[m-dev.] fix for bytecode generator

[Bert Thompson]

|Hi Bert,
|
|> Is it possible to avoid the test "FunctorList = [Char]"  by changing
|> the declaration of sym_name prog_data.m to:
|> 
|> :- type sym_name 	---->	unqualified(string)	% Can remove?
|> 			;	qualified(module_specifier, string)
|> 			;	char_const(char).	% Added
|
|If we were to add a specialised representation of characters, it would go
|into the cons_id data type, as was done for byte_cons_id.

Yes, that makes much more sense. 

|> This simplifies the code, but means changing other code that uses
|> `unqualified(string)' to represent a char constant.
|
|Should characters be treated differently to other enumeration types?

For the sake of minimality, no. For efficiency, yes. 

|Currently the only places characters are treated separately are
|code_util__cons_id_to_tag and bytecode_gen__map_cons_id. If we
|did change the representation there would need to be a similar check
|in the code to do the conversion during mode analysis (overloading
|isn't resolved until after type-checking), 

Yep, the change is not localised.

|so I don't think the change
|would actually save much.

You're right. It doesn't save much at all. 

Miniscule though the saving may be, it's still true that less-than-optimal
fixes tend to beget other less-than-optimal-er fixes. Anyway, this isn't 
a big deal, and I don't foresee Mercury dying a death of a thousand cuts 
any time soon. 8^)

Bert