I recently stumbled upon an example where Gnuplot produces (at least for me) an unexpected result so I was wondering whether it is indeed the desired behavior. The setup is following:
set terminal pngcairo
set output plot3d?"plot1.png":"plot0.png"
unset key
set xtics nomirror
set ytics nomirror
set xr [0.5:2.5]
set yr [-0.5:4.5]
set multiplot layout 1,2
$data0 << EOD
1 2 1
2 4 1
1 3 1
1 0 1
2 2 1
1 1 1
EOD
$data1 << EOD
1 2 1
2 4 1
1 3 1
1 0 1
2 2 1
1 1 1
EOD
if(plot3d){
set view map
splot $data0 w lp, \
'' u 1:2:3:(sprintf("%d", $0)) w labels offset char 0, char -0.5
splot $data1 w lp, \
'' u 1:2:3:(sprintf("%d", $0)) w labels offset char 0, char -0.5
}else{
plot $data0 w lp, \
'' u 1:2:(sprintf("%d", $0)) w labels offset char 0, char -0.5
plot $data1 w lp, \
'' u 1:2:(sprintf("%d", $0)) w labels offset char 0, char -0.5
}
The two data blocks differ only in white space, $data0 has two blank lines, while $data1 just one. With plot3d=1, the script produces (5.0 patchlevel 6)

Interestingly, also points 1 and 4 in data block $data1 are connected. This edge seems to disappear when the empty line is commented out. Digressing to plot (plot3d=0) yields an expected result:

I am probably missing something obvious, although the documentation seems to state that (perhaps it's somehow overridden by splot):
Single blank records designate discontinuities in a
plot; no line will join points separated by a blank records (if they are plotted with a line style).