[antlr-interest] philosophy about translation

[Anthony W. Youngman]

In message <D677B863-2607-4CBF-B907-32E176F9C846 at cs.usfca.edu>, Terence 
Parr <parrt at cs.usfca.edu> writes
>Anyway, now make it 400 of those rules.  Now, add a bunch of  arbitrary 
>actions that test "am I in this function and did I see this  variable 
>defined before by walking backwards?" in addition to these  syntactic 
>rules and I am certain that your brain must be very large  to figure 
>out the emergent behavior of all of these rules.   This is  the 
>standard Prolog programming problem.  Andy confirmed for me on  the 
>phone that debugging why the huge list of rules is not giving you  the 
>right translation requires tools that generate nice HTML reports  and 
>lots of thinking sometimes.  Further, the constant threat of  infinite 
>rule-application loops frightens me.    Perhaps he can give  us a 
>better idea of how often he hits a landmine.  I may be incorrect  to 
>focus on this problem.

Linus and git may be an interesting example here ...

Never mind why he had to write it, Linus wrote git as a rush job source 
code control system. And did it very quickly because he basically threw 
all previous received wisdom out the window and started again.

It stores files based on their MD5 hash. It's theoretically possible for 
several files to share a hash, and Linus said to himself "loads of sccs 
code is based on dealing with the possibility of a collision, and seeing 
as collisions are rare this code rarely gets tested".

So he made the *engineering* decision that (1) you never corrupt data, 
so you never overwrite a file already in the sccs, and (2) the chances 
of having a collision were less than the chances of the error-recovery 
code going wrong, so ...

Should you get an MD5 collision, it's a fatal error. The merge fails and 
the user is left to sort out the mess by hand. It seems a pretty poor 
way to handle things, but at the end of the day it's a rational decision 
- attempting to handle a corner case automatically is a high risk 
strategy and no rational person would try. The trouble is people aren't 
rational, much as we like to think we are :-)

Cheers,
Wol
-- 
Anthony W. Youngman - anthony at thewolery.demon.co.uk