Re: [Libmesh-devel] Why not use boost::bimap to implement string_to_enum ?

[Roy S. <roy...@ic...>]

On Wed, 18 Nov 2009, John Peterson wrote:

> On Wed, Nov 18, 2009 at 9:26 PM, Liu Kai <liu...@gm...> wrote:
>> Hi all,
>>
>>  I find that things like string_to_enum and enum_to_string in src/utils
>> are exactly bi-direction map, which has been well implemented in boost
>> library. So why not use the boost::bimap to implement these
>> functionality, instead of implement the same thing by hand?
>>
>>  I just want to know if the reason is boost::bimap has some potential
>> problems, or is it a historical reason?
>
> We don't use boost at all in the library.  There are probably
> thousands of places where we could and should.

Note also that boost::bimap postdates string_to_enum/enum_to_string by
a year or two.  Looks like Ben's options in 2005 were using
codeproject::bimap, or building his own bimap on top of
boost::multi_index_container...

I'm not trying to advocate a total Not-Invented-Here attitude, but
tracking APIs that might be still in flux is a pain.  That's
worthwhile for something like PETSc; not so much for a single little
lookup table.

Of course, that was 2005; this is 2009.  It looks like the Boost folks
have been happy with the bimap interface for a couple years now, and
they generally do very good work.
---
Roy