usr/bin/env.1 python" as a path (it doesn't split arguments)
So the only way I see is to write your env.1 in C
EDIT: seems like no one belives me ^^, so I've written a simple and dirty env.1.c:
#include <stdlib.h>
#include <stdio.h>
#include <string.h>
#include <unistd.h>
const char* prependpath = "/your/prepend/path/here:";
int main(int argc, char** argv){
int args_len = argc + 1;
char* args[args_len];
const char* env = "/usr/bin/env";
int i;
/* arguments: the same */
args[0] = env;
for(i=1; i<argc; i++)
args[i] = argv[i];
args[argc] = NULL;
/* environment */
char* p = getenv("PATH");
char* newpath = (char*) malloc(strlen(p)
+ strlen(prependpath));
sprintf(newpath, "%s%s", prependpath, p);
setenv("PATH", newpath, 1);
execv(env, args);
return 0;
}

Related

Is there a way to know how the user invoked a program from bash?

Here's the problem: I have this script foo.py, and if the user invokes it without the --bar option, I'd like to display the following error message:
Please add the --bar option to your command, like so:
python foo.py --bar
Now, the tricky part is that there are several ways the user might have invoked the command:
They may have used python foo.py like in the example
They may have used /usr/bin/foo.py
They may have a shell alias frob='python foo.py', and actually ran frob
Maybe it's even a git alias flab=!/usr/bin/foo.py, and they used git flab
In every case, I'd like the message to reflect how the user invoked the command, so that the example I'm providing would make sense.
sys.argv always contains foo.py, and /proc/$$/cmdline doesn't know about aliases. It seems to me that the only possible source for this information would be bash itself, but I don't know how to ask it.
Any ideas?
UPDATE How about if we limit possible scenarios to only those listed above?
UPDATE 2: Plenty of people wrote very good explanation about why this is not possible in the general case, so I would like to limit my question to this:
Under the following assumptions:
The script was started interactively, from bash
The script was start in one of these 3 ways:
foo <args> where foo is a symbolic link /usr/bin/foo -> foo.py
git foo where alias.foo=!/usr/bin/foo in ~/.gitconfig
git baz where alias.baz=!/usr/bin/foo in ~/.gitconfig
Is there a way to distinguish between 1 and (2,3) from within the script? Is there a way to distinguish between 2 and 3 from within the script?
I know this is a long shot, so I'm accepting Charles Duffy's answer for now.
UPDATE 3: So far, the most promising angle was suggested by Charles Duffy in the comments below. If I can get my users to have
trap 'export LAST_BASH_COMMAND=$(history 1)' DEBUG
in their .bashrc, then I can use something like this in my code:
like_so = None
cmd = os.environ['LAST_BASH_COMMAND']
if cmd is not None:
cmd = cmd[8:] # Remove the history counter
if cmd.startswith("foo "):
like_so = "foo --bar " + cmd[4:]
elif cmd.startswith(r"git foo "):
like_so = "git foo --bar " + cmd[8:]
elif cmd.startswith(r"git baz "):
like_so = "git baz --bar " + cmd[8:]
if like_so is not None:
print("Please add the --bar option to your command, like so:")
print(" " + like_so)
else:
print("Please add the --bar option to your command.")
This way, I show the general message if I don't manage to get their invocation method. Of course, if I'm going to rely on changing my users' environment I might as well ensure that the various aliases export their own environment variables that I can look at, but at least this way allows me to use the same technique for any other script I might add later.
No, there is no way to see the original text (before aliases/functions/etc).
Starting a program in UNIX is done as follows at the underlying syscall level:
int execve(const char *path, char *const argv[], char *const envp[]);
Notably, there are three arguments:
The path to the executable
An argv array (the first item of which -- argv[0] or $0 -- is passed to that executable to reflect the name under which it was started)
A list of environment variables
Nowhere in here is there a string that provides the original user-entered shell command from which the new process's invocation was requested. This is particularly true since not all programs are started from a shell at all; consider the case where your program is started from another Python script with shell=False.
It's completely conventional on UNIX to assume that your program was started through whatever name is given in argv[0]; this works for symlinks.
