Stroked path width #17198

brunobeltran · 2020-04-20T11:53:07Z

PR Summary

Introduces code to path.py and bezier.py that allows us to compute the extents of stroked Paths analytically. Required for bugfix PRs #17182 and #17199, which fixes #16606.

Previous Behavior

The following code should nominally get the extents of the path, but the current get_extents only takes into account the control points:

markeredgewidth = 10
leg_length = 1
joinstyles = ['miter', 'round', 'bevel']
capstyles = ['butt', 'round', 'projecting']
# The common miterlimit defaults are 0, :math:`\sqrt{2}`, 4, and 10. These
# angles are chosen so that each successive one will trigger one of these
# default miter limits.
angles = [np.pi, np.pi/4, np.pi/8, np.pi/24]
# Each column tests one join style and one butt style, each row one angle
# and iterate through orientations
fig, axs = plt.subplots(len(joinstyles), len(angles), sharex=True,
                        sharey=True)
# technically it *can* extend beyond this depending on miterlimit....
axs[0, 0].set_xlim([-1.5*leg_length, 1.5*leg_length])
axs[0, 0].set_ylim([-1.5*leg_length, 1.5*leg_length])
for i, (joinstyle, capstyle) in enumerate(zip(joinstyles, capstyles)):
    for j, corner_angle in enumerate(angles):
        rot_angle = (i*len(angles) + j) * 2*np.pi/12
        # A path with two caps and one corner. the corner has:
        # path.VertexInfo(apex=(0,0), np.pi + rot_angle + corner_angle/2,
        #                 corner_angle)
        vertices = leg_length*np.array(
            [[1, 0], [0, 0], [np.cos(corner_angle), np.sin(corner_angle)]])
        path = Path(vertices, [Path.MOVETO, Path.LINETO, Path.LINETO])
        path = path.transformed(Affine2D().rotate(rot_angle))
        patch = PathPatch(path, linewidth=markeredgewidth,
                            joinstyle=joinstyle, capstyle=capstyle)
        axs[i, j].add_patch(patch)
        # plot the extents
        data_to_pts = (Affine2D().scale(72)
                        + fig.dpi_scale_trans.inverted()
                        + axs[i, j].transData)
        bbox = path.get_extents()
        axs[i, j].plot([bbox.x0, bbox.x0, bbox.x1, bbox.x1, bbox.x0],
                        [bbox.y0, bbox.y1, bbox.y1, bbox.y0, bbox.y0],
                        'r-.')
        axs[i, j].axis('off')

New Behavior

Now we can use the Path.get_stroked_extents method:

markeredgewidth = 10
leg_length = 1
joinstyles = ['miter', 'round', 'bevel']
capstyles = ['butt', 'round', 'projecting']
# The common miterlimit defaults are 0, :math:`\sqrt{2}`, 4, and 10. These
# angles are chosen so that each successive one will trigger one of these
# default miter limits.
angles = [np.pi, np.pi/4, np.pi/8, np.pi/24]
# Each column tests one join style and one butt style, each row one angle
# and iterate through orientations
fig, axs = plt.subplots(len(joinstyles), len(angles), sharex=True,
                        sharey=True)
# technically it *can* extend beyond this depending on miterlimit....
axs[0, 0].set_xlim([-1.5*leg_length, 1.5*leg_length])
axs[0, 0].set_ylim([-1.5*leg_length, 1.5*leg_length])
for i, (joinstyle, capstyle) in enumerate(zip(joinstyles, capstyles)):
    for j, corner_angle in enumerate(angles):
        rot_angle = (i*len(angles) + j) * 2*np.pi/12
        # A path with two caps and one corner. the corner has:
        # path.VertexInfo(apex=(0,0), np.pi + rot_angle + corner_angle/2,
        #                 corner_angle)
        vertices = leg_length*np.array(
            [[1, 0], [0, 0], [np.cos(corner_angle), np.sin(corner_angle)]])
        path = Path(vertices, [Path.MOVETO, Path.LINETO, Path.LINETO])
        path = path.transformed(Affine2D().rotate(rot_angle))
        patch = PathPatch(path, linewidth=markeredgewidth,
                            joinstyle=joinstyle, capstyle=capstyle)
        axs[i, j].add_patch(patch)
        # plot the extents
        data_to_pts = (Affine2D().scale(72)
                        + fig.dpi_scale_trans.inverted()
                        + axs[i, j].transData)
        bbox = path.get_stroked_extents(markeredgewidth, data_to_pts,
                                        joinstyle, capstyle)
        bbox = bbox.transformed(data_to_pts.inverted())
        axs[i, j].plot([bbox.x0, bbox.x0, bbox.x1, bbox.x1, bbox.x0],
                        [bbox.y0, bbox.y1, bbox.y1, bbox.y0, bbox.y0],
                        'r-.')
        axs[i, j].axis('off')

Saving as a PDF or PNG produces the following:

Remaining bugs

If we instead output an SVG, it's wrong (because the default miterlimit is different). To fix this, we would need to deal with #9830, because matplotlib doesn't set the miterlimit right now. This feels acceptably out of scope for this PR. This PR contains all the code required to deal with an arbitrary miterlimit, but it has been hardcoded to 10, since that's the nominal miterlimit for Agg, which is in charge of rasterizing, allowing us to fix #17182.

