Macintosh Transperancy (PR#11511)

Simon,

On May 24, 2008, at 10:28 PM, Simon Urbanek wrote:

Bradley,

On May 24, 2008, at 9:07 PM, Bradley Vance wrote:

	The bug is that the point that's being plotted is a single point,  
not a combination of two points, one that's just the outline, and  
one that's just the fill.  It also doesn't look like we'd expect  
when you're debugging a plot.  For example, let's say I create a  
plot, and notice that I can't see some of the points (due to other  
points being drawn on top of them).  So I decide to make them  
transparent, so you can see where points over-lay each other, and  
the relative size/colors/etc of the over-layed points.  Now,  
rather than just changing the opacity of the points/colors I see,  
I'm adding in another color to each point where outline and  
background meet, making the plot more confusing than it needs to  
be...
	In looking at the documentation for bg in points (plot says to  
look at points for bg) it says
bgbackground (fill) color for the open plot symbols given by  
pch=21:25.
and the way it is now, the color is not a fill (ie - inside the  
lines) - it's super-imposed over part of the line.

I think you're still missing the point - it is the fill. I suspect  
you're misunderstanding the effect of using semi-transparent  
strokes. Moreover I get *exactly* the same result using X11 (both  
Mac *and* Linux) as well as other devices (PDF etc.). If you see  
anything else then *that* is a bug.

A point is represented by a shape (path) which is filled and then  
drawn (line stroke). As I was trying to explain earlier the effect  
is that half of the line is in the filled area. Imagine a rectangle  
(0,0)-(1,1). When you fill it, you're filling strictly the area  
between 0 and 1 (in both x and y). However, when you strike let's  
say the line (0,0)-(1,0) with the width 0.25, you are filling the  
area (0,-0.25)-(1,+0.25) which means that half the line is outside  
(y<0) the filled are and another half is inside (y>=0). Given the  
alpha-blending mechanics [target=alpha*color + (1-alpha)*old] it  
implies that if the fill color is different from the background  
color and the line is not opaque, you get different color outside  
and inside the rectangle. You can verify that this is true for all  
devices that support alpha blending. This is how it works and is  
supposed to be, that's not a bug.

While I agree that it should look that way if you draw it as you  
describe, I don't think you're describing the drawing process the  
way people expect.  I don't think FILL means that you draw the shape/ 
fill first, then draw the perimeter (which is what you describe).  I  
think it means that you draw the perimeter first, and then fill-in  
everything that's inside the perimeter.  From my experience, that's  
the way all the drawing programs I've used work.
	If a user were to draw the point themselves, they would draw the  
perimeter, then use a fill tool to fill what's inside the  
perimeter.  They would NOT draw a filled-shape, and then draw the  
empty-shape.
	I also think that the users shouldn't need to be familiar with the  
complexities of drawing each point with different strokes.  Perhaps  
the best way to do this would be to have a separate alpha-parameter  
for the point itself.  This would hide the complexities of drawing  
the point, and allow the users to use regular colors (without alpha- 
extensions) for their points.  I think that most times users want  
transparency, they want the point that was drawn without  
transparency to have transparency, and not to have to think about  
the transparency of individual strokes/etc...

	Also, there's the fact that it does something different on  
different systems.  When I do the same thing at work on a Linux  
box (we don't have R.2.7 yet, we use mostly R2.4 but I've tried it  
on R2.6,

Well, 2.6 didn't have alpha support in X11 AFAIR, so I have no idea  
what you are trying to compare there ...

Look at the next line - I used the Linux version to print out to  
PDF...   You are correct that X11 didn't/doesn't have alpha support,  
but a PDF device has had alpha-support for quite a while.

and we don't have Quartz, but I did output to PDF), it looks like  
a single point whose colors were made transparent, and not a point  
made of two overlapping transparent parts.