You can even see standard UNIX tools doing this:
$ ls '*.txt' # sample command to generate an error message; note "ls:" at the front
ls: *.txt: No such file or directory
$ (exec -a foobar ls '*.txt') # again, but tell it that its name is "foobar"
foobar: *.txt: No such file or directory
$ alias somesuch=ls # this **doesn't** happen with an alias
$ somesuch '*.txt' # ...the program still sees its real name, not the alias!
ls: *.txt: No such file
If you do want to generate a UNIX command line, use pipes.quote() (Python 2) or shlex.quote() (Python 3) to do it safely.
try:
from pipes import quote # Python 2
except ImportError:
from shlex import quote # Python 3
cmd = ' '.join(quote(s) for s in open('/proc/self/cmdline', 'r').read().split('\0')[:-1])
print("We were called as: {}".format(cmd))
Again, this won't "un-expand" aliases, revert to the code that was invoked to call a function that invoked your command, etc; there is no un-ringing that bell.
That can be used to look for a git instance in your parent process tree, and discover its argument list:
def find_cmdline(pid):
return open('/proc/%d/cmdline' % (pid,), 'r').read().split('\0')[:-1]
def find_ppid(pid):
stat_data = open('/proc/%d/stat' % (pid,), 'r').read()
stat_data_sanitized = re.sub('[(]([^)]+)[)]', '_', stat_data)
return int(stat_data_sanitized.split(' ')[3])
def all_parent_cmdlines(pid):
while pid > 0:
yield find_cmdline(pid)
pid = find_ppid(pid)
def find_git_parent(pid):
for cmdline in all_parent_cmdlines(pid):
if cmdline[0] == 'git':
return ' '.join(quote(s) for s in cmdline)
return None
See the Note at the bottom regarding the originally proposed wrapper script.
A new more flexible approach is for the python script to provide a new command line option, permitting users to specify a custom string they would prefer to see in error messages.
For example, if a user prefers to call the python script 'myPyScript.py' via an alias, they can change the alias definition from this:
alias myAlias='myPyScript.py $#'
to this:
alias myAlias='myPyScript.py --caller=myAlias $#'
If they prefer to call the python script from a shell script, it can use the additional command line option like so:
#!/bin/bash
exec myPyScript.py "$#" --caller=${0##*/}
Other possible applications of this approach:
bash -c myPyScript.py --caller="bash -c myPyScript.py"
myPyScript.py --caller=myPyScript.py
For listing expanded command lines, here's a script 'pyTest.py', based on feedback by #CharlesDuffy, that lists cmdline for the running python script, as well as the parent process that spawned it.
If the new -caller argument is used, it will appear in the command line, although aliases will have been expanded, etc.
#!/usr/bin/env python
import os, re
with open ("/proc/self/stat", "r") as myfile:
data = [x.strip() for x in str.split(myfile.readlines()[0],' ')]
pid = data[0]
ppid = data[3]
def commandLine(pid):
with open ("/proc/"+pid+"/cmdline", "r") as myfile:
return [x.strip() for x in str.split(myfile.readlines()[0],'\x00')][0:-1]
pid_cmdline = commandLine(pid)
ppid_cmdline = commandLine(ppid)
print "%r" % pid_cmdline
print "%r" % ppid_cmdline
After saving this to a file named 'pytest.py', and then calling it from a bash script called 'pytest.sh' with various arguments, here's the output:
$ ./pytest.sh a b "c d" e
['python', './pytest.py']
['/bin/bash', './pytest.sh', 'a', 'b', 'c d', 'e']
NOTE: criticisms of the original wrapper script aliasTest.sh were valid. Although the existence of a pre-defined alias is part of the specification of the question, and may be presumed to exist in the user environment, the proposal defined the alias (creating the misleading impression that it was part of the recommendation rather than a specified part of the user's environment), and it didn't show how the wrapper would communicate with the called python script. In practice, the user would either have to source the wrapper or define the alias within the wrapper, and the python script would have to delegate the printing of error messages to multiple custom calling scripts (where the calling information resided), and clients would have to call the wrapper scripts. Solving those problems led to a simpler approach, that is expandable to any number of additional use cases.
Here's a less confusing version of the original script, for reference:
#!/bin/bash
shopt -s expand_aliases
alias myAlias='myPyScript.py'
# called like this:
set -o history
myAlias $#
_EXITCODE=$?