PR Checklist

Has Pytest style unit tests
Code is Flake 8 compliant
New features are documented, with examples if plot related
Documentation is sphinx and numpydoc compliant
Added an entry to doc/users/next_whats_new/ if major new feature (follow instructions in README.rst there)
Documented in doc/api/api_changes.rst if API changed in a backward-incompatible way

jkseppan · 2020-08-14T06:17:33Z

Looks good to me. I wonder what's going to with the AppVeyor and Azure checks.

jkseppan

Approving modulo CI passing

jkseppan · 2020-08-14T06:29:30Z

The failure diff on appveyor:

Is this an Agg version difference?

jkseppan · 2020-08-14T06:30:04Z

The test result:

And the expected image:

jkseppan · 2020-08-14T06:33:06Z

Once this is merged, we could use this in writeMarkers in backend_pdf.py.

QuLogic

I only really read docs so far.

QuLogic · 2020-08-25T20:51:56Z

lib/matplotlib/path.py

+    form a corner, the angle swept out by the two lines that meet at the
+    vertex.


I'm confused what this means with relation to the next member, especially since that one has corner in the name.

Good catch. This was bad copy/paste from an older version where the names were too difficult to remember (phi/theta, as below). I fixed the description here to actually make sense.

lib/matplotlib/path.py

QuLogic · 2020-08-25T21:21:30Z

lib/matplotlib/path.py

+    ----------
+    extents : 4*[float]
+        The extents (xmin, ymin, xmax, ymax) of the `~.transforms.Bbox` of the
+        vertices. Modified in place so that the corner described by *vinfo*


The type should be more explicit then; it has to a mutable reference of some sort to work (i.e., not tuples.)

Is (4,) array_like of float better?

Maybe, I'm not sure we have a convention for must-be-mutable, @timhoffm?

lib/matplotlib/path.py

brunobeltran · 2020-08-31T18:58:22Z

Thanks @QuLogic as usual for your detailed documentation read, I'm glad you caught these lingering copy/paste errors, as they made the code hard to go back to even for me. Hopefully these documentation fixes help the code make lots more sense now.

Let me know if you have any issues parsing what something is supposed to be doing.

I assume you'll ask this question when you actually read the code itself:

"why did he get the angle information in global coordinates, then include the logic to pretend all corners point in the positive x direction right at the end?"

The formulas are much simpler if you fix the corners to be aligned in on specific direction, it turns out. I tried both ways (the other way being to include the switching logic for which direction the corner is pointing directly in the code that's currently in "stroke_x_overflow") and I found this one much more readable.

brunobeltran · 2020-09-01T03:52:27Z

The original idea was that ``4*[float]`` was at least not ``4*(float, )``...

…

On Mon, Aug 31, 2020 at 4:42 PM Elliott Sales de Andrade < ***@***.***> wrote: ***@***.**** commented on this pull request. ------------------------------ In lib/matplotlib/path.py <#17198 (comment)> : > + elif joinstyle == 'round': + return width/2 # hemispherical cap, so always same padding + else: + raise ValueError(f"Unknown joinstyle: {joinstyle}") + + +def _pad_extents_stroked_vertex(extents, vinfo, markeredgewidth, joinstyle, + capstyle): + """ + Accumulator for building true extents from `.VertexInfo`s. + + Parameters + ---------- + extents : 4*[float] + The extents (xmin, ymin, xmax, ymax) of the `~.transforms.Bbox` of the + vertices. Modified in place so that the corner described by *vinfo* Maybe, I'm not sure we have a convention for must-be-mutable, @timhoffm <https://github.com/timhoffm>? — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#17198 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AALIGPUEAIH5ZFF7Q26TRTTSDQYPHANCNFSM4MML7XCA> .

QuLogic

This was pretty-well commented, which made it easier to review. I mean, it still took hours, but probably less than if it weren't.

QuLogic · 2020-09-01T00:01:43Z

lib/matplotlib/path.py

+                    cap_angles = [first_tan_angle, prev_tan_angle]
+                    cap_vertices = [first_vertex, prev_vertex]
+                    for cap_angle, cap_vertex in zip(cap_angles, cap_vertices):
+                        yield VertexInfo(cap_vertex, cap_angle, None)


Just yielding the two things seems more straightforward?

