Re: Changing mkinstalldirs to avoid using mkdir -p

[Alexandre Duret-Lutz]

Nathanael Nerode <address@hidden> writes:
> +# Solaris 8's mkdir -p isn't thread-safe!  If you mkdir -p a/b and
> +# mkdir -p a/c at the same time, both will detect that a is missing,
> +# one will create a, then the other will try to create a and die with
> +# a "File exists" error.  So we have to disable this code on Solaris 8.
> +#
> +#case $dirmode in
> +#  '')
> +#    if mkdir -p -- . 2>/dev/null; then
> +#      echo "mkdir -p -- $*"
> +#      exec mkdir -p -- "$@"
> +#    fi
> +#    ;;
> +#  *)
> +#    if mkdir -m "$dirmode" -p -- . 2>/dev/null; then
> +#      echo "mkdir -m $dirmode -p -- $*"
> +#      exec mkdir -m "$dirmode" -p -- "$@"
> +#    fi
> +#    ;;
> +#esac

Does anybody see a way to disable this only when needed?

An idea would be to skip this code if
   1. " -$MAKEFLAGS" matches `" "-[BdeikrRsSw]*j' or `" "---?jobs'
and
   2. `mkdir -p --help' doesn't work.

That way we would always GNU's mkdir -p, and we would ignore
other mkdir -p implementations only in parallel builds.

Unfortunately, the problem is not only in mkinstalldirs, it is
also in AM_PROG_MKDIR_P, and we cannot check $MAKEFLAGS there.
I presume we will end up setting `mkdir_p = mkdir -p' only if
mkdir supports --help.  I find this quite frustrating.

If you consider a system such as Solaris 8, whose make does not
support parallel builds, there is no reason not to use `mkdir
-p'.  The problem only exist when users mix tools, doesn't it?
It's a bit of a shame, we have to disable `mkdir -p' everywhere
because of such setups.

Other ideas?
-- 
Alexandre Duret-Lutz