CALL_HISTORY=( `history` )
_CALLING_MODE=${CALL_HISTORY[1]}
case "$_EXITCODE" in
0) # no error message required
;;
1)
echo "customized error message #1 [$_CALLING_MODE]" 1>&2
;;
2)
echo "customized error message #2 [$_CALLING_MODE]" 1>&2
;;
esac
Here's the output:
$ aliasTest.sh 1 2 3
['./myPyScript.py', '1', '2', '3']
customized error message #2 [myAlias]
There is no way to distinguish between when an interpreter for a script is explicitly specified on the command line and when it is deduced by the OS from the hashbang line.
Proof:
$ cat test.sh
#!/usr/bin/env bash
ps -o command $$
$ bash ./test.sh
COMMAND
bash ./test.sh
$ ./test.sh
COMMAND
bash ./test.sh
This prevents you from detecting the difference between the first two cases in your list.
I am also confident that there is no reasonable way of identifying the other (mediated) ways of calling a command.
I can see two ways to do this:
The simplest, as suggested by 3sky, would be to parse the command line from inside the python script. argparse can be used to do so in a reliable way. This only works if you can change that script.
A more complex way, slightly more generic and involved, would be to change the python executable on your system.
Since the first option is well documented, here are a bit more details on the second one:
Regardless of the way your script is called, python is ran. The goal here is to replace the python executable with a script that checks if foo.py is among the arguments, and if it is, check if --bar is as well. If not, print the message and return.
In every other case, execute the real python executable.
Now, hopefully, running python is done trough the following shebang: #!/usr/bin/env python3, or trough python foo.py, rather than a variant of #!/usr/bin/python or /usr/bin/python foo.py. That way, you can change the $PATH variable, and prepend a directory where your false python resides.
In the other case, you can replace the /usr/bin/python executable, at the risk of not playing nice with updates.
A more complex way of doing this would probably be with namespaces and mounts, but the above is probably enough, especially if you have admin rights.
Example of what could work as a script:
#!/usr/bin/env bash
function checkbar
{
for i in "$#"
do
if [ "$i" = "--bar" ]
then
echo "Well done, you added --bar!"
return 0
fi
done
return 1
}
command=$(basename ${1:-none})
if [ $command = "foo.py" ]
then
if ! checkbar "$#"
then
echo "Please add --bar to the command line, like so:"
printf "%q " $0
printf "%q " "$#"
printf -- "--bar\n"
exit 1
fi
fi
/path/to/real/python "$#"
However, after re-reading your question, it is obvious that I misunderstood it. In my opinion, it is all right to just print either "foo.py must be called like foo.py --bar", "please add bar to your arguments" or "please try (instead of )", regardless of what the user entered:
If that's an (git) alias, this is a one time error, and the user will try their alias after creating it, so they know where to put the --bar part
with either with /usr/bin/foo.py or python foo.py:
If the user is not really command line-savvy, they can just paste the working command that is displayed, even if they don't know the difference
If they are, they should be able to understand the message without trouble, and adjust their command line.
I know it's bash task, but i think the easiest way is modify 'foo.py'. Of course it depends on level of script complicated, but maybe it will fit. Here is sample code:
#!/usr/bin/python
import sys
if len(sys.argv) > 1 and sys.argv[1] == '--bar':
print 'make magic'
else:
print 'Please add the --bar option to your command, like so:'
print ' python foo.py --bar'
In this case, it does not matter how user run this code.
$ ./a.py
Please add the --bar option to your command, like so:
python foo.py --bar
$ ./a.py -dua
Please add the --bar option to your command, like so:
python foo.py --bar
$ ./a.py --bar
make magic
$ python a.py --t
Please add the --bar option to your command, like so:
python foo.py --bar
$ /home/3sky/test/a.py
Please add the --bar option to your command, like so:
python foo.py --bar
$ alias a='python a.py'
$ a
Please add the --bar option to your command, like so:
python foo.py --bar
$ a --bar
make magic

How to execute Rust code directly on Unix systems? (using the shebang)

From reading this thread, it looks like its possible to use the shebang to run Rust *.
#!/usr/bin/env rustc
fn main() {
println!("Hello World!");
}
Making this executable and running does compile, but not run the code.
chmod +x hello_world.rs
./hello_world.rs
However this only compiles the code into hello_world.
