Re: more wishes for chrismas

[Emin Gun Sirer <egsirer@phoenix.Princeton.EDU>]

>From: Dave Mason <dmason@plg.uwaterloo.ca>
>Date: 	Sat, 5 Dec 1992 11:21:13 -0500
>
>> Date: Fri, 4 Dec 1992 09:02:51 -0500
>> From: Alan Watson <alan@oldp.astro.wisc.edu>
>> 
>> There are a few characters which only make sense at certain places --
>> like "=" which has to be quoted in dd commands -- but at least the
>> current rules have a simplicity about them.
>
>This is the only spot where I get bitten occasionally and wouldn't
>mind a change.  If local assignments were only allowed at the
>beginning of lines, then:
>	dd count=1 bs=512
>wouldn't cause a problem.  In fact, looking at the grammar in the rc
>manual, assignments *are* only allowed at the beginning of lines, and
>I don't see the rule that causes the problem.

Yes, promotion is inconsistent. For example:
;~ a sdfdsfg
[status 1]
;ls file1 ~ file3
;ls file1 = file3
[syntax error]

;in arg1
syntax error
;cmnd in asfd
;(elem1 in elem3)

So the rule in the latter part is that the keyword gets promoted to a string
unless it is the first word on a line. But the same thing is not true of "=",
which does not get promoted anywhere. However, "~" gets promoted around
*in the lexer*, which is not a good thing.

The solutions I see, in order of desirability as far as I'm concerned:

	1. Fix dd and awk (I was at Bell Labs when someone changed the plan9
dd to take "-infile fname" and "-outfile" as opposed to "infile=fname". People
were furious that a lot of their scripts had to be fixed. But it only
happens once and saves a lot of headache later on).

	1. Promote '=' as if it was a keyword. Put '~' promotion in the
grammer along with '='.

	1. Promote all syntax error causing keyword constructs to strings 
(this'll require following multiple paths through the parser but is doable)
and retry.

	Duff's rc paper praises sh for using recursive descent, but criticizes
it because a lot of the routines change their behaviour according to some
flags. The counterpart to this in a yacc-based program is the special
treatment of certain tokens in the lexer. Yes, it's nice that rc has a
yacc grammer that anyone can understand, but does everyone know when a given
character represents token TWIDDLE or CHAR and is this treatment uniform
for all characters that are more or less the same ??

	Gun.