On Tue, Jul 07, 2020 at 07:41:51AM -0400, Greg Wooledge wrote:
> On Mon, Jul 06, 2020 at 09:45:59PM -0400, Dale R. Worley wrote:
> > bug-b...@trodman.com writes:
> > > foo | tee >(cat >&2) | bar
> >
> > I do wonder how portable >( ... ) is in practice, versus the portability
> > of /dev/stderr. Maybe I worry about the former because I'm not
> > practiced in named-FIFO programming and so think of it as non-universal.
>
> On Linux and BSD systems, >( ) will use a /dev/fd/ entry. On most
> commercial Unix systems, where /dev/fd/ does not exist, it will use a
> named pipe in /var/tmp. On a hypothetical system where neither one is
> available (Microsoft Windows?), I believe it may use a temp file. That
> decision is made at bash's compile time.
>
> The semantics of /dev/fd/* and named pipes are not quite identical, so
> if you're relying on some very *special* mechanisms, then there could
> indeed be portability issues. For most scripts, however, it shouldn't
> matter.
Oh, the other thing I forgot to mention is that the semantics of /dev/fd/*
differ between Linux and BSD. I don't have a reference available off
the top of my head, but at some time in the past few years, it came up
on one of the bug-bash or help-bash mailing lists -- someone's script
acted differently on BSD than it did on Linux because of the different
implementation of /dev/fd/ as used by a process substitution.
People do *weird* shit with bash. The really surprising part is that
they don't *think* it's weird.