Can *.rs files be executed directly, similar to a shell script?
* This references rustx, I looked into this, but its a bash script which compiles the script every time (without caching) and never removes the file from the temp directory, although this could be improved. Also it has the significant limitation that it can't use crates.
There's cargo-script. That also lets you use dependencies.
After installing cargo-script via cargo install cargo-script, you can create your script file (hello.rs) like this:
#!/usr/bin/env run-cargo-script
fn main() {
println!("Hello World!");
}
To execute it, you need to:
$ chmod +x hello.rs
$ ./hello.rs
Compiling hello v0.1.0 (file://~/.cargo/.cargo/script-cache/file-hello-d746fc676c0590b)
Finished release [optimized] target(s) in 0.80 secs
Hello World!
To use crates from crates.io, please see the tutorial in the README linked above.
This seems to work:
#!/bin/sh
//usr/bin/env rustc $0 -o a.out && ./a.out && rm ./a.out ; exit
fn main() {
println!("Hello World!");
}
I have written a tool just for that: Scriptisto. It is a fully language agnostic tool and it works with other compiled languages or languages that have expensive validation steps (Python with mypy).
For Rust it can also fetch dependencies behind the scenes or build entirely in Docker without having a Rust compiler installed. scriptisto embeds those templates into the binary so you can bootstrap easily:
$ scriptisto new rust > ./script.rs
$ chmod +x ./script.rs
$ ./script.rs
Instead of new rust you can do new docker-rust and the build will not require Rust compiler on your host system.
#!/bin/sh
#![allow()] /*
exec cargo-play --cached --release $0 -- "$#"
*/
Needs cargo-play. You can see a solution that doesn't need anything here:
#!/bin/sh
#![allow()] /*
# rust self-compiler by Mahmoud Al-Qudsi, Copyright NeoSmart Technologies 2020
# See <https://neosmart.net/blog/self-compiling-rust-code/> for info & updates.
#
# This code is freely released to the public domain. In case a public domain
# license is insufficient for your legal department, this code is also licensed
# under the MIT license.
# Get an output path that is derived from the complete path to this self script.
# - `realpath` makes sure if you have two separate `script.rs` files in two
# different directories, they get mapped to different binaries.
# - `which` makes that work even if you store this script in $PATH and execute
# it by its filename alone.
# - `cut` is used to print only the hash and not the filename, which `md5sum`
# always includes in its output.
OUT=/tmp/$(printf "%s" $(realpath $(which "$0")) | md5sum | cut -d' ' -f1)
# Calculate hash of the current contents of the script, so we can avoid
# recompiling if it hasn't changed.
MD5=$(md5sum "$0" | cut -d' ' -f1)
# Check if we have a previously compiled output for this exact source code.
if !(test -f "${OUT}.md5" && test "${MD5}" = "$(cat ${OUT}.md5)"); then
# The script has been modified or is otherwise not cached.
# Check if the script already contains an `fn main()` entry point.
if grep -Eq '^\s*(\[.*?\])*\s*fn\s*main\b*' "$0"; then
# Compile the input script as-is to the previously determined location.
rustc "$0" -o ${OUT}
# Save rustc's exit code so we can compare against it later.
RUSTC_STATUS=$?
else
# The script does not contain an `fn main()` entry point, so add one.
# We don't use `printf 'fn main() { %s }' because the shebang must
# come at the beginning of the line, and we don't use `tail` to skip
# it because that would result in incorrect line numbers in any errors
# reported by rustc, instead we just comment out the shebang but leave
# it on the same line as `fn main() {`.
printf "fn main() {//%s\n}" "$(cat $0)" | rustc - -o ${OUT}
# Save rustc's exit code so we can compare against it later.
RUSTC_STATUS=$?
fi
# Check if we compiled the script OK, or exit bubbling up the return code.
if test "${RUSTC_STATUS}" -ne 0; then
exit ${RUSTC_STATUS}
fi
# Save the MD5 of the current version of the script so we can compare
# against it next time.
printf "%s" ${MD5} > ${OUT}.md5
fi
# Execute the compiled output. This also ends execution of the shell script,
# as it actually replaces its process with ours; see exec(3) for more on this.
exec ${OUT} $#
# At this point, it's OK to write raw rust code as the shell interpreter
# never gets this far. But we're actually still in the rust comment we opened
# on line 2, so close that: */

Making Sublime Text 2 command on linux behave as it does on MacOS X

There are many questions asking about accessing the Sublime Text 2 editor from the command line. The responses, in summary, are to make a symlink, alias or simple shell script to run the appropriate sublime_text command. I can do that. What I want is to make the linux version behave like the MacOS version.
