Re: [boost] uBLAS formal review

From: Kresimir Fresl (fresl_at_[hidden])
Date: 2002-06-26 14:23:06

I vote for the acceptance of the uBLAS library into Boost.

I've been using uBLAS' real dense and sparse vectors and matrices
(and complex v&m in toy examples) for almost a year. Interface is
rather intuitive (in many cases one can use straightforward
mathematical notation) and I think that uBLAS attains good
compromise between abstraction and efficiency.

It should be stressed that uBLAS provides not only functionality
of `classical' dense Fortan BLAS, but also of sparse BLAS which
is AFAIK not yet well established (eg.: http://math.nist.gov/spblas/).

All examples that I tried, including CLAPACK's BLAS tests with uBLAS
wrappers which Joerg Walter was kind enough to send, can now be
compiled with g++ 2.96, 3.0.4 and 3.1 (switches: -pedantic -ansi -W
-Wall -Wno-unused -Wno-sign-compare), Comeau's como 4.2.45
and 4.3 (in strict mode) and Intel's icc 6.0. For all problems that I had
[either compilation problems (e.g. missing `typename's) or bugs in
computations] Joerg provided fast replies and fixes.

Documentation is well laid out and easy to survey although
rather parsimonious -- sometimes one must read very carefully
(and ocassionally even look into the code, which is not very
simple task ;o). Also, more (and more complex ;o) examples are
needed; short tutorial can facilitate the acceptance of the library.

fres