csh bug is probably quite old, see 6th Edition shell

[forsyth]

panda!fluffy!keith complained that the following failed in a
version of csh:

	cat /etc/passwd | (cat) &

If he has the csh source (I haven't), I think he will find that
it is actually a bug derived from 6th Edition shell(!), and that csh
has borrowed some of the older shell's methods, if not its code.
The following explanation based on the 6th Edition source might help
to find it in csh.

In the function `execute', there is code that
handles parenthesised commands: those with tree type TPAR.
The code propagates the flag FINT, which controls suppression of
signals, to the descendent of TPAR that describes the
parenthesised command, before a recursive call to execute that command:

		if(t[DTYP] == TPAR) {
			if(t1 = t[DSPR])
				t1[DFLG] =| f&FINT;
			execute(t1);
			exit(255);
		}

Fine, but it should also include FPIN, which states that
input should be taken from a pipe.  The f&FINT should be f&(FINT|FPAR).
Otherwise the following code to handle `&' will close the
standard input and open /dev/null, since FPIN will not be set:

		if((f&FINT)!=0 && t[DLEF]==0 && (f&FPIN)==0) {
			close(0);
			open("/dev/null", 0);
		}

This code implements the convention that
	command&
will receive /dev/null as standard input by default, to prevent it stealing
terminal input.  This might well be different in csh owing to job control.
Csh will also undoubtably use structure references instead of array
references like t[DLEF]!

A copy of the old shell was in /usr/src/cmd/osh.c on 7th Edition systems.
Just for fun, I compiled it on a VAX for a PDP-11, and it runs quite
happily in PDP-11 compatibility mode, except of course for
	% echo *
	glob: cannot execute
	%
$ size11 bin/v6sh
5080+700+1396 = 7176b = 0x1c08b
It's amazing what can be done in 7k (it doesn't grow), with 844 lines of
code which even includes a few comments (but don't press your luck).

	ukc!minster!forsyth

PS.  net.bugs.v6??