On Mon, Sep 05, 2011 at 02:35:42PM -0700, Phil Steitz wrote: > On 9/5/11 2:23 PM, Gilles Sadowski wrote: > > On Mon, Sep 05, 2011 at 08:12:53PM +0200, Luc Maisonobe wrote: > >> Le 05/09/2011 15:23, Gilles Sadowski a écrit : > >>> Hello. > >> Hi Gilles, > >> > >>> There might be a typo in the value for the default singularity threshold > >>> value in class "LUDecompositionImpl" (at line 37): Was > >>> 10E-12 > >>> the intended value, or was it > >>> 1e-12 > >>> ? > >> It is probably a typo, but the exact value was certainly chosen > >> almost arbitrarily. However, since it has now been published and > >> changing it may break code, I would suggest to write it 1e-11. > > Done in revision 1165416. > > Should we maybe make the default public?
The fewer "public" things, the better. [As it becomes part of the API, and thus ... (known story).] I think that we should view those simplified constructors as "syntactic sugar" for those users and situations for which the exact value does not matter. If the default is made public, it seems that there was a rationale behind choosing the particular value. Which is not the case here, according to Luc. There are some other occurences in CM where I already indicated my preference for keeping "private" the variable holding the default value, for the same reason. Regards, Gilles --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
