RUN-PROGRAM and error output

[Elias Pipping]

> On 9 Sep 2016, at 20:35, Jason Miller <jason at milr.com> wrote:
> 
> I did a bit more digging and found some interesting behavior on sbcl.
> 
> Whether or not the output is propagated to a stream in the event of an
> error is dependant on shell vs. exec:
> 
> (uiop:run-program "echo hi >&2; false" :error-output t) ; No output
> (uiop:run-program '("/bin/sh" "-c" "echo hi >&2; false" ) :error-output t) ; "hi\n" on *error-output*
> 
> -Jason

Dear list (I’ve already talked to Jason in person on IRC so that he can savely ignore this e-mail),

the difference in behaviour seen here:

> (uiop:run-program "echo hi >&2; false" :error-output t) ; No output
> (uiop:run-program '("/bin/sh" "-c" "echo hi >&2; false" ) :error-output t) ; "hi\n" on *error-output*

boils down to the use of %run-program vs. %system as a result of the snippet

    (when (stringp command)
      (unless force-shell-suppliedp
        #-(and sbcl os-windows) ;; force-shell t isn't working properly on windows as of sbcl 1.2.16
        (setf force-shell t)))

in run-program.

What currently happens is that we have code of the form

  (let (exit-code)
    (with-program-output …
      (setf exit-code (%check-result (%wait-process-result)))))

This means: If %check-result yields a condition, the with-program-output will not complete. But there’s no good reason for that. All %check-result needs is the exit-code and we keep that around anyway, so that we can just move %check-result out of the with-program-output block. This should not have any negative side effects I’m aware of and leave e.g. :error-output, if it’s a stream, in a deterministic state.

I’ve put up a merge request here:

  https://gitlab.common-lisp.net/asdf/asdf/merge_requests/8

and would welcome comments.


Elias

PS: There are also a few other merge requests at https://gitlab.common-lisp.net/asdf/asdf/merge_requests