I've just been reviewing a really useful PR ( https://github.com/matplotlib/matplotlib/pull/1531) from Pierre Haessig which speeds up the drawing of non-visible artists by bringing the following line to the top of the LineArtist's draw method: if self.get_visible() is False: return This *does* fix the problem (and will fix the problem for all other artists if applied in the same way), but it relies on a developer remembering the rule of thumb that they must always start their draw method with these two (simple) lines. Additionally, testing this functionality is actually quite hard without resorting to timing the execution. It made we wonder if there was a better approach to fixing this. Having a decorator to do this for you is a good idea, except that a developer would need to remember to decorate their subclass' draw method, so the next level up is to use a metaclass to *always* wrap the draw method with the "if visible" lines. An example of implementing this (apologies if the code doesn't come out well in the email): class ArtistMeta(type): def __new__(cls, classname, bases, class_dict): # replace the draw method with one which short-circuits if self.visible is False draw_method = class_dict['draw'] def draw(self, *args, **kwargs): if self.visible is False: print 'draw **not** called with visible={}'.format(self.visible) return else: return draw_method(self, *args, **kwargs) class_dict['draw'] = draw return type.__new__(cls, classname, bases, class_dict) class Artist(object): __metaclass__ = ArtistMeta def __init__(self, visible=True): self.visible = visible def draw(self, renderer=None): print 'draw called with visible={}'.format(self.visible) return 'foobar' class SubArtist(Artist): def draw(self, renderer=None): print "subclass' draw method" return Artist.draw(self, renderer=renderer) With the following results: >>> a = Artist().draw('wibble') draw called with visible=True >>> b = Artist(False).draw('wibble') draw **not** called with visible=False >>> c = SubArtist(True).draw('wibble') subclass' draw method draw called with visible=True >>> d = SubArtist(False).draw('wibble') draw **not** called with visible=False In my eyes this makes testing the functionality possible without timing (and is therefore an improvement), but I wanted to know how others felt about the approach, and in particular, using more metaclasses in matplotlib (a simple tutorial which I found useful: http://www.voidspace.org.uk/python/articles/metaclasses.shtml). Cheers, Phil
The problem is that I don't think we can do this for all artists. Some may need to create groupings, or push and pop state even if they are "invisible". For instance, this is used in the SVG backend to create named groupings (possibly empty) that are referenced from Javascript to provide interactivity. I think I'd rather keep this to the contained solution in the PR and not try to generalize it beyond that. If we did want to generalize, this would only apply to "leaf node" artists, and not artists that simply exist to contain other artists -- and conceivably we could implement that using either a decorator or explicit chaining to a base class, but in any event it would have to be a manual process to determine which artists this would apply to. We could insert a class in the heirarchy of "ConcreteArtist" (or somesuch) to handle this. Mike On 11/26/2012 06:17 AM, Phil Elson wrote: > I've just been reviewing a really useful PR > (https://github.com/matplotlib/matplotlib/pull/1531) from Pierre > Haessig which speeds up the drawing of non-visible artists by bringing > the following line to the top of the LineArtist's draw method: > > if self.get_visible() is False: > return > > This *does* fix the problem (and will fix the problem for all other > artists if applied in the same way), but it relies on a developer > remembering the rule of thumb that they must always start their draw > method with these two (simple) lines. Additionally, testing this > functionality is actually quite hard without resorting to timing the > execution. > > It made we wonder if there was a better approach to fixing this. > Having a decorator to do this for you is a good idea, except that a > developer would need to remember to decorate their subclass' draw > method, so the next level up is to use a metaclass to *always* wrap > the draw method with the "if visible" lines. > > An example of implementing this (apologies if the code doesn't come > out well in the email): > > > class ArtistMeta(type): > def __new__(cls, classname, bases, class_dict): > # replace the draw method with one which short-circuits if > self.visible is False > draw_method = class_dict['draw'] > def draw(self, *args, **kwargs): > if self.visible is False: > print 'draw **not** called with > visible={}'.format(self.visible) > return > else: > return draw_method(self, *args, **kwargs) > class_dict['draw'] = draw > > return type.