From mboxrd@z Thu Jan  1 00:00:00 1970
From: fche@redhat.com (Frank Ch. Eigler)
To: cgen@sources.redhat.com
Subject: Re: escaping in syntax strings
Date: Mon, 26 Mar 2001 08:49:00 -0000
Message-id: <o5snk01nnf.fsf@toenail.toronto.redhat.com>
References: <20010326081015.D13900@redhat.com> <o5wv9c1ojl.fsf.cygnus.local.cgen@toenail.toronto.redhat.com>
X-SW-Source: 2001-q1/msg00210.html

I wrote:

: [...]
: By the way, why do you need a "$" in the syntax string?  Are you
: confusing register name keywords or hexadecimal constants with
: the markers that actually belong into the syntax strings?

I just found what Ben was talking about.  It seems that some
instructions in the assembly language for his target includes dummy
operands that make explicit the implicit (not-encoded) operands.  For
example, given a register keyword set such as "$0 .. $7", there are
some instructions which are hard-wired to use only $0 as an input or
output, in addition to other inputs/outputs.  Rather than leave the $0
implicit, this assembly language expects the programmer to specify it.
Say,
        compare $0,$2,$3
instead of
        compare $2,$3

So, Ben is trying to implement this in the assembler by including the
literal string "$0" in the syntax bytes of the instruction.  This is
not too bad, though a possibly better way would be to associate a
synthetic cgen operand with that slot, and to give it a parser that
accepts only "$0", and emits a helpful error message for anything
else.

- FChE