Suggested change

cap_angles = [first_tan_angle, prev_tan_angle]

cap_vertices = [first_vertex, prev_vertex]

for cap_angle, cap_vertex in zip(cap_angles, cap_vertices):

yield VertexInfo(cap_vertex, cap_angle, None)

yield VertexInfo(first_vertex, first_tan_angle, None)

yield VertexInfo(prev_vertex, prev_tan_angle, None)

QuLogic · 2020-09-01T00:16:45Z

lib/matplotlib/path.py

+                # deal with capping ends of previous polyline, if it exists
+                if prev_tan_angle is not None and is_capped:


Capping occurs if the polyline was unclosed, so I'm 50/50 whether is_capped should be is_closed (with opposite values), or this comment should mention that.

QuLogic · 2020-09-01T00:22:32Z

lib/matplotlib/path.py

+            for cap_angle, cap_vertex in [(first_tan_angle, first_vertex),
+                                          (prev_tan_angle, prev_vertex)]:
+                yield VertexInfo(cap_vertex, cap_angle, None)


Same here.

Suggested change

for cap_angle, cap_vertex in [(first_tan_angle, first_vertex),

(prev_tan_angle, prev_vertex)]:

yield VertexInfo(cap_vertex, cap_angle, None)

yield VertexInfo(first_vertex, first_tan_angle, None)

yield VertexInfo(prev_vertex, prev_tan_angle, None)

QuLogic · 2020-09-01T00:25:06Z

lib/matplotlib/path.py

+                # often CLOSEPOLY is used when the curve has already reached
+                # it's initial point in order to prevent there from being a


Suggested change

# often CLOSEPOLY is used when the curve has already reached

# it's initial point in order to prevent there from being a

# Often CLOSEPOLY is used when the curve has already reached

# its initial point in order to prevent there from being a

QuLogic · 2020-09-01T00:34:04Z

lib/matplotlib/path.py

+        ----------
+        markeredgewidth : float
+            Width, in points, of the stroke used to create the marker's edge.
+            For ``markeredgewidth = 0``, same as `.get_extents`.


Suggested change

For ``markeredgewidth = 0``, same as `.get_extents`.

For ``markeredgewidth = 0``, the result will become the same as `.get_extents`.

QuLogic · 2020-09-01T05:32:02Z

lib/matplotlib/path.py

+        phi1 = phi + theta/2
+        phi2 = phi - theta/2
+        return (width/2) * max(np.abs(np.cos(phi1)), np.abs(np.cos(phi2)))
+    # finally, _joinstyle = "round" is just _joinstyle = "bevel" but with


?

Suggested change

# finally, _joinstyle = "round" is just _joinstyle = "bevel" but with

# finally, joinstyle = "round" is just joinstyle = "bevel" but with

QuLogic · 2020-09-01T05:40:10Z

lib/matplotlib/path.py

+            cap_perp = vinfo.incidence_angle + perp_dir
+            x += (markeredgewidth/2) * np.cos(cap_perp)
+            if x < extents[xmin]:
+                extents[xmin] = x
+            if x > extents[xmax]:
+                extents[xmax] = x
+            y += (markeredgewidth/2) * np.sin(cap_perp)
+            if y < extents[ymin]:
+                extents[ymin] = y
+            if y > extents[ymax]:
+                extents[ymax] = y


I'm not sure if it looks better this way (ditto the others), or worse and slower...

cap_perp = vinfo.incidence_angle + perp_dir x += (markeredgewidth/2) * np.cos(cap_perp) extents[xmin] = min(extents[xmin], x) extents[xmax] = max(extents[xmax], x) y += (markeredgewidth/2) * np.sin(cap_perp) extents[ymin] = min(extents[ymin], y) extents[ymax] = max(extents[ymax], y)

QuLogic · 2020-09-01T05:44:38Z

lib/matplotlib/path.py

+    # of the bbox, and see if the stroked vertex expands the extents...
+    x, y = vinfo.apex
+    if np.cos(vinfo.incidence_angle) > 0:
+        incidence_angle = vinfo.incidence_angle + np.pi/2


In these cases, I'd name the local variable theta, as the reference for the angle is different (I think?), so it's different than incidence_angle.

QuLogic · 2020-09-01T05:50:44Z

lib/matplotlib/path.py

+        For vertices with one incoming line, *phi* is the incidence angle that
+        line forms with the positive y-axis.  For corners (vertices with two


Clockwise, or counter-clockwise?

QuLogic · 2020-09-01T05:53:19Z

lib/matplotlib/tests/test_path.py

+@image_comparison(['stroked_bbox'], remove_text=True,
+                  extensions=['pdf', 'svg', 'png'])


These are the default extensions?