__new__(cls, classname, bases, class_dict) > > > class Artist(object): > __metaclass__ = ArtistMeta > > def __init__(self, visible=True): > self.visible = visible > > def draw(self, renderer=None): > print 'draw called with visible={}'.format(self.visible) > return 'foobar' > > > class SubArtist(Artist): > def draw(self, renderer=None): > print "subclass' draw method" > return Artist.draw(self, renderer=renderer) > > > > > With the following results: > > > >>> a = Artist().draw('wibble') > draw called with visible=True > > >>> b = Artist(False).draw('wibble') > draw **not** called with visible=False > > >>> c = SubArtist(True).draw('wibble') > subclass' draw method > draw called with visible=True > > >>> d = SubArtist(False).draw('wibble') > draw **not** called with visible=False > > > > In my eyes this makes testing the functionality possible without > timing (and is therefore an improvement), but I wanted to know how > others felt about the approach, and in particular, using more > metaclasses in matplotlib (a simple tutorial which I found useful: > http://www.voidspace.org.uk/python/articles/metaclasses.shtml). > > > Cheers, > > Phil > > > > > > > > > > > > > > ------------------------------------------------------------------------------ > Monitor your physical, virtual and cloud infrastructure from a single > web console. Get in-depth insight into apps, servers, databases, vmware, > SAP, cloud infrastructure, etc. Download 30-day Free Trial. > Pricing starts from 795ドル for 25 servers or applications! > http://p.sf.net/sfu/zoho_dev2dev_nov > > > _______________________________________________ > Matplotlib-devel mailing list > Mat...@li... > https://lists.sourceforge.net/lists/listinfo/matplotlib-devel
On 2012年11月26日 7:12 AM, Michael Droettboom wrote: > The problem is that I don't think we can do this for all artists. Some > may need to create groupings, or push and pop state even if they are > "invisible". For instance, this is used in the SVG backend to create > named groupings (possibly empty) that are referenced from Javascript to > provide interactivity. I think I'd rather keep this to the contained > solution in the PR and not try to generalize it beyond that. > > If we did want to generalize, this would only apply to "leaf node" > artists, and not artists that simply exist to contain other artists -- > and conceivably we could implement that using either a decorator or > explicit chaining to a base class, but in any event it would have to be > a manual process to determine which artists this would apply to. We > could insert a class in the heirarchy of "ConcreteArtist" (or somesuch) > to handle this. I think we should be rather conservative about this sort of thing. Sometimes it is better to just explicitly put the two lines in each method than to come up with machinery to do it for you. Each level of depth in an inheritance hierarchy or "meta" chain is an additional level of complexity for someone reading the code. And if someone forgets to put in those lines, the penalty is typically from small to nil; but if they are put in automatically by fancy methods, and they are not really wanted or something else goes wrong, it can make debugging painful. Eric > > Mike > > On 11/26/2012 06:17 AM, Phil Elson wrote: >> I've just been reviewing a really useful PR >> (https://github.com/matplotlib/matplotlib/pull/1531) from Pierre >> Haessig which speeds up the drawing of non-visible artists by bringing >> the following line to the top of the LineArtist's draw method: >> >> if self.get_visible() is False: >> return >> >> This *does* fix the problem (and will fix the problem for all other >> artists if applied in the same way), but it relies on a developer >> remembering the rule of thumb that they must always start their draw >> method with these two (simple) lines. Additionally, testing this >> functionality is actually quite hard without resorting to timing the >> execution. >> >> It made we wonder if there was a better approach to fixing this. >> Having a decorator to do this for you is a good idea, except that a >> developer would need to remember to decorate their subclass' draw >> method, so the next level up is to use a metaclass to *always* wrap >> the draw method with the "if visible" lines. >> >> An example of implementing this (apologies if the code doesn't come >> out well in the email): >> >> >> class ArtistMeta(type): >> def __new__(cls, classname, bases, class_dict): >> # replace the draw method with one which short-circuits if >> self.visible is False >> draw_method = class_dict['draw'] >> def draw(self, *args, **kwargs): >> if self.visible is False: >> print 'draw **not** called with >> visible={}'.format(self.visible) >> return >> else: >> return draw_method(self, *args, **kwargs) >> class_dict['draw'] = draw >> >> return type.__new__(cls, classname, bases, class_dict) >> >> >> class Artist(object): >> __metaclass__ = ArtistMeta >> >> def __init__(self, visible=True): >> self.visible = visible >> >> def draw(self, renderer=None): >> print 'draw called with visible={}'.format(self.visible) >> return 'foobar' >> >> >> class SubArtist(Artist): >> def draw(self, renderer=None): >> print "subclass' draw method" >> return Artist.draw(self, renderer=renderer) >> >> >> >> >> With the following results: >> >> >> >>> a = Artist().draw('wibble') >> draw called with visible=True >> >> >>> b = Artist(False).draw('wibble') >> draw **not** called with visible=False >> >> >>> c = SubArtist(True).draw('wibble') >> subclass' draw method >> draw called with visible=True >> >> >>> d = SubArtist(False).draw('wibble') >> draw **not** called with visible=False >> >> >> >> In my eyes this makes testing the functionality possible without >> timing (and is therefore an improvement), but I wanted to know how >> others felt about the approach, and in particular, using more >> metaclasses in matplotlib (a simple tutorial which I found useful: >> http://www.voidspace.org.uk/python/articles/metaclasses.shtml). >> >> >> Cheers, >> >> Phil
