Is `BigFloat` loss of precision intended?

Internals & Design
,
1

when doing any operation on a BigFloat, the precision of the result always reverts back to the default one, instead of getting it from the operands.

Is this behavior intended and if yes, why ?

julia> a=BigFloat("1.1", precision=512);

julia> a.prec
512

julia> (a+a).prec
256

julia> (0+a).prec
256
2

I don’t know, and I don’t know why it’s like this, but maybe

setprecision(512) do
    x = big"1.1"
    x+x
end

solves your problem?

3

I know about setprecision, but I was wondering the point of the precision keyword if it doesn’t propagate

1 Like
4

Hm. base/mpfr.jl defines these functions, and it’s an immediate ccall. I’m not sure there’s much one can do without changing it to BigFloat{512} etc. But I don’t know a lot about it.

5

No, it is not an immediate ccall. It first allocates the result using BigFloat(), which uses the global precision and not the precision of the input (which it could).

PS.: So yes, this is intended. The precision of the result is always the global precision. This is the same behavior as MPFR itself.

2 Likes
6

You must use setprecision(BigFloat, nbits) for precision to work properly (at nbits). The precision keyword is there because over time we have been considering how to rework the whole interface.

1 Like