Re: Eliminating conflicts of parenthesized subexpressions

[Hans Aberg]

At 08:53 +0200 2004/09/14, Akim Demaille wrote:
>>>>> So a simple form of type checking might also do the trick here;
>>>>> however, type checking is better not done in the grammar, I think.
>>>
>>>> I looked a bit in Aho, Sethi & Ullman, "Compilers..." (the Dragon
>>>> book), p.  351 ff., and they do it in the actions.
>>>
>>> Choose your "better" and then the discussion will be meaningful.  If
>>> you're not interested in performances, Vincent is definitely right.
>>
>> I find it hard to parse this one :-): If one should be able to catch
>> semantic type errors, then the actions must contain some code
>> indicating
>> how those types should be computed. Then that must be checked at
>> sometime.
>> What is the performance issue you have in your mind here?
>
>One pass is faster than two.  Multiplying passes over the text/AST is
>more modular, but more CPU, and less performance.  That's why most
>industrial strength compilers are not very modular.

I am sure Vincent had a multipass in his mind. I did not, in any case.

>>> And let's face it at least once: The Dragon Book is obsolete (another
>>> edition is expected though).
>>
>> People say that, but it probably depends on what one wants to do. It
>> probably only wants to provide the basics, anyhow.
>
>Even for the basics I do not recommend it to newcomers.
>
>> And I figure computer
>> language constructors and implementers do better to not stare to
>> deeply in
>> books, but use their own mind, as well hanging out in suitable
>> newsgroups
>> and mailing lists.
>
>There are many more books, some much better for beginners, IMHO.
>>
>> I recently read a book that for some reason was on my bookshelf:
>>   David A. Watt, "Programming Language Concepts and Paradigms".
>> The interesting thing with this book, is even though it is also
>> obsolete,
>> being written in 1989 and oblivious about the developments in the
>> 1990'ies,
>> it has some insightful remarks about the limitations of the computer
>> languages of the day, which are still valid today in newer computer
>> languages. So, one can get some information out of also older books.
>>
>The dragon books also has nice pieces of information.  If you have spare
>time to read longish books, you will be rewarded.  But if you don't,
>let's take a more efficient book.

You clearly have a textbook in your mind, perhaps for a course or
something. I have found it suitable as a reference on a few topics, on the
basic level. If you have some suggestion in your mind, you may drop it off
here, and/or to John Levine, the comp.compilers moderator, so he can put
into the comp.compilers FAQ.

  Hans Aberg