On Mon, Nov 26, 2012 at 12:23 PM, Eric Firing <ef...@ha...> wrote: > On 2012年11月26日 7:12 AM, Michael Droettboom wrote: > > The problem is that I don't think we can do this for all artists. Some > > may need to create groupings, or push and pop state even if they are > > "invisible". For instance, this is used in the SVG backend to create > > named groupings (possibly empty) that are referenced from Javascript to > > provide interactivity. I think I'd rather keep this to the contained > > solution in the PR and not try to generalize it beyond that. > > > > If we did want to generalize, this would only apply to "leaf node" > > artists, and not artists that simply exist to contain other artists -- > > and conceivably we could implement that using either a decorator or > > explicit chaining to a base class, but in any event it would have to be > > a manual process to determine which artists this would apply to. We > > could insert a class in the heirarchy of "ConcreteArtist" (or somesuch) > > to handle this. > > I think we should be rather conservative about this sort of thing. > Sometimes it is better to just explicitly put the two lines in each > method than to come up with machinery to do it for you. Each level of > depth in an inheritance hierarchy or "meta" chain is an additional level > of complexity for someone reading the code. And if someone forgets to > put in those lines, the penalty is typically from small to nil; but if > they are put in automatically by fancy methods, and they are not really > wanted or something else goes wrong, it can make debugging painful. I think you and Mike are skirting around a key point here. You can always add the line if you need it, but if you don't need it (or can't use it), by use of a metaclass, there's no way to "opt out" so to speak. I'll also add that we don't need to add any more indirection (i.e. another Python function call) to our drawing stack--we really need to be doing everything possible to take every last millisecond out of the call to draw(). Ryan -- Ryan May Graduate Research Assistant School of Meteorology University of Oklahoma
Note that we already use a decorator for a similar purpose (allow_rasterization). Also, please note that the "draw" method is not just for drawing things. There are other things being done within the draw method, and I think some of them still need to be done even though the artist is invisible. My personal inclination on this issue is to refactor the "draw" method, the only method being called during the drawing time. But, yes there are sideeffects. Regards, -JJ On Tue, Nov 27, 2012 at 3:40 AM, Ryan May <rm...@gm...> wrote: > On Mon, Nov 26, 2012 at 12:23 PM, Eric Firing <ef...@ha...> wrote: > >> On 2012年11月26日 7:12 AM, Michael Droettboom wrote: >> > The problem is that I don't think we can do this for all artists. Some >> > may need to create groupings, or push and pop state even if they are >> > "invisible". For instance, this is used in the SVG backend to create >> > named groupings (possibly empty) that are referenced from Javascript to >> > provide interactivity. I think I'd rather keep this to the contained >> > solution in the PR and not try to generalize it beyond that. >> > >> > If we did want to generalize, this would only apply to "leaf node" >> > artists, and not artists that simply exist to contain other artists -- >> > and conceivably we could implement that using either a decorator or >> > explicit chaining to a base class, but in any event it would have to be >> > a manual process to determine which artists this would apply to. We >> > could insert a class in the heirarchy of "ConcreteArtist" (or somesuch) >> > to handle this. >> >> I think we should be rather conservative about this sort of thing. >> Sometimes it is better to just explicitly put the two lines in each >> method than to come up with machinery to do it for you. Each level of >> depth in an inheritance hierarchy or "meta" chain is an additional level >> of complexity for someone reading the code. And if someone forgets to >> put in those lines, the penalty is typically from small to nil; but if >> they are put in automatically by fancy methods, and they are not really >> wanted or something else goes wrong, it can make debugging painful. > > > I think you and Mike are skirting around a key point here. You can always > add the line if you need it, but if you don't need it (or can't use it), by > use of a metaclass, there's no way to "opt out" so to speak. > > I'll also add that we don't need to add any more indirection (i.e. another > Python function call) to our drawing stack--we really need to be doing > everything possible to take every last millisecond out of the call to > draw(). > > Ryan > > -- > Ryan May > Graduate Research Assistant > School of Meteorology > University of Oklahoma > > > ------------------------------------------------------------------------------ > Monitor your physical, virtual and cloud infrastructure from a single > web console. Get in-depth insight into apps, servers, databases, vmware, > SAP, cloud infrastructure, etc. Download 30-day Free Trial. > Pricing starts from 795ドル for 25 servers or applications! > http://p.sf.net/sfu/zoho_dev2dev_nov > _______________________________________________ > Matplotlib-devel mailing list > Mat...@li... > https://lists.sourceforge.net/lists/listinfo/matplotlib-devel > >