In further debug today, I actually found out that what you see  
depends on the PDF-viewer you use.  Since I used acroread at work/ 
linux, and preview at home/mac, I assumed the files they created  
were different.  Actually, it looks like they are the same, but what  
they look like is different.  I verified this by downloading Adobe  
Reader for my mac (latest version), and comparing it to what is  
shown in Preview.   I then took snap-shots of the entire windows to  
a PNG, so that you can see how they draw them differently.  This may  
actually need to ticket to go to Apple.  But as you can see from  
these snapshots, I think that how Adobe (PDF-spec-creator) shows the  
plot is correct (First is Adobe Reader, 2nd is Preview).  On my work- 
Linux box, acroread shows the same as Adobe Reader on my mac.  My  
guess is that Adobe uses FILL to mean fill-inside-the-perimeter- 
shape, the way I described above, and Preview uses it to mean fill- 
inside-the-shape-then-draw-the-perimeter that you're advocating.   
Myself, I think that adobe is correct...

<Rplot_AdobeReader.png><Rplot_Preview.png>


	Here's 3-points at the top, zoomed in (Adobe first, Preview 2nd).   
Personally, I find that the first looks much cleaner / less noisy  
than the second.  Maybe it's just because that's what I expect...
<Rplot_AdobeReader_zoom.png><Rplot_Preview_zoom.png>

	As I implied in my original ticket, I'd like to be able to choose  
whether to have it look the way it does now on the mac, and the  
way it does on Linux.  This might actually be something that needs  
to be addressed further by the core team, as that opacity is  
planning to be added for X11() for the next revision (and  
hopefully, will work into PNG and other device types that can  
support it)...
	Perhaps the best way to implement the options/choices/etc is to  
add a separate alpha/opacity/whatever options that would control  
the opacity of each object it plots.  Thus, if you define an  
opacity to either col or bg, it would work the way it does on the  
Mac (ie - the colors are opaque and can thus you can see the other  
point-color through the top one).  However, if you have a point- 
opacity defined, the entire point will get the opacity, the way it  
is currently done on Linux.
	Or, you can just say that in order to get it look the way that it  
currently does on the mac, you need to plot each point twice, once  
with no bg, and once with no line...

	Also, I don't think this is a critical bug.  It's just something  
I noticed that behaves differently on what I have at work (Linux)  
and what I have at home (mac).  I'm impressed with R, and am  
thinking of getting more involved in the r-project community, so  
figured I'd start with something I noticed was different across  
platforms...

Again, I think you are missing something - it behaves consistently  
on Linux and Mac in the latest release and it's not a bug.

Again, I think it is a bug, though the root of the bug may actually  
be in Apple's QUARTZ code.  One major reason I think it's a bug is  
that for the way R currently draws it, there is NO-WAY for R to  
create the plot that Adobe Reader shows in Quartz.  However, if R  
draws it the way I think is right, then R can create both plots, as  
the way it currently draws it can be emulated by plotting each point  
twice (once for fill-only, and once for perimeter-only)...
	In short the bug is this...  I cannot make 2-color-points  
(perimeter/fill), and make a point that looks exactly like it, but  
is transparent...

Cheers,
Simon

On May 24, 2008, at 6:17 PM, Simon Urbanek wrote:

Brad,

can you, please, let us know what is the "bug" here? If you paint  
semi-transparent line over the same filled shape (same vertices),  
clearly half of the line will be inside and and half outside of  
the shape, so each part will have a different color - that is the  
whole point of semi-transparency. Therefore I fail to see any bug  
or unexpected behavior here.

Cheers,
Simon


On May 24, 2008, at 10:55 AM, brdvance at gmail.com wrote:

Full_Name: Brad Vance
Version: 2.7.0
OS: 10.5.2
Submission from: (NULL) (71.123.195.202)


Problem : When drawing transperant points with lines thick- 
lines(lwd>1) + fill
(pch=21-25), it looks like the line is transperantly overlayed  
over the fill,
making it look like 2-lines surround the fill (each a different  
color).  I
actually think that this is a nice OPTIONAL way to draw the  
points (even as
default), but since I could not discover how to turn it off, I  
decided to issue
the ticket.  In this option can see on the default-quartz device  
as well as
PDF-device.

Solution : Have the entire point-object designed/drawn as if  
there is
no-transparency, then make the composite object transparent.    
Optionally figure
out an option to determine whether to apply transperancy as a  
group (to the
entire object) or separately (as it is now).

Code :

df <- data.frame(X=rnorm(50),Y=rnorm(50))
plot 
(X 
~ 
Y 
,data 
= 
df 
,pch 
=c(21,22,23,24,25),cex=5,col="#00FF0055",bg="#FF000055",lwd=6)

Thanks,
-Brad Vance