gnuplot Git Repository

OK to have missing corners in a image represented by a sparse matrix

When a sparse matrix is populated, missing entries are flagged UNDEFINED.
However we do know their indices, and this is sufficient to generate
coordinates for plotting. The image rendering code does a quick check
to see if the first pixel of an image is UNDEFINED and fails if so.
We don't want this to be the case for a sparse matrix with a missing corner,
so make sure the corners are flagged INRANGE even if the content is NaN
because the pixel is missing.

Bug 2791

more fixups for placement of minor tics when logscale axis uses linear tic spacing

fixup for commit 3e103a4c3
The same adjustment made to handle minor tics when the program forces
major tics on a logscale axis to use linear spacing must also be made
when the user explicitly requests "set tics nolog".

Bug 2765

"set boxwidth" is now applied to "plot ... bins ... with hsteps"

Commit 22feb809b2 introduced changes allowing the "hsteps" style to use
the "set boxwidth" setting, but it did not fully address cases where the
`bins` filter is applied.

This commit modifies the behavior so that even when the `bins` filter is used,
the "hsteps" style can still respect the width setting specified by "set boxwidth".

When the `bins` filter is applied, the number of members in each bin is stored
in "points[i].z". However, "plot_hsteps()" expects "points[i].z" to represent
the width of each segment. Since the `bins` filter provides invalid data in
"points[i].z" as the width, this caused incorrect handling of "set boxwidth".

This commit ensures that when the `bins` filter is applied, "plot_hsteps()"
no longer uses "points[i].z" for determining the segment width.

when logscale tic placement is adjusted, mtics placement must change also

Commits d4b92d1e9 and 4543908a9 modifed placement of logscale axis
ticmarks by introducing a sanity check for how many marks would be
placed. If fewer than 3 marks would be placed using the usual code,
marks are instead placed at equal linear increments.

However minor tic generation was not modified appropriately to match this.
Now it is.

Test case:
set xrange [1:14]
set logscale x
set xtics 10 logscale
set mxtics 10
plot x

Bug #2765

Fix broken cases of "set colorbox invert"

1) The "invert" flag was honored only for a vertical colorbox,
ignored for a horizontal colorbox.

2) The draw_inside_colorbox_bitmap_smooth__image() special case
routine used by the qt and cairo terminals (commit 6af1bda4cc)
does not work for an inverted colorbox.

boxplot: using (start):2:3:4 every :::n::n should work

The "every" clause meant that the starting x position for boxplot category
labels was stored in plot->points[n].x rather than plot->points[0].x,
which confused the tic label generation code.

Bug #2789

boxplot: using 1:2:3:xticlabel("foo") should work

Fixes an unnoticed regression from long ago when commit 30668d20
added a four column variant of "with boxplot".

docs: error in example plot command for "with boxplot"

Guarantee that comma operator is a sequence point in serial evaluation

This commit adds an operation SERIAL_COMMA to the evaluation stack
corresponding to the position of a comma in an expression such as
A = ( <exp1> , <exp2> )
The intent is that any error flags from evaluation of <exp1> should not
persist across evaluation of <exp2>. In particular, if <exp2> is
well-defined but <exp1> is not then A should receive the value of <exp2>
rather than being set to <undefined>.

gnuplot> show at A = (B,C)

pushc "A"
push B
pop
serial comma
push C
assign

This change allows serial evaluation in a "using" clause to reference a missing
or NaN value as part of <exp1> without affecting the value return for <exp2>.
plot FOO using 1 : (TEMP=$3, $2)
The point at [ $1, $2 ] should be plotted even if column 3 is missing, NaN,
or undefined. Further use of the value in TEMP later in the plot command
may be problematic, however.

Bug #2787

add ability for user to set XML <desc>