On Sep 8, 2009 2:41am, Rutger ter Borg &lt;rutger@terborg.net&gt; wrote:<br />&gt; <br />&gt; What would be nice is to be able to work with expression trees. In this case<br />&gt; I&#39;m referring to Proto&#39;s semantic actions, or expression transformation.<br />&gt; E.g., rewrite an expression to make use of BLAS calls (or, some other<br />&gt; &quot;backend&quot;). Pattern matching on the expression, rewrite, and transparently<br />&gt; an other expression is used. All in compile time.<br /><br />It definitely would be nice.  I have found that as soon as you start down the path of generic and functional programming in C++, you start wishing you could mix/match this stuff.  For example, lambdas binding to specific locations in vectors, auto-differentiation of vectors/matrices, etc.  While I am not a good enough programmer to understand the details (and hope I never do...), it sure smells like having a consistent expression template library would make this stuff  fit together seamlessly.<br /><br />That said, I fear that if we worry too much about the long-term goal, we won&#39;t make the necessary patches to make ublas a good enough solution to hold us over.<br /><br />If the GLAS guys are watching this list, I would love to hear their thoughts on the longterm vision and when they think we might be in a position to evolve boost&#39;s matrix library to the next level.  Based on my experiences, it sure seems to me that GLAS is the right approach, but it is a hell of a lot of work.<br /><br />-Jesse