On MacOS, I have the following:
ln -s /Applications/Sublime\ Text\ 2.app/Contents/SharedSupport/bin/subl ~/bin/subl
Then in my .zshrc:
alias subl="$HOME/bin/subl -n"
export EDITOR="$HOME/bin/subl -n -w"
This does two things. It gives me a subl command that opens any files given on the command line in a new window. The subl command does not block the terminal. It also sets up my editor to open sublime text to edit the arguments, but this time it does block. In particular, $EDITOR blocks until its arguments are closed. It does not block on unrelated sublime text windows.
I can achieve a similar effect on linux with the following:
In ~/bin/subl:
#! /bin/zsh
$HOME/Sublime\ Text\ 2/sublime_text -n $# &
and then in ~/bin/subl_wait: (think mate_wait for TextMate users)
#! /bin/zsh
exec $HOME/Sublime\ Text\ 2/sublime_text -n -w $#
I can then set EDITOR to subl_wait, and things almost work. subl opens files for editing and doesn't block. subl_wait opens files for editing and does block.
The problem is that subl_wait is waiting until all open files are closed, not just its arguments.
Is it possible to get this working perfectly?
Looks like I've found the issue. (Thanks to this post: http://www.sublimetext.com/forum/viewtopic.php?f=2&t=7003 )
Basic point: sublime behaves differently depending upon whether an instance is already running!
If an instance is already running then sublime on linux behaves similarly to MacOS. If no instance is running then the terminal blocks until you exit sublime.
With that in mind, we just need to modify the scripts to make sure sublime is running:
in ~/bin/subl_start:
#! /bin/zsh
if [ ! "$(pidof sublime_text)" ] ; then
# start sublime text main instance
# echo "Starting Sublime Text 2"
$HOME/Sublime\ Text\ 2/sublime_text &
sleep 1 # needed to avoid a race condition
fi
in ~/bin/subl:
#! /bin/zsh
. $HOME/bin/subl_start
exec $HOME/Sublime\ Text\ 2/sublime_text -n $#
in ~/bin/subl_wait:
#! /bin/zsh
. $HOME/bin/subl_start
exec $HOME/Sublime\ Text\ 2/sublime_text -n -w $#
Note that I've used the -n flags everywhere. This might not be your cup of tea. If you are using -n then you possibly also want to look at your close_windows_when_empty setting.
Inspired by the OP's answer, I've created a bash wrapper script for Sublime Text that incorporates all your findings and runs on both OSX and Linux.
Its purpose is threefold:
provide a unified subl CLI that works like ST's own subl on OSX: invoke ST without blocking, unless waiting is explicitly requested.
encapsulate a workaround for the waiting-related bug on Linux.
when saved or symlinked to as sublwait, provide a sublwait CLI that automatically applies the --wait and --new-window options so as to make it suitable for use with $EDITOR (note that some programs, e.g. npm, require the $EDITOR to contain the name of an executable only - executables + options are not supported); also makes sure that at least one file is specified.
The only open question is whether the OP's approach to avoiding the race condition - sleep 1 - is robust enough.
Update:
Note that subl on OSX is by default NOT placed in the $PATH - you normally have to do that manually. If you haven't done so, the script will now locate subl inside ST's application bundle; (it tries app names in the following sequence: 'Sublime Text', 'Sublime Text 2', 'Sublime Text 3', first in /Applications, then in ~/Applications.)
Here's the output from running the script with -h:
Multi-platform (OSX, Linux) wrapper script for invocation of Sublime Text (ST)
from the command line.
Linux:
Works around undesired blocking of the shell (unless requested)
and a bug when waiting for specific files to be edited.
Both platforms:
When invoked as `sublwait`, automatically applies the
--wait --new-window
options to make it suitable for use with $EDITOR.
Therefore, you can to the following:
- Name this script `subl` for a CLI that supports ALL options.
(On OSX, this will simply defer to the `subl` CLI that came with ST.)
- Place the script in a directory in your $PATH.
- In the same directory, create a symlink to the `subl` script named
`sublwait`:
ln -s subl sublwait
and, if desired, add
export EDITOR=sublwait
to your shell profile.
Note that if you only use OSX, you can make do with ST's own subl and just save this script directly as sublwait.
Script source:
#!/usr/bin/env bash
# Multi-platform (OSX, Linux) wrapper script for invocation of Sublime Text (ST)
# from the command line. Invoke with -h for details.
[[ $1 == '-h' || $1 == '--help' ]] && showHelpOnly=1 || showHelpOnly=0
[[ $(basename "$BASH_SOURCE") == 'sublwait' ]] && invokedAsSublWait=1 || invokedAsSublWait=0
[[ $(uname) == 'Darwin' ]] && isOsX=1 || isOsX=0
# Find the platform-appropriate ST executable.
if (( isOsX )); then # OSX: ST comes with a bona-fide CLI, `subl`.
# First, try to find the `subl` CLI in the $PATH.
# Note: This CLI is NOT there by default; it must be created by symlinking it from
# its location inside the ST app bundle.
# Find the `subl` executable, ignoring this script, if named subl' as well, or a
# script by that name in the same folder as this one (when invoked via symlink 'sublwait').
stExe=$(which -a subl | fgrep -v -x "$(dirname "$BASH_SOURCE")/subl" | head -1)
# If not already in the path, look for it inside the application bundle. Try several locations and versions.
if [[ -z $stExe ]]; then
for p in {,$HOME}"/Applications/Sublime Text"{,' 2',' 3'}".app/Contents/SharedSupport/bin/subl"; do
[[ -f $p ]] && { stExe=$p; break; }
done
fi
[[ -x $stExe ]] || { echo "ERROR: Sublime Text CLI 'subl' not found." 1>&2; exit 1; }
else # Linux: `sublime_text` is the only executable - the app itself.
stExe='sublime_text'
which "$stExe" >/dev/null || { echo "ERROR: Sublime Text executable '$stExe' not found." 1>&2; exit 1; }
fi
# Show command-line help, if requested.
# Add preamble before printing ST's own help.
# Note that we needn't worry about blocking the
# shell in this case - ST just outputs synchronously
# to stdout, then exits.
if (( showHelpOnly )); then
bugDescr=$(
cat <<EOF
works around a bug on Linux (as of v2.0.2), where Sublime Text,
if it is not already running, mistakenly blocks until it is exited altogether.
EOF
)
if (( invokedAsSublWait )); then
# We provide variant-specific help here.
cat <<EOF
Wrapper script for Sublime Text suitable for use with the \$EDITOR variable.
Opens the specified files for editing in a new window and blocks the
invoking program (shell) until they are closed.
In other words: the --wait and --new-window options are automatically
applied.
Aside from encapsulating this functionality without the need for options
- helpful for tools that require \$EDITOR to be an executable name only -
$bugDescr
Usage: sublwait file ...
EOF
# Note: Adding other options doesn't make sense in this scenario
# (as of v2.0.2), so we do NOT show ST's own help here.
else
cat <<EOF
Multi-platform (OSX, Linux) wrapper script for invocation of
Sublime Text (ST) from the command line.
Linux:
Works around undesired blocking of the shell (unless requested)
and a bug when waiting for specific files to be edited.
Both platforms:
When invoked as \`sublwait\`, automatically applies the
--wait --new-window
options to make it suitable for use with \$EDITOR.
Therefore, you can to the following:
- Name this script \`subl\` for a CLI that supports ALL options.
(On OSX, this will simply defer to the \`subl\` CLI that came with ST.)
- Place the script in a directory in your \$PATH.
- In the same directory, create a symlink to the \`subl\` script named
\`sublwait\`:
ln -s subl sublwait
and, if desired, add
export EDITOR=sublwait
to your shell profile.
