Re: [boost] [Review] ITL review starts today, February 18th

Subject: Re: [boost] [Review] ITL review starts today, February 18th
From: Jeff Flinn (TriumphSprint2000_at_[hidden])
Date: 2010-02-22 15:39:36

Hartmut Kaiser wrote:
 > Hi all,
 >
 > The formal review of the Interval Template Library (ITL) starts today,
 > February 18th, 2010 and will end February 27th, 2010. ITL is being 
developed by Joachim Faulhaber.
 >
 > ------------------------------------------------
 >
 > About the library:
 >
 > The Interval Template Library (ITL) is a library of interval 
containers that
 > facilitates computations on collections of intervals and associated 
values.
...
Hmm, not much discussion yet. While this is not my review I'll make a 
few comments.

I was able to port at least my necessary portions of split interval map 
to CodeWarrior 9.4 and was successful in using it to generate a column 
coverage graph of what is effectively the columns of possibly large 
sparse matrices.

I would have expected the containers to be templated on an interval 
type, negating the need for run-time determination of boundtype. With 
the current design, a rather heavy interval is required with it's 
boundtype data member. My domain type is a simple int, an index into a 
row of a sparse matrix. Our in-house Interval class is always 
right_open, so there is only the need for a begin and end value leading 
to minimal memory usage. The current library design requires me to 
convert between ITL intervals and my own application's intervals. I'd 
have like to be able to provide a model of an interval concept that the 
ITL container would use.

I wasn't able to find any description of the class invariants for 
interval. For example in our own Interval class the upper/end value must 
be greater than/equal to the lower/begin value which is a precondition 
to the constructor taking a pair of domain values. We do provide a 
static makeNormalized method to produce a proper interval when the order 
of the domain arguments is unknown, which is not very often in our usage 
so you don't pay that cost. In my case I hadn't the need for 'reversed' 
intervals, but I could see that being required for some use cases. Does 
ITL handle that?

Along the same lines I've found little need for the mutators. An 
interval really only needs default and copy constructors, and one taking 
the upper and lower values. All mutations merely construct a new 
interval using calculated values and begin/end values from source intervals.

Jeff