Paolo Bonzini <bonzini <at> gnu.org> writes:
> > At any rate, that means we should probably teach m4sh to try "print -r --"
> > as one of its options, prior to "printf %s\\n", during _AS_ECHO_PREPARE.
>
> Maybe, also because in that case we have "print -r -n --" too. But:
>
> 1) these tests do cost a few subshells (which can be as expensive as a
> fork on bash, even if the executed command is a builtin). In the
> attached patch I conditionalized it on ${TMOUT} so that it is not
> executed unless we're on ksh.
Sorry for cross-committing; I checked in a different patch before seeing
your mail, with a very similar idea. Your filter based on TMOUT is different
than my filter on {BASH,ZSH}_VERSION; I could go either way (I tested that
pdksh also supplies both $TMOUT and print).
Things to consider: How likely is TMOUT to be exported in bash (causing a false
positive), vs. BASH_VERSION to be exported in ksh (which causes way more
problems than a spurious export of TMOUT)? On the other hand, using a positive
test (TMOUT being set) vs. a negative test (BASH_VERSION is not set) means that
your version avoids 2 forks on ash or Solaris /bin/sh while mine does not.
Should we document $TMOUT as a reliable way to detect ksh, the way we already
use {BASH,ZSH}_VERSION as reliable witnesses of those two shells?
> 2) I'm not extremely fond of making the M4sh prolog even longer.
> To fix that you could rewrite
> _AS_RUN to use here documents, and change the remaining uses of AS_ECHO
> in error messages to cat. Then you don't need at all anymore
> _AS_ECHO_PREPARE in _AS_SHELL_SANITIZE.
Yes, avoiding $as_echo in the prolog would be nice, if the use of here-docs in
_AS_RUN doesn't add an overall forking penalty. Would you like to write that
patch?
--
Eric Blake
_______________________________________________
Autoconf mailing list
Autoconf@xxxxxxx
http://lists.gnu.org/mailman/listinfo/autoconf
