I want to stabilize the source tree again. My goal is to ensure that all optimizations introduced so far work and make at least two big pieces of software work, Maxima and OpenAxiom. In the process, though, I try will check whether some other libraries break due to the type inference and optimization algorithms.
As far as I can tell, after turning the type propagator messages into style warnings, Maxima works again. Executing "make check" produces some new errors that I was not aware of, but I am not sure whether this comes from further development in the Maxima sources.
One thing that I have not yet implemented is the requested "fix" to one special function where greater accuracy is expected for complex-integer input, while ECL interprets that it has to work in single float precision. I am still doubtful about the semantics. Furthermore, the code for that extra precision in one marginal case has a great impact in the performance of the function, adding more branches, which is why I am still hesitating to do a quick "fix".
I would welcome some other warnings about stuff that does not work but which previously did.
Instituto de Física Fundamental, CSIC
c/ Serrano, 113b, Madrid 28006 (Spain)