MasterLee wrote:
Given an language L1 that is faster as an language L2 in some tasks an slower in other tasks. An language L3 will be faster as booth language if (optimal) translators are available that can translate code in L3 to L1 and L2.
I'm not sure code generation and optimizations are what's being discussed. First of all, given very intelligent compilers, they should be able to output equivalent yet optimal code regardless of language provided they are both Turing-complete---L3 is rendered useless if we start talking about optimal compilers. We mostly want to talk about language capabilities and productivity. The question is whether a good L3 could be designed...
Language support for every possible programming paradigm is not a good idea, as no one likes a bloated language (e.g., C++ doesn't even try that and look how huge it is: the latest C++11 draft, which became FIDS, is 1353 pages long). The other extreme is clearly not good way to go either (e.g., assembly is the most capable but the least productive).
Next, let's remember that we don't (yet) have such intelligent compilers and that we design languages in such a way that they will make it easy for compiler implementors to come up with efficient implementations.
Is your L3 feasible?