No-op shell command that preserves $? - sh

Question

Using only the features of the POSIX shell, is there a "simple command" that does nothing and does not change the value of $? People usually describe : as a no-op command for shell, but this always sets $? to zero, so it's not what I want.
This is needed by a program that generates shell scripts. In several places it needs to emit an if-then-else block
if CONDITION
then
TRUE-COMMANDS
else
FALSE-COMMANDS
fi
but, because of further complications that I'd rather not try to explain right now, it cannot reliably tell whether TRUE-COMMANDS and FALSE-COMMANDS are empty. An empty then- or else-clause will be a shell syntax error.
: can be put at the beginning of the then-clause to take care of TRUE-COMMANDS being empty, but that won't work for FALSE-COMMANDS because it clobbers $? and the FALSE-COMMANDS might want to look at the value $? was set to by the condition. For the same reason, : can't be put after FALSE-COMMANDS—the code after the if-then-else might want to see $? from the last operation within the if-then-else.
Bonus points if you can avoid:
Forking: (exit $?) does the job, but there are so many of these conditional blocks in a generated script that it produces a measurable slowdown.
Functions: given nop () { return $? } then nop does the job, but due to more complications that I'd rather not get into, it's not practical to have nop defined sufficiently early for all of the places that would need it.

Answer 1

The easiest would be to make use of a simple assignment. Instead of using :, do _rc=$?.
if condition; then
[ list-true ] # optional
_rc=$?
else
[ list-false ] # optional
_rc=$?
fi
( exit $_rc ) # optional
list-post-if
Using this variable _rc, you have stored the exit status of the last executed command, whether this is condition or the last command in list-true or list-false.
The arguments in favour of this method is the low overhead of an assignment.
The argument against is the need to at least rewrite list-post-if to make use of _rc instead of $?.
If the latter is not possible, or too tedious, you might concider to add a (exit $_rc) just after the conditional statement. This, however, requires a sub-shell, but it is only one.

Answer 2

Don't let anyone tell you there are only functions and sub-shells.
Can you create or distribute another tiny file? If so, you can create a file with just
return $?
and then source it as an "empty" command keeping the exit status:
$ echo 'return $?' > keepstatus
$ ls foobar
ls: fooobar: No such file or directory
$ . ./keepstatus
$ echo $?
2
$ sleep 100
^C
$ . ./keepstatus
$ echo $?
130
Doesn't fork, doesn't use functions, no extra variable, keeps the status and is as POSIXly as they come.
There's even a fourth way I can see, when I sacrifice the bonus points for forking and assuming, since you are in autoconf m4 territory, that finding and using a host compiler is a breeze.
cat > keepstatus.c <<EOF
#include <stdlib.h>
int main(int argc, char **argv) {
return argv[1] ? atoi(argv[1]) : 0;
}
EOF
$CC -o keepstatus keepstatus.c
Then use /path/to/keepstatus $?.

Answer 3

It’s certainly not possible in a POSIX-compliant shell to write a command that does not touch $?. By reading through IEEE Std 1003.1-2017 vol. 3 chapter 2 ‘Shell Command Language’, we learn that:
§2.8.2 says that ‘[e]ach command has an exit status’ (so no command has no exit status).
§2.5.2 ‘Special Parameters’ says that $? ‘[e]xpands to the decimal exit status of the most recent pipeline (see Section 2.9.2)’.
§2.9.2 ‘Pipelines’ says that a pipeline is a sequence of one or more commands, optionally preceded (as a whole) by !, with commands joined by |. About exit status, it says
If the pipeline does not begin with the ! reserved word, the exit status shall be the exit status of the last command specified in the pipeline. Otherwise, the exit status shall be the logical NOT of the exit status of the last command.
§2.9 ‘Shell Commands’ defines a ‘command’ and says that ‘[u]nless otherwise stated, the exit status of a command shall be that of the last simple command executed by the command’. A ‘command’ may be any of the following:
A simple command (§2.9.1), which is simply an external program, a shell built-in or a variable assignment (without a command name). The former will of course return the exit status of the command executed. About the latter, the specification says:
If there is no command name, but the command contained a command substitution, the command shall complete with the exit status of the last command substitution performed. Otherwise, the command shall complete with a zero exit status.
A pipeline, described by §2.9.2 mentioned above.
A compound-list (§2.9.3), which is a sequence of one or more {sequences of one or more {sequences of one or more pipelines (see above), joined by && or ||}, each terminated by ; or & (with a final ; optional)}, joined by newline characters. A compound-list returns the exit status of the last pipeline executed synchronously; asynchronously-executed pipelines (those terminated by &) set exit status to zero. None of the sequences may be empty, which guarantees at least one pipeline will be executed.
A compound command (§2.9.4), which is either:
A subshell or a braced compound-list (§2.9.4.1), which returns the exit status of the underlying compound-list (see above).
A conditional construct (case, §2.9.4.3 or if, §2.9.4.4) or a loop (for, §2.9.4.2; while, §2.9.4.5; until, §2.9.4.6). If the body of this construct is not executed, it returns an exit status of zero.
A function definition (§2.9.5), which returns an exit status of zero if the definition is accepted, and a non-zero status otherwise.
And finally, §2.9.4.4 ‘The if Conditional Construct’ defines the condition and the body of the if construct to be a compound-list, described by §2.9.3 mentioned above. This means that the body of an if construct will always contain at least one command that will clobber $?. The specification does not even leave any wiggle room for ‘implementation-defined behaviour’.
All of the above means that the only way to write a no-op command that preserves $? is by reading out the value in $? and returning it back. There are three constructs in POSIX shell that are capable of this, two of which are already mentioned in the question body: an exit from a subshell or a function invocation.
The third is .: the shell sourcing statement mentioned in #Jens’ answer. By sourcing a script containing just a return "$?" command, the value of $? can be preserved. However, this requires you to arrange that a suitable script is found at some known location, which I assume to be just as inconvenient as ensuring that a no-op function has been defined early enough in the file (if not in fact more).
If you bend the strict POSIX requirement a little, even this can be overcome, though:
. /dev/stdin <<EOF
return "$?"
EOF
/dev/stdin is not a POSIX feature, but is widely available; it is explicitly listed in IEEE Std 1003.1-2017 vol. 1 §2.1.1 as an extension. The above snippet has been tested to work in bash 5.0.8, dash 0.5.11 and busybox sh 1.30.1 on Linux (with appropriately set-up /dev and so on).