Sublime Text's own help:
------------------------
EOF
# Finally, print ST's own help and exit.
exec "$stExe" "$#"
fi
exit 0
fi
# Invoked as `sublwait`? -> automatically apply --wait --new-window options.
if (( invokedAsSublWait )); then
# Validate parameters.
# - We expect NO options - to keep things simple and predictable, we do NOT allow
# specifying additional options (beyond the implied ones).
# - We need at least 1 file argument.
# - As a courtesy, we ensure that no *directories* are among the arguments - ST doesn't support
# that properly (always waits for ST exit altogether); beyond that, however, we leave input
# validation to ST.
if [[ "$1" =~ ^-[[:alnum:]]+$ || "$1" =~ ^--[[:alnum:]]+[[:alnum:]-]+$ ]]; then # options specified?
{ echo "ERROR: Unexpected option specified: '$1'. Use -h for help." 1>&2; exit 1; }
elif (( $# == 0 )); then # no file arguments?
{ echo "ERROR: Missing file argument. Use -h for help." 1>&2; exit 1; }
else # any directories among the arguments?
# Note: We do NOT check for file existence - files could be created on demand.
# (Things can still go wrong - e.g., /nosuchdir/mynewfile - and ST doesn't
# handle that gracefully, but we don't want to do too much here.)
for f in "$#"; do
[[ ! -d "$f" ]] || { echo "ERROR: Specifying directories is not supported: '$f'. Use -h for help." 1>&2; exit 1; }
done
fi
# Prepend the implied options.
set -- '--wait' '--new-window' "$#"
fi
# Finally, invoke ST:
if (( isOsX )); then # OSX
# `subl` on OSX handles all cases correctly; simply pass parameters through.
exec "$stExe" "$#"
else # LINUX: `sublime_text`, the app executable itself, does have a CLI, but it blocks the shell.
# Determine if the wait option was specified.
mustWait=0
if (( invokedAsSublWait )); then
mustWait=1
else
# Look for the wait option in the parameters to pass through.
for p in "$#"; do
[[ $p != -* ]] && break # past options
[[ $p == '--wait' || $p =~ ^-[[:alnum:]]*w[[:alnum:]]*$ ]] && { mustWait=1; break; }
done
fi
if (( mustWait )); then # Invoke in wait-for-specified-files-to-close mode.
# Quirk on Linux:
# If sublime_text isn't running yet, we must start it explicitly first.
# Otherwise, --wait will wait for ST *as a whole* to be closed before returning,
# which is undesired.
# Thanks, http://stackoverflow.com/questions/14598261/making-sublime-text-2-command-on-linux-behave-as-it-does-on-macos-x
if ! pidof "$stExe" 1>/dev/null; then
# Launch as BACKGROUND task to avoid blocking.
# (Sadly, the `--background` option - designed not to activate the Sublime Text window
# on launching - doesn't actually work on Linux (as of ST v2.0.2 on Ubuntu 12.04).)
("$stExe" --background &)
# !! We MUST give ST some time to start up, otherwise the 2nd invocation below will be ignored.
# ?? Does a fixed sleep time of 1 second work reliably?
sleep 1
fi
# Invoke in blocking manner, as requested.
exec "$stExe" "$#"
else # Ensure invocation in NON-blocking manner.
if ! pidof "$stExe" 1>/dev/null; then # ST isn't running.
# If ST isn't running, invoking it *always* blocks.
# Therefore, we launch it as a background taks.
# Invocation via a subshell (parentheses) suppresses display of the
# background-task 'housekeeping' info.
("$stExe" "$#" &)
else # ST is already running, we can safely invoke it directly without fear of blocking.
exec "$stExe" "$#"
fi
fi
fi
On Ubuntu Gnu/Linux 13.04 64-bit:
I just keep subl running pretty much all the time. So my git config has:
core.editor=/usr/bin/subl -n -w
And that's all I need. I save the git commit file with ctrl-s, close the window with ctrl-w and I'm done. But I then have to really close the window by hitting the X in the upper corner... 96% perfect.

Bash config file or command line parameters

If I am writing a bash script, and I choose to use a config file for parameters. Can I still pass in parameters for it via the command line? I guess I'm asking can I do both on the same command?
The watered down code:
#!/bin/bash
source builder.conf
function xmitBuildFile {
for IP in "{SERVER_LIST[#]}"
do
echo $1#$IP
done
}
xmitBuildFile
builder.conf:
SERVER_LIST=( 192.168.2.119 10.20.205.67 )
$bash> ./builder.sh myname
My expected output should be myname#192.168.2.119 and myname#10.20.205.67, but when I do an $ echo $#, I am getting 0, even when I passed in 'myname' on the command line.
