Fix errorbar property cycling to match plot. #17930

QuLogic · 2020-07-15T06:41:59Z

PR Summary

It would not cycle if a color were specified. However, this does not match plot, which does not advance the cycle only if all properties in the cycle are specified. Notably, this means if your property cycle was for line style, specifying a color would ignore the cycle in errorbar, but not in plot.

Fixes #7074.

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/next_api_changes/* if API changed in a backward-incompatible way

WeatherGod · 2020-07-15T14:25:05Z

lib/matplotlib/axes/_axes.py

-            base_style = next(self._get_lines.prop_cycler)
-
+        base_style = self._get_lines._getdefaults(
+            set(), {**fmt_style_kwargs, **kwargs})


Is this where the property cycler gets implicitly advanced?

Yes, I don't really like the private API of a private member, but that's how it is right now.

jklymak · 2020-07-15T14:59:36Z

Is it the case that errorbar was wrong or that the behaviour for plot changed, but it didn't get changed in errorbar? I suspect the latter, in which case this may still get out of sync. Does it make sense to have some sort of helper for this or add the functionality into _get_lines._getdefaults so this doesn't get out of sync in the future?

QuLogic · 2020-07-17T05:11:32Z

I think _get_lines._getdefaults is that helper.

My preferred method would be to use something higher, like _get_lines() directly or _get_lines._makeline(), but we need the property dict to copy to the error bars. I don't think we have an easy way to copy from the Line2D to the error bar line. Though maybe since this is private, I can just make it return both the line and its properties?

jklymak · 2020-07-17T14:26:38Z

I'm not blocking on that comment - just seems if we want the two to have the same behaviour they should have the same method of getting the behaviour...

QuLogic · 2020-07-18T03:48:36Z

Well, it is, sort of. self._get_lines is the object that implements the behaviour. The plot call tree is plot -> self._get_lines() -> self._get_lines._plot_args -> self._get_lines._makeline -> (self._get_lines._getdefaults for kwargs and creates Line2D). Previously, errorbar used the self._get_lines.prop_cycle and only looked for colour; in this PR it uses _getdefaults like plot ultimately does.

I experimented a little bit with calling one of those higher methods and then .update_from(data_line) + overriding any properties after, but this is complicated by the kwarg inheritance in errorbar. It's currently:

_getdefaults + fmt_style_kwargs -> base_style
- + kwargs = plot_line_style -> the main line
- + e* kwargs - markers/lines = eb_lines_style -> error bars
- + cap settings - markers/lines = eb_cap_style -> error caps

However, the main line is Line2D while the error bars are LineCollection, so they have different properties and bars.update_from(data_line) doesn't work.

I think at best, we could have _makeline return both the line and its kwargs, which could then be used as base_style, or maybe do that on _plot_args.

QuLogic · 2020-07-18T07:14:36Z

The closest I could do is call self._get_lines._plot_args, as calling self.plot or self._get_lines directly would call _process_unit_info again:

matplotlib/lib/matplotlib/axes/_base.py

Line 210 in bb9d11d

self.axes._process_unit_info(kwargs=kwargs)

or process data

matplotlib/lib/matplotlib/axes/_base.py

Lines 220 to 263 in bb9d11d

    
           if data is None:  # Process dict views 
        
               args = [cbook.sanitize_sequence(a) for a in args] 
        
           else:  # Process the 'data' kwarg. 
        
               replaced = [mpl._replacer(data, arg) for arg in args] 
        
               if len(args) == 1: 
        
                   label_namer_idx = 0 
        
               elif len(args) == 2:  # Can be x, y or y, c. 
        
