Jeff Prothero on elidable terminators issue from before LogFest

[Guy Steele <cbmvax!uunet!Think.COM!gls>]

Prothero's note is very clear and provides a good framework for
discussing the question, but it does not yet answer my objections.
The strongest objection is this: while he has argued the possibility
of parsing the augmented language by LALR(1) techniques, he has not
provided an effective procedure for constructing the necessary tables.
Therefore he has not addressed the question of how complex these tables
must be compared to the tables for the unaugmented language.  Indeed,
his argument has not yet ruled out the possibility that the tables
must be infinite in size!  For he argues that the parser can implicitly
regenerate "a]Bc" from "aBc"; but it must be able to do this for any
string "a" such that "aBc" in not in the original language.  There may
be an infinite number of such strings "a".  He has not yet shown that
a finite number of table entries will be able to cover all the possible
decisions to be made.  I conjecture that in the general case a finite
table will not suffice.  My original arguments suggested that we ought
to answer the question definitively for the specific case of Lojban,
rather than merely handwaving.

--Guy Steele