Re: Bug in 5.4.rc2

[Ethan A M. <me...@uw...>]

On Tuesday, 23 June 2020 11:24:24 PDT Dima Kogan wrote:
> I finished a patch just as you wrote this. The patch is attached. I
> tried several operations, and it appears to do what I want, and not
> break anything. The code has a lot of confusion about the meaning of
> xlow/xhigh/ylow/yhigh in arguments to store2d_point() and again, in the
> *cp structure this function fills. The previous code was passing the x,y
> extents to store2d_point(), but if you do that, you run out of
> arguments, and can't pass the y extents even if you wanted to. This
> patch computes the extents inside store2d_point(), so there are
> arguments left over. 

That won't work.
The eventual extent of a cicle along the y axis will not be known
until after the scaling on both x and y is complete.  For an auto-scaled
plot that will not be true until all the plot components, including functions,
have been processed.  There is not enough information yet at the time
store2d_point() is called.

"with circles" is not the only plot style for which autoscaling must
be delayed till later.  That is the reason for the extra routines
box_range_fiddling()  impulse_range_fiddling() histogram_range_fiddling()
and so on.   However for those plot styles the extra routine can
be called following the individual plot component "with <whatever>".
To handle circles you'd have to wait until _all_ plot components had
contributed to the autoscaling.  That could be done, but it would introduce
a new stage of processing that has not been needed before this.
And even so it would be imprecise, because if you must increase
the y range to fit the circles that will itself increase the y-radius of
the circles so you would need to iterate or estimate the asymptote.

So not impossible, but more difficult than autoscaling other plot styles.

	Ethan