[Firebird-devel] Vulcan: Question about parsing large integers

[Tom C. <tom...@sa...>]

In Vulcan, the SQLParse::yylex code doesn't handle very large numbers  
that could be represented as floating point values.  For example, in  
ISQL, something like the following inserts into a column that is  
DOUBLE PRECISION result in tokenization errors:

     insert into foo values ( 1.000000000000000000000000 );
     insert into foo values ( 1000000000000000000000000  );

This is essentially because the digits are accumulated in a 64-bit  
integer which is (correctly) predicted to overflow, resulting in a  
perceived precision error.  Both are representable as floating point  
values easily enough, but the behavior of the lexer prevents this.   
I'd like to make a change that essentially recognizes insignificant  
trailing zeroes in an other-wise precise representation and process  
the result as a FLOAT_NUMBER type.

However, since this is changing the way numbers are handled (at least  
at parse time) for these very large integers, I thought I'd better  
check with the community to make sure there's not a legacy/standards/ 
compatibility issue here that I need to know about.    Testing the  
fix appears to work, but additionally if there are issues with  
numeric conversion or representation later in runtime after parsing  
that I should be testing for, I'd appreciate any pointers.

(A quick scan of the Firebird2 code seems to indicate that this same  
issue exists there as well, though I am far less familiar with that  
code base and may be misinterpreting the code behavior from simple  
inspection.)


-- 
Tom Cole, Platform R&D, SAS Institute Inc.
"There are 10 kinds of people.  Those who understand binary, and  
those who don't."