Return-Path: Message-Id: <9204021458.AA17304@relay1.UU.NET> Date: Thu Apr 2 11:11:49 1992 Reply-To: "Mark E. Shoulson" Sender: Lojban list From: "Mark E. Shoulson" Subject: MEX, Operator Precedence X-To: lojban@cuvmb.cc.columbia.edu To: John Cowan , Eric Raymond , Eric Tiedemann In-Reply-To: Chris Handley's message of Thu, 2 Apr 1992 10:27:49 GMT+1200 Status: RO X-From-Space-Date: Thu Apr 2 11:11:49 1992 X-From-Space-Address: cbmvax!uunet!CUVMA.BITNET!LOJBAN Chris Handley suggests that "if this [{bo}] merely raised it [the precedence] above the _preceding_ operator, (even if that had already been "bo"ed) if may work," thus supporting left-precedence in general but *right* precedence among {bo}ed operators. To my understanding, that is the current state of affairs. So unmarked operators group left to right, bo-ed operators have higher precedence than those and group right-to-left. So Chris will be happy, but we still need a more general plan. Jacques Guy proposes we forget about getting precedence any better than what we have and concentrate on getting RPN. From what I've seen, it seems that in Lojban, infix order is unmarked, *as* *is* Polish notation (prefix notation), and Reverse Polish (postfix) is easily gotten by the marker {fu'a}. So we have RPN already, and that's happy. But it still doesn't solve our problem, merely proposes a workaround. Let's face it, people are going to want to have their nice familiar infix (try getting someone who isn't used to it to try an HP calculator!). I'm not sure I like John's sub-rules and sub-selma'o proposal. It seems like overcomplication of the grammar. Then again, its results are certainly the best that could be hoped for. Needless to say, it is important that those rules be modifyible by {ti'o} or whatever. When I first heard of {bo} I thought that maybe you could indicate as many shades of higher and higher precedence as you wanted simply by repeating the {bo} (this was before I looked at the grammar or anything...), so you could get very high-precedence addition with {*li pa su'ibobobobo pa}. This plan is imperfectly general, counterintutive (the *more* words between the operators, the *tighter* they bind!?), and probably impossible to put into the grammar. Also, people want their precedences built in when they say things. I'll have to think about this whole problem some more. ObControversy: We already do it unofficially, and people are prone to do it all the time, so we might as well officially recognize {h} as a possible, albeit discouraged and deplored, alloglyph of {'}. It wouldn't preak the A-V isomorphism or anything, and as I said, it should be discouraged. But we already use it in selma'o names and I think I've seen it used in UNIX filenames which, though they can have quotes, do ugly things to the shell if they do. So let's make our lives a little easier... ~mark