                   # Figure out what the second argument is. 
        
                   # 1) If the second argument cannot be a format shorthand, the 
        
                   #    second argument is the label_namer. 
        
                   # 2) Otherwise (it could have been a format shorthand), 
        
                   #    a) if we did perform a substitution, emit a warning, and 
        
                   #       use it as label_namer. 
        
                   #    b) otherwise, it is indeed a format shorthand; use the 
        
                   #       first argument as label_namer. 
        
                   try: 
        
                       _process_plot_format(args[1]) 
        
                   except ValueError:  # case 1) 
        
                       label_namer_idx = 1 
        
                   else: 
        
                       if replaced[1] is not args[1]:  # case 2a) 
        
                           cbook._warn_external( 
        
                               f"Second argument {args[1]!r} is ambiguous: could " 
        
                               f"be a format string but is in 'data'; using as " 
        
                               f"data.  If it was intended as data, set the " 
        
                               f"format string to an empty string to suppress " 
        
                               f"this warning.  If it was intended as a format " 
        
                               f"string, explicitly pass the x-values as well.  " 
        
                               f"Alternatively, rename the entry in 'data'.", 
        
                               RuntimeWarning) 
        
                           label_namer_idx = 1 
        
                       else:  # case 2b) 
        
                           label_namer_idx = 0 
        
               elif len(args) == 3: 
        
                   label_namer_idx = 1 
        
               else: 
        
                   raise ValueError( 
        
                       "Using arbitrary long args with data is not supported due " 
        
                       "to ambiguity of arguments; use multiple plotting calls " 
        
                       "instead") 
        
               if kwargs.get("label") is None: 
        
                   kwargs["label"] = mpl._label_from_arg( 
        
                       replaced[label_namer_idx], args[label_namer_idx]) 
        
               args = replaced

which are not done in errorbar.

jklymak

This looks good to me - just a small comment...

lib/matplotlib/axes/_axes.py

QuLogic · 2020-08-14T20:29:14Z

With #8031 in, I'd like to apply this to the 3D version as well before merging.

timhoffm · 2020-08-18T23:48:13Z

This seems rather complete (modulo some restructuring of the clipping). IMHO it would be easier for reviewing to merge this now and do the 3D stuff in a separate PR.

QuLogic · 2020-08-19T00:09:20Z

Sure, I ran into other issues just doing that directly, so I'll mark this Ready for Review now, and do 3D in a followup PR. Not sure what you mean about clipping though?

timhoffm · 2020-08-19T00:28:37Z

Sorry, I meant popping, not clipping.

It would not cycle if a color were specified. However, this does not match `plot`, which does not advance the cycle only if _all_ properties in the cycle are specified. Notably, this means if your property cycle was for line style, specifying a color would ignore the cycle in `errorbar`, but not in `plot`. Fixes matplotlib#7074.

Move iterable checks before working on styling. In the styling section, the lines and caps dictionary modifications are intermixed, making them a bit confusing. Also do the same marker style removal from the cap style, which would normally be replaced later.

We can't directly use `self.plot` or `self._get_lines` because they would do `self._process_unit_info` and/or *data* keyword argument processing.

QuLogic · 2020-08-19T04:23:01Z

Ah, okay, the last commit should clear that up, I hope.

As with 2D before matplotlib#17930, it would not cycle if a color were specified. However, this does not match `plot`, which does not advance the cycle only if _all_ properties in the cycle are specified. Notably, this means if your property cycle was for line style, specifying a color would ignore the cycle in `errorbar`, but not in `plot`. This is a 3D version of 149e7fb and 0782c74.

QuLogic added the API: consistency label Jul 15, 2020

QuLogic added this to the v3.4.0 milestone Jul 15, 2020

QuLogic force-pushed the errorbar-cycler branch from d810348 to 1c59fb0 Compare July 15, 2020 06:42

WeatherGod reviewed Jul 15, 2020

View reviewed changes

QuLogic force-pushed the errorbar-cycler branch from 8eea274 to c548924 Compare July 18, 2020 07:11

QuLogic mentioned this pull request Jul 24, 2020

Add errorbars to mplot3d #8031

Merged

QuLogic force-pushed the errorbar-cycler branch from c548924 to a61a034 Compare July 24, 2020 06:39

jklymak approved these changes Aug 14, 2020

View reviewed changes

lib/matplotlib/axes/_axes.py Outdated Show resolved Hide resolved

QuLogic marked this pull request as draft August 14, 2020 20:28

QuLogic mentioned this pull request Aug 18, 2020

Unit handling needs many more tests #18278

Open

QuLogic force-pushed the errorbar-cycler branch 2 times, most recently from a7c085d to 32c3ece Compare August 19, 2020 00:06

QuLogic marked this pull request as ready for review August 19, 2020 00:08

QuLogic added 5 commits August 18, 2020 23:44

Return used kwargs from _process_plot_var_args._plot_args.

defa63c

Make errorbar's main Line2D closer to plot.

