Re: syntactic completion

[Hans Aberg]

At 11:39 +0200 2004/08/02, Vincent Zweije wrote:
>||  In general, though, there seems to be a growing need for
>||  more exact error recovery, which suggests Bison would
>||  benefit to in the future support LR(1). This completions
>||  request is just one in that batch.
>
>Given that LALR grammars will do the job for most of the parsing
>tasks and that it produces smaller tables, its use is not going
>to go away.
>
>However with the coming of larger hardware, LR parsing becomes
>feasible, and it has certain advantages over LALR, such as more
>exact error reporting and, it turns out now, completion.
>
>What I mean to say is, yes, LR has its useful sides too, but the
>compactness of LALR will still make it the preferred technique
>for the majority of parsing applications.

The main point is that LR(1) seems to be important when one needs more
exact error recovery. If one merely wants to capture more grammars, then
one can with Bison now use the GLR version of LALR(1) (see the manual). In
view of this, the motivation of LR(1) because importance in error recovery
is somewhat unexpected.

>  Just don't try to
>make bison an LR-only parser.  (As if I thought you would.)

Bison has been modularized to some extent in order to admit more than one
parsing algorithm. So the most likely future is that the default Bison will
stay compatible with POSIX Yacc, which then includes LALR(1), but then
additional options are added onto that.

  Hans Aberg