[Gnuplot-info] Filtering + histogram

[<kev...@fr...>]

Given the following data (simplified example):

3 2
2 3
3 2
3 2

I can plot a vertical-bar histogram with 1-unit buckets with the data in column 2 as follows:

hist(x,width)=width*floor(x/width)+width/2.0
set boxwidth 1
plot 'gp' u (hist($2,1)):(1.0) sm f w boxes

producing the expected result, with one bar of height 3 from 2-3 and one bar of height 1 from 3-4.

Now I want to show data only from rows with 3 in column 1.  Most info I've seen on this recommends using ?: to turn unwanted rows into garbage, e.g., 1/0:

plot 'gp' u (hist((($1==3)?$2:(1/0)),1)):(1.0) sm f w boxes

Unfortunately, I don't get what I expect (a single bar of height 3 from 2-3).  Instead, I get a bar of height 2 with a horizontal line in the middle at y=1.  I think it's actually superimposing two bars, one of height 1 and one of height 2, because if I add another row '3 2' at the end, the height goes up to 3 (instead of 4) and the line is still at y=1.  It's as if the "bad" data in row 2 "breaks up" the data and plot doesn't realize row 1 and rows 3-4 really belong to the same bucket, and it draws two bars, one in front of the other.

Is there another way to do this?  (Would prefer within gnuplot, rather than "filter the data before going into gnuplot.)  Or is this a known problem recently fixed?  (I'm using 4.6.0.)