OSH is a Unix shell designed to run existing shell scripts. More precisely, it's
It's designed to be "stricter" than other shells. To avoid programs that don't behave as intended,
"Batch" programs are most likely to run unmodified under OSH. Interactive
programs like .bashrc
and bash completion scripts may require small changes.
This manual covers the differences between OSH and other shells. It leaves
the details of each construct to the help
builtin and the Quick
Reference (Warning: both are incomplete). It also
doesn't cover the Oil language, which is a newer part of the
Oil project.
Existing educational materials for the Unix shell apply to OSH, because they generally don't teach the quirks that OSH disallows. For example, much of the information and advice in BashGuide can be used without worrying about which shell you're using. See the end of this manual for more resources.
The releases page links to source tarballs for every release. It also links to the documentation tree, which includes this manual.
After running the instructions in INSTALL, run:
mkdir -p ~/.config/oil
osh_history
there, to store your command history.oshrc
there.On startup, the interactive shell sources only ~/.config/oil/oshrc
.
Other shells have a confusing initialization sequence involving many
files (original). It's very hard to tell when and if
/etc/profile
, ~/.bashrc
, ~/.bash_profile
, etc. are executed.
OSH intentionally avoids this. If you want those files, simply source
them
in your oshrc
.
$LANG
may not get set without
/etc/profile
. Add source /etc/profile
to your oshrc
may solve this
problem.~/.config/oil/oshrc
, symlink it to ~/.oshrc
.I describe my own oshrc
file on the wiki: How To Test
OSH.
Strict options disallow certain parts of the language with fatal runtime errors.
They are used like this:
shopt -s strict-array # Set this option. I want more fatal errors.
shopt -u strict-array # Unset it. Ignore errors and keep executing.
You can turn all of them on or off at once:
shopt -s all:strict
shopt -u all:strict
This line turns all strict modes on, but is portable to other shells:
shopt -s all:strict 2>/dev/null || true # suppress errors
strict-argv
. Empty argv
arrays are disallowed, since there's no practical
use for them.
x=''; $x
results in a fatal error.strict-array
. No implicit conversions between string an array. In other
words, turning this on gives you a "real" array type. (NOTE: Only partially
implemented.)
strict-control-flow
. break
and continue
outside of a loop are fatal
errors.
strict-errexit
. The errexit
setting is inherited in subshells, AND it
can cause fatal errors in the parent process.
For example, echo 0; echo $(touch one; false; touch two); echo 3
will print
0
and touch the file one
.
false
, andThis is even stricter than bash 4.4's inherit_errexit
, which stops at false
in the command sub, but keeps running the parent process.
strict-word-eval
. More word evaluation errors are fatal.
${s: -1}
and ${s: 1 : -1}
result in a fatal error. (NOTE: In array slices, negative start indices are
allowed, but negative lengths are always fatal, regardless of
strict-word-eval
.)${#s}
) and slices.On by default:
strict-arith
. Strings that don't look like integers cause a fatal error in
arithmetic expressions. NOTE: This option may be removed if no scripts rely on
the old, bad behavior.
The -n
flag tells OSH to parse the program rather than executing it. By
default, it prints an abbreviated abstract syntax tree:
$ bin/osh -n -c 'ls | wc -l'
(command.Pipeline children:[(C {(ls)}) (C {(wc)} {(-l)})] negated:F)
You can also ask for the full text
format:
$ bin/osh -n --ast-format text -c 'ls | wc -l'
(command.Pipeline
children: [
(command.Simple
words: [
(word.Compound
parts: [(word_part.Literal
token:(token id:Lit_Chars val:ls span_id:0))]
)
]
)
(command.Simple
words: [
(word.Compound
parts: [(word_part.Literal
token:(token id:Lit_Chars val:wc span_id:4))]
)
(word.Compound
parts: [(word_part.Literal
token:(token id:Lit_Chars val:-l span_id:6))]
)
]
)
]
negated: F
spids: [2]
)
This format is subject to change. It's there for debugging the parser, but sophisticated users may use it to interpret tricky shell programs without running them.
OSH_HIJACK_SHEBANG
This environment variable can be set to the path of a shell. Before OSH executes a program, it will inspect the shebang line to see if it looks like a shell script. If it does, it will use this shell instead of the one specified in the shebang line.
For example, suppose you have myscript.sh
:
#!/bin/sh
# myscript.sh
./otherscript.sh --flag ...
and otherscript.sh
:
#!/bin/sh
# otherscript.sh
echo 'hello world'
Then you can run myscript.sh
like this:
OSH_HIJACK_SHEBANG=osh osh myscript.sh
and otherscript.sh
will be executed with OSH rather than the /bin/sh
.
Note that osh
appears twice in that command line: once for the initial
run, and once for all recursive runs.
(This is an environment variable rather than a flag because it needs to be inherited.)
--debug-file
Print internal debug logs to this file. It's useful to make it a FIFO:
mkfifo _tmp/debug
osh --debug-file _tmp/debug
Then run this in another window to see logs as you type:
cat _tmp/debug
Related:
OSH_DEBUG_DIR
environment variable is the inherited version of
--debug-file
. A file named $PID-osh.log
will be written in that
directory for every shell process.--xtrace-to-debug-file
flag sends set -o xtrace
output to that file
instead of to stderr
.OSH_CRASH_DUMP_DIR
This is implemented, but a JSON library isn't in the release build.
The completion API is modeled after the bash completion API
However, an incompatibility is that it deals with argv
entries and not
command strings.
OSH moves the responsibility for quoting into the shell. Completion plugins should not do it.
compadjust
builtin. Derived from a cleanup of the
bash-completion
project.0
for success.1
for runtime errors. Examples:
echo foo > out.txt
and out.txt
can't be opened.fg
and there's not job to put in the foreground.2
for parse errors. This means that we didn't attempt to do
anything, rather than doing something and it failing. Examples:
echo $(
.read -z
.0
for true, and 1
for false. Example:
test -f foo
and foo
isn't a file.126
for permission denied when running a command (errno EACCES
)127
for command not foundShell programs should be encoded in UTF-8 (or its ASCII subset). Unicode characters can be encoded directly in the source:
echo 'μ'
or denoted in ASCII with C-escaped strings, i.e. $''
:
echo $'[\u03bc]'
(This construct is preferred over echo -e
because it's statically parsed.)
The data they operate on should also be UTF-8 / ASCII.
For example, the length operator ${#s}
and slicing ${s:1:3}
perform UTF-8
decoding. Decoding errors are fatal if shopt -s strict-word-eval
is on.
The GNU iconv
program converts text from one encoding to another.
Also see Notes on Unicode in Shell.
External: