Suggestion for "errexit" equivalent for $(shell ...)

[paul d]

Hi,

This is my first posting to the GNU Make mailing lists, so please let me know 
if i've contravened any expectations.

## Background

I am a somewhat heavy user of GNU Make, and frequently want to do something 
like the following:

    FILE_SHA = $(shell sha256sum src/file.c)
    deploy-frobnicate:
        rsync src/file.c remote:dist/file-${FILE_SHA}.c

Obviously this is a toy example, but bear with me.  In almost all of my own 
use-cases it makes no sense to continue execution if the command passed to 
$(shell ...) exits nonzero.  In fact, it is often destructive, because empty 
strings being used in subsequent commands frequently risk doing the wrong thing 
for my situation.

I'm aware that there are workarounds such as the pattern of testing 
.SHELLSTATUS after having invoked $(shell), but i find this extremely 
cumbersome.  For example, this can work,

    FOO := $(shell exit 12)
    ifneq ($(.SHELLSTATUS),0)
    $(error Shell failed with $(.SHELLSTATUS))
    endif

However, that only works if i am disciplined and never use lazy assignment -- 
for slow commands that's not a good trade-off.  There's also the approach of 
setting .ONESHELL, but i'm not sure i'd like to rewrite all my Makefiles such 
that that's feasible.  Also, as far as i understand, that won't help me when 
i'm wanting to set a "top-level" variable to the stdout of a shell command, 
which is what i'm most often doing.

## Finally getting to the point

I'm writing to ask whether there's any chance that a patch might be accepted if 
i add a configurable setting (for example, similar to .ONESHELL, maybe an 
option called .SHELL_ERREXIT, to match Bash's `-o errexit` option) which will 
cause any $(shell ...) invocation exiting nonzero to halt Make with an error 
message.  Of course such a flag would have to default to Make's current 
behaviour of entirely ignoring exit codes.

I haven't looked at the code extremely thoroughly, but it looks like i'd have 
to modify func_shell_base in function.c, as well as do whatever needs to be 
done to introduce, document, and handle an option flag such as i've described.  
I'd be happy to have a stab at it, but i wouldn't want to start working on a 
patchset if there's zero chance the maintainers would be interested in the 
feature.

Hope that all makes sense, kind regards, happy holidays,

paul