Suggested change

@image_comparison(['stroked_bbox'], remove_text=True,

extensions=['pdf', 'svg', 'png'])

@image_comparison(['stroked_bbox'], remove_text=True)

github-actions · 2023-07-26T01:59:14Z

Since this Pull Request has not been updated in 60 days, it has been marked "inactive." This does not mean that it will be closed, though it may be moved to a "Draft" state. This helps maintainers prioritize their reviewing efforts. You can pick the PR back up anytime - please ping us if you need a review or guidance to move the PR forward! If you do not plan on continuing the work, please let us know so that we can either find someone to take the PR over, or close it.

brunobeltran force-pushed the stroked_path_width branch 2 times, most recently from 5a401ee to 7d84966 Compare April 20, 2020 11:58

This was referenced Apr 20, 2020

BUGFIX: use true bbox for rasters in backend_mixed #17182

Draft

Line2d window extents #17199

Draft

brunobeltran force-pushed the stroked_path_width branch from 7d84966 to 0a92c2f Compare April 20, 2020 15:05

brunobeltran added New feature topic: path handling labels Apr 22, 2020

brunobeltran force-pushed the stroked_path_width branch from 0a92c2f to 265caec Compare June 22, 2020 19:09

jkseppan approved these changes Aug 14, 2020

View reviewed changes

jkseppan requested a review from QuLogic August 14, 2020 06:19

QuLogic reviewed Aug 25, 2020

View reviewed changes

FEATURE: Path.get_stroked_extents

a055b24

brunobeltran force-pushed the stroked_path_width branch from 265caec to 48fe47a Compare August 31, 2020 18:54

brunobeltran force-pushed the stroked_path_width branch from 48fe47a to efceb57 Compare August 31, 2020 19:21

DOCS: @QuLogic's suggestions for stroked extents docs

dad23ba

brunobeltran force-pushed the stroked_path_width branch from efceb57 to dad23ba Compare August 31, 2020 20:10

QuLogic reviewed Sep 1, 2020

View reviewed changes

QuLogic added the status: needs revision label Oct 9, 2020

jklymak added the status: needs rebase label Apr 23, 2021

jklymak marked this pull request as draft April 23, 2021 15:31

github-actions bot added the status: inactive Marked by the “Stale” Github Action label Jul 26, 2023

anntzer mentioned this pull request Aug 11, 2023

get_tightbbox doesn't account for markeredgewidth #16606

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Stroked path width #17198

Stroked path width #17198

brunobeltran commented Apr 20, 2020 •

edited

Loading

jkseppan commented Aug 14, 2020

jkseppan left a comment

jkseppan commented Aug 14, 2020

jkseppan commented Aug 14, 2020

jkseppan commented Aug 14, 2020

QuLogic left a comment

QuLogic Aug 25, 2020

brunobeltran Aug 31, 2020

QuLogic Aug 25, 2020

brunobeltran Aug 31, 2020

QuLogic Aug 31, 2020

brunobeltran commented Aug 31, 2020

brunobeltran commented Sep 1, 2020 via email

QuLogic left a comment

QuLogic Sep 1, 2020

QuLogic Sep 1, 2020

QuLogic Sep 1, 2020

QuLogic Sep 1, 2020

QuLogic Sep 1, 2020

QuLogic Sep 1, 2020

QuLogic Sep 1, 2020

QuLogic Sep 1, 2020

QuLogic Sep 1, 2020

QuLogic Sep 1, 2020

github-actions bot commented Jul 26, 2023

		form a corner, the angle swept out by the two lines that meet at the
		vertex.

		# deal with capping ends of previous polyline, if it exists
		if prev_tan_angle is not None and is_capped:

		# often CLOSEPOLY is used when the curve has already reached
		# it's initial point in order to prevent there from being a

	For ``markeredgewidth = 0``, same as `.get_extents`.
	For ``markeredgewidth = 0``, the result will become the same as `.get_extents`.

	# finally, _joinstyle = "round" is just _joinstyle = "bevel" but with
	# finally, joinstyle = "round" is just joinstyle = "bevel" but with

		For vertices with one incoming line, phi is the incidence angle that
		line forms with the positive y-axis. For corners (vertices with two

		@image_comparison(['stroked_bbox'], remove_text=True,
		extensions=['pdf', 'svg', 'png'])

Stroked path width #17198

Are you sure you want to change the base?

Stroked path width #17198

Conversation

brunobeltran commented Apr 20, 2020 • edited Loading

PR Summary

PR Checklist

jkseppan commented Aug 14, 2020

jkseppan left a comment

Choose a reason for hiding this comment

jkseppan commented Aug 14, 2020

jkseppan commented Aug 14, 2020

jkseppan commented Aug 14, 2020

QuLogic left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

brunobeltran commented Aug 31, 2020

brunobeltran commented Sep 1, 2020 via email

QuLogic left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

github-actions bot commented Jul 26, 2023

brunobeltran commented Apr 20, 2020 •

edited

Loading