[Maxima-discuss] Use of cl:pi in numeric code

[Raymond T. <toy...@gm...>]

In a discussion with Barton and Stavros, it came up that ecl was getting 
some type errors in running the testsuite. In particular, |elliptic_e| 
and |lambert_w| were producing long-float values.

In maxima, float values are always supposed to be double-floats. 
However, in ecl and clisp, |cl:pi| is a long-float because both ecl and 
clisp have a long-float type separate from double-float.

This means any code that uses |cl:pi| causes the use of long-floats. 
That's not want we want to happen.

I'm not sure what the best solution is, but we do have |bigfloat:%pi| to 
return a value of pi of the appropriate type. However, that's a method 
so there's some cost to using this. I'm not sure if a compiler macro 
could be written to remove the cost.

An alternative is to change |bigfloat:%pi| to be a function. Then I can 
definitely write a compiler macro to return the appropriate type if the 
arg has a known value. A quick grep shows that currently |bigfloat:%pi| 
is always called with a variable instead of a constant.

There are also quite a few places where we do |(coerce pi 'flonum)| and 
|(float pi 1e0)|. These should be updated to use |bigfloat:%pi| if we 
choose to do this.

This might be ok. I'll do some experimentation.

​