0782c74

We can't directly use `self.plot` or `self._get_lines` because they would do `self._process_unit_info` and/or *data* keyword argument processing.

Simplify creating style dict for errorbar bars/caps.

da8d510

QuLogic force-pushed the errorbar-cycler branch from 32c3ece to da8d510 Compare August 19, 2020 04:10

timhoffm approved these changes Aug 19, 2020

View reviewed changes

timhoffm merged commit c7e4448 into matplotlib:master Aug 19, 2020

QuLogic deleted the errorbar-cycler branch August 19, 2020 18:46

QuLogic mentioned this pull request Sep 8, 2020

Sync 3D errorbar with 2D #18436

Merged

4 tasks

This was referenced Sep 24, 2020

Accept same types to errorevery as markevery #18556

Merged

plt.errorbar raises error when given marker=<matplotlib.markers.MarkerStyle object> #18600

Closed

smartlixx mentioned this pull request Mar 30, 2021

FIX errorbar problem with fillstyle #19784

Merged

3 tasks

timhoffm mentioned this pull request Apr 6, 2021

Using "drawstyle" raises AttributeError in errorbar, when yerr is specified. #19879

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Uh oh!

Fix errorbar property cycling to match plot. #17930

Fix errorbar property cycling to match plot. #17930

Uh oh!

QuLogic commented Jul 15, 2020 •

edited

Loading

Uh oh!

WeatherGod Jul 15, 2020

Uh oh!

QuLogic Jul 17, 2020

Uh oh!

jklymak commented Jul 15, 2020

Uh oh!

QuLogic commented Jul 17, 2020

Uh oh!

jklymak commented Jul 17, 2020

Uh oh!

QuLogic commented Jul 18, 2020 •

edited

Loading

Uh oh!

QuLogic commented Jul 18, 2020

Uh oh!

jklymak left a comment

Uh oh!

Uh oh!

QuLogic commented Aug 14, 2020

Uh oh!

timhoffm commented Aug 18, 2020

Uh oh!

QuLogic commented Aug 19, 2020 •

edited

Loading

Uh oh!

timhoffm commented Aug 19, 2020

Uh oh!

QuLogic commented Aug 19, 2020

Uh oh!

Uh oh!

Uh oh!

Fix errorbar property cycling to match plot. #17930

Fix errorbar property cycling to match plot. #17930

Uh oh!

Conversation

QuLogic commented Jul 15, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

PR Summary

PR Checklist

Uh oh!

WeatherGod Jul 15, 2020

Choose a reason for hiding this comment

Uh oh!

QuLogic Jul 17, 2020

Choose a reason for hiding this comment

Uh oh!

jklymak commented Jul 15, 2020

Uh oh!

QuLogic commented Jul 17, 2020

Uh oh!

jklymak commented Jul 17, 2020

Uh oh!

QuLogic commented Jul 18, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

QuLogic commented Jul 18, 2020

Uh oh!

jklymak left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

QuLogic commented Aug 14, 2020

Uh oh!

timhoffm commented Aug 18, 2020

Uh oh!

QuLogic commented Aug 19, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

timhoffm commented Aug 19, 2020

Uh oh!

QuLogic commented Aug 19, 2020

Uh oh!

Uh oh!

QuLogic commented Jul 15, 2020 •

edited

Loading

QuLogic commented Jul 18, 2020 •

edited

Loading

QuLogic commented Aug 19, 2020 •

edited

Loading