On 09/07/13 18:55, Ondřej Čertík wrote: > I agree with you that removing mpmath brings more cons than benefits > overall. > > However, I just want to understand the root of the particular > problem you mentioned. If sympy removes mpmath, then sympy cannot be > configured to use its own mpmath. In other words, you always do just > "import mpmath as mp". What things go seriously awry? Is it some > incompatibility between mpmath and sympy? > > Or did you mean that if sympy ships its own mpmath (as it does now), > and the user installs systemwide mpmath and you start mixing the two? > In other words, that would be an argument for removing mpmath from > sympy.
If sympy removes mpmath then we would indeed switch to "import mpmath as mp". The issues arise at the moment when people accidentally mix a system-wide mpmath install with the one in sympy. Hence why we only support using the copy embedded in sympy. Unbundling mpmath would mean that we would have to start -- at least in the interim -- supporting the current version of sympy (which will use the system wide mpmath) and legacy versions (via their embedded copy). At this point I would need to check when I import mpmath that I am pulling in the same version being used by sympy and no accidentally using the system version when I should be using the embedded version. Polemically yours, Freddie.
signature.asc
Description: OpenPGP digital signature