matplotlib-devel Mailing List for matplotlib

On Monday 10 April 2006 12:26, John Hunter wrote:
> >>>>> "Darren" == Darren Dale <dd...@co...> writes:
>
> Darren> Sorry about that, I didnt see that one coming. This
> Darren> should be fixed in svn 2278.
>
> We seem to be repeatedly bitten by this type of bug in texmanager:
> first with subprocess and now with the dvipng version check. I think
> it would be worthwhile to refactor backend_ps and backend_agg to only
> import the texmanager when needed. There is no reason a person with
> usetex=False should fail.
>
> I suggest something like (in backend_ps and backend_agg).
>
> class RendererPS:
> def __init__(self, blah):
> self.texmanager = None
>
> def get_texmanager(self):
> if self.texmanager is not None: return self.texmanager
> from matplotlib.texmanager import TexManager
> self.texmanager = TexManager()
>
>
> and then only call get_texmanager if rcParams['text.usetex']. We
> could also put this functionality in backend_bases.RendererBase
Ok, I just committed this. get_texmanager is in backend_bases.
> BTW, do we need renderer specific texmanagers. Would anything break
> if we used module level texmanagers? Something tells me I've thought
> about this before and concluded we needed renderer (ie figure) level
> managers, but am now drawing a blank.
I need to think about this. I'm pretty sure we can use module-level 
texmanagers, since texmanagers dont keep any state info (no args to __init__, 
no set_* functions ...). On the other-hand, module-level texmanagers may be a 
problem for someone who wants to do rcParams['text.usetex']=True after they 
have imported pylab.