Assuming the "config file" is just a piece of shell sourced into the main script (usually containing definitions of some variables), like this:
. /etc/script.conf
of course you can use the positional parameters anywhere (before or after ". /etc/..."):
echo "$#"
test -n "$1" && ...
you can even define them in the script or in the very same config file:
test $# = 0 && set -- a b c
Yes, you can. Furthemore, it depends on your architecture of script. You can overwrite parametrs with values from config and vice versa.
By the way shflags may be pretty useful in writing such script.

Bash script execution with and without shebang in Linux and BSD

How and who determines what executes when a Bash-like script is executed as a binary without a shebang?
I guess that running a normal script with shebang is handled with binfmt_script Linux module, which checks a shebang, parses command line and runs designated script interpreter.
But what happens when someone runs a script without a shebang? I've tested the direct execv approach and found out that there's no kernel magic in there - i.e. a file like that:
$ cat target-script
echo Hello
echo "bash: $BASH_VERSION"
echo "zsh: $ZSH_VERSION"
Running compiled C program that does just an execv call yields:
$ cat test-runner.c
void main() {
if (execv("./target-script", 0) == -1)
perror();
}
$ ./test-runner
./target-script: Exec format error
However, if I do the same thing from another shell script, it runs the target script using the same shell interpreter as the original one:
$ cat test-runner.bash
#!/bin/bash
./target-script
$ ./test-runner.bash
Hello
bash: 4.1.0(1)-release
zsh:
If I do the same trick with other shells (for example, Debian's default sh - /bin/dash), it also works:
$ cat test-runner.dash
#!/bin/dash
./target-script
$ ./test-runner.dash
Hello
bash:
zsh:
Mysteriously, it doesn't quite work as expected with zsh and doesn't follow the general scheme. Looks like zsh executed /bin/sh on such files after all:
greycat#burrow-debian ~/z/test-runner $ cat test-runner.zsh
#!/bin/zsh
echo ZSH_VERSION=$ZSH_VERSION
./target-script
greycat#burrow-debian ~/z/test-runner $ ./test-runner.zsh
ZSH_VERSION=4.3.10
Hello
bash:
zsh:
Note that ZSH_VERSION in parent script worked, while ZSH_VERSION in child didn't!
How does a shell (Bash, dash) determines what gets executed when there's no shebang? I've tried to dig up that place in Bash/dash sources, but, alas, looks like I'm kind of lost in there. Can anyone shed some light on the magic that determines whether the target file without shebang should be executed as script or as a binary in Bash/dash? Or may be there is some sort of interaction with kernel / libc and then I'd welcome explanations on how does it work in Linux and FreeBSD kernels / libcs?
Since this happens in dash and dash is simpler, I looked there first.
Seems like exec.c is the place to look, and the relevant functionis are tryexec, which is called from shellexec which is called whenever the shell things a command needs to be executed. And (a simplified version of) the tryexec function is as follows:
STATIC void
tryexec(char *cmd, char **argv, char **envp)
{
char *const path_bshell = _PATH_BSHELL;
repeat:
execve(cmd, argv, envp);
if (cmd != path_bshell && errno == ENOEXEC) {
*argv-- = cmd;
*argv = cmd = path_bshell;
goto repeat;
}
}
So, it simply always replaces the command to execute with the path to itself (_PATH_BSHELL defaults to "/bin/sh") if ENOEXEC occurs. There's really no magic here.
I find that FreeBSD exhibits identical behavior in bash and in its own sh.
The way bash handles this is similar but much more complicated. If you want to look in to it further I recommend reading bash's execute_command.c and looking specifically at execute_shell_script and then shell_execve. The comments are quite descriptive.
(Looks like Sorpigal has covered it but I've already typed this up and it may be of interest.)
According to Section 3.16 of the Unix FAQ, the shell first looks at the magic number (first two bytes of the file). Some numbers indicate a binary executable; #! indicates that the rest of the line should be interpreted as a shebang. Otherwise, the shell tries to run it as a shell script.
Additionally, it seems that csh looks at the first byte, and if it's #, it'll try to run it as a csh script.

Resources