SYNOPSIS
arx (-h | -[?] | --help)
arx (-v | --version)
arx shdat (-b <size>)? (-o <output file>)? < input
arx shdat (-b <size>)? (-o <output file>)? <input file>+
arx tmpx <option,archive>* (//+ <command> (//+ <option,archive>*)?)?
DESCRIPTION
A UNIX executable is a simple thing -- a file the kernel can execute,
one way or another, via an interpreter or directly as object code.
Every executable induces a family of executions -- instances of execu-
tion with different command line arguments, with different files in the
working directory and with different environment variables present.
The arx tool captures the parameters of an execution and encodes them
as an executable, making for easy, consistent transfer and repetition
of a particular run. The generated executable ensures that each run
occurs in a freshly allocated temporary directory, with only the
desired files in scope; it uses traps to ensure the cleanup of this
directory; and its format is a simple POSIX shell script, relying on
just a few shell tools.
DEPENDENCIES
The arx tool relies on the presence of sed, tr, date, head, tar, hex-
dump and sh. When unpacking tar archives, it may use the -j or -z
(bzip2 and gzip, respectively) options of tar. Scripts have been tested
with dash and the GNU tools as well as the sh implementation and user-
land tools that are part of busybox.
APPLICATION
The tmpx subcommand of arx offers a variety of options for bundling
code and a task to run. The shdat subcommand exposes the lower-level
functionality of encoding binary data in a shell script that outputs
that binary data, using HERE documents and some odd replacement rules
for nulls.
Scripts generated by tmpx and shdat may be fed to sh over STDIN to exe-
cute them. This can be helpful when using ssh and sudo to set up an
execution context; for example:
arx tmpx ... | ssh user@host.com sudo sh
Scripts generated by tmpx will pass their arguments to the contained
script or command. To pass arguments when piping to sh, use -s:
arx tmpx ... | ssh user@host.com sudo sh -s a b c
Some arguments to the generated script will be treated specially,
namely, --extract, --no-rm and --no-run. Please see the section on
Passing Arguments, below, for more information about these options.
ARX COMMANDLINE PROCESSING
For all subcommands, when options overlap in their effect -- for exam-
ple, setting the output with -o -- the rightmost option takes prece-
dence. Whenever -h, -? or --help is present on the command line, help
is displayed and the program exits.
When paths are specified on an arx command line, they must be quali-
fied, starting with /, ./ or ../. This simplifies the command line syn-
tax, overall, without introducing troublesome ambiguities.
TMPX
The tmpx subcommand bundles together archives, environment settings and
an executable or shell command in to a Bourne-compatible script that
runs the command or executable in a temporary directory, after having
unpacked the archives and set the environment.
Any number of file path arguments may be specified; they will be inter-
preted as tar archives to include in bundled script. If - is given,
then STDIN will be included as an archive stream. If no arguments are
given, it is assumed that no archives are desired and only the command
and environment are bundled.
The temporary directory created by the script is different for each
invocation, with a name of the form /tmp/tmpx-<timestamp>-<randomhex>.
The timestamp format is %Y.%m.%dT%H.%M.%SZ, in UTC. One happy conse-
quence of this is that earlier jobs sort ASCIIbetically before later
jobs. After execution, the temporary directory is removed (or not,
depending on the -rm[10!_] family of options).
-rm0, -rm1, -rm_, -rm!
By default, the temporary directory created by the script
will be deleted no matter the exit status status of the task.
These options cause a script to be generated that deletes the
temporary directory only on success, only on failure, always
(the default) or never.
--shared
Causes the temporary directory to be identified by a hash of
the ARX archive, instead of by date and time. Different runs
of the same archive will share the same directory. Note that
this implies shared state and every disadvantage thereof.
-b <size>
Please see the documentation for this option, shared with
shdat, below.
-o <path>
By default, the generated script is sent to STDOUT. With -o,
output is redirected to the given path.
-e <path>
Causes the file specified to be packaged as the task to be
run. A binary executable, a Ruby script or a longish shell
script all fit here.
In addition to these options, arguments of the form VAR=VALUE are rec-
ognized as environment mappings and stored away in the script, to be
sourced on execution.
Without -e, the tmpx subcommand tries to find the task to be run as a
sequence of arguments delimited by a run of slashes. The following
forms are all recognized:
arx tmpx ...some args... // ...command...
arx tmpx ...some args... // ...command... // ...more args...
arx tmpx // ...command... // ...some args...
The slash runs must have the same number of slashes and must be the
longest continuous runs of slashes on the command line. The command
will be included as-is in a Bourne shell script.
SHDAT
The shdat subcommand translates binary data in to a shell script which
outputs the binary data. The data is encoded in HERE documents in such
a way that data without NULs is not changed and that data with NULs is
minimally expanded: about 1% for randomish data like compressed tar-
balls and about 10% in pathological cases.
The shdat subcommand can be given any number of paths, which will be
concatenated in the order given. If no path is given, or if - is given,
then STDIN will be read.
-b <size>
The size of data chunks to place in each HERE document. The
argument is a positive integer followed by suffixes like B,
K, KiB, M and MiB, in the manner of dd, head and many other
tools. The default is 4MiB. This is unlikely to make a dif-
ference for you unless the generated script is intended to be
run on a memory-constrained system.
-o <path>
By default, the generated script is sent to STDOUT. With -o,
output is redirected to the given path.
EXAMPLES
# Installer script that preserves failed builds.
git archive HEAD | bzip2 | arx tmpx -rm0 - // make install > go.sh
# Now install as root; but don't log in as root.
cat ./go.sh | ssh joey@hostname sudo /bin/sh
# Variation of the above.
git archive HEAD | bzip2 | arx tmpx -rm0 - -e ./build-script.py > go.sh
# Bundle an instance of an application with DB credentials and run it.
arx tmpx -rm! ./app.tbz ./stage-info.tgz // rake start | ssh ...
# Get dump of linking info for build that works here but not there.
arx tmpx ./server-build.tgz LD_DEBUG=files // ./bin/start | ssh ...
# Test out Cabal source distribution of this package:
arx tmpx // 'cd arx-* && cabal configure && cabal build' // \
-rm0 ./dist/arx-0.0.0.tar.gz | sh
PASSING ARGUMENTS TO GENERATED SCRIPTS
The scripts generated by tmpx treat some arguments as special, internal
options, to allow for inspecting them should there be a need to deter-
mine their contents.
--extract
Unpack the data in the present directory and do nothing else.
--no-rm
Run the script as normal but do not delete the generated tem-
porary directory.
--no-run
Unpack into a temporary directory as normal but do not run
the user's command.
To prevent arguments from being specially treated, use // in the argu-
ment list:
a-tmpx-script.sh --no-rm // a b c --extract
In the above example, --extract will be passed to the inner command, in
the same way as a, b, c. The following example causes ab, c and --no-rm
to be printed one after another, each on their own line.
arx tmpx // printf "'%s\n'" '"$@"' | sh -s // ab c --no-rm
NOTES
The timestamp is not the common ISO 8601 format, %Y-%m-%dT%H:%M:%SZ,
because of software and build processes that attach special meaning to
colons in pathnames.
BUGS
The command line parser offers no hints or help of any kind; it fails
with the simple message "argument error". The two most common mistakes
I make are:
o Not qualifying paths with /, ./ or ../.
o Not specifying a subcommand (tmpx or shdat).
