Re: Faster token table lookups

[Hans Aberg]

At 14:42 +0100 0-12-19, Akim Demaille wrote:
>> Token table lookups in the parser that Bison generates (cf. the
>> manual, the %token_table option & sec 4.2.1) can be speeded up
>> if Bison writes a token table with the names sorted
>> alphabetically
...
>I don't understand your point.  AFAIK we never lookup for anything in
>this table.  It's used the other way round: we have the token numbers,
>we want to know their `human' names, and that a simple matter of
>reading a cell of an array.
>
>So I don't know what you are talking about.

It would merely be a convenience to those that want to make a lot of name
table lookups. I think there is a suggestion one can implement it in the
lexer function yylex: instead of returning a macro number directly, it
makes a table lookup, and returns that number to Bison.

For example,
  int lookup(const char*);  // Make a string lookup & return the Bison macro
                            // number
  %token implies "=>"
and instead of
  yylex() {
    ...
    return implies;
    ...
  }
one writes
  yylex() {
    ...
    return lookup("=>");
    ...
  }

This makes the macro name "implies" unnecessary, and one oculd have skipped
it, if
  %token "=>"
would have been legal.

This could be convenient if speed isn't crucial, if one does not want to
invent a lot of macro names.

-- I haven't used that setup myself. This is how I figure it might be used
though.

-- Normally, this table would not be used at all, and it should be either
an option to have it written out, or one could let the linker remove it.

  Hans Aberg