[ENH]: Configureable default padding around figure when using constrained_layout

[ecciv]

Problem

Constrained layout figures fill out their frames too well, and adding padding during savefig makes the figures no longer adhere to the figsize given in my stylesheet, and isn't a by-default configured amount of padding. When these figures will be used (sometimes two half page-width figures side by side) in my latex document, the contents sit way too close. I don't want to change spacing within the subplot group -- that's all fine, I just want constrained layout to add space around the figure (incl. labels and maybe titles) while respecting the outer figure dimensions.

Proposed solution

Add padding arguments to rcParams so that they may be configured on a stylesheet. These padding arguments should not influence the spacing of elements within the figure -- only around the figure.

[rcomer]

If I have understood, this is already possible with rcParams["figure.constrained_layout.h_pad"] and rcParams["figure.constrained_layout.w_pad"].

See https://matplotlib.org/stable/api/layout_engine_api.html#matplotlib.layout_engine.ConstrainedLayoutEngine

[jklymak]

@rcomer h/w_space puts space between axes, but not on the outer sides. h/w_pad puts a padding around the axes, including the outer sides. but if you set a large h_pad, it will put a big space between the axes as well as on the outer sides.

@ecciv, the rect parameter gives you want you want with a touch of trivial math:

w, h = 7, 5
margin = 1

fig, axs = plt.subplots(1, 3, layout='constrained', figsize=(w, h))
fig.get_layout_engine().set(rect=[margin / w, margin / h,
                                  (w-2*margin) / w, (h-2*margin) / h])
plt.show()

It would be pretty easy to wrap into your own helper, so I don't necessarily think we should add extra API for this, particularly as this is largely to service layout outside of Matplotlib (there are pretty easy ways to horizontally separate figures in LaTeX, for instance).

def my_subplots(nrows, ncols, figsize=None, margin=0.5):
    fig, axs = plt.subplots(nrows, ncols, layout='constrained', figsize=figsize)
    w = fig.get_figwidth()
    h = fig.get_figheight()
    fig.get_layout_engine().set(rect=[margin / w, margin / h,
                                      (w-2*margin) / w, (h-2*margin) / h])
    return fig, axs

[ecciv]

Thanks @jklymak, that's a great way of doing it and it achieves the result I'm after, so I might use that :)

I'll try to explain a bit where I'm coming from with this issue, maybe that will change your mind.

I'm trying to standardize the appearance of plots from different tools in separate repos in our business, and a stylesheet gets us a long way. Our stylesheet is it's own importable module, which makes style compliance and also updating our common business style a breeze. The way I view it is that any added code (past creating the figure and plotting data), like manually touching the grid or what it may be, marks a deviation from our style. A minimal amount of configuration should be as adherent to our style as possible. That's why I don't find a helper function to be the ideal solution, even if in our case we could deploy it quite easily to all our repos via our stylesheet module.

Our stylesheet uses the constrained layout by default. These plots are meant to end up in different places -- probably in Word based reports, in automatically generated Latex design documentation, and in PPoints. The current layout (with basically no margins) is something we can work with in all of those places. But I'd much rather have one solution in our stylesheet than one workaround in each of the places where we want to use these figures (or worse, one workaround in each of the places where we generate a figure).

[jklymak]

Easy would be to add an rcParam for the default rect. However rect is relative to the figure size, so it won't give you a 0.5 inch margin (for instance), but rather one in the fraction of the figure size. Perhaps that is adequate for your needs, and I'm sure we'd welcome a PR.

If you want to add a physical-space margin on the outside (eg in inches), that is not available in the way constrained layout is made currently. It could be made to work by various methods, but I think it would be substantial work. My immediate idea would be to vary "rect" dynamically as the figure changes size, but not 100% sure that would work. One could also add another layer of margins that simply default to zero width - thats not conceptually too hard, but a bit of plumbing would be necessary.

I would still say that this is something I would set up in the final document's stylesheet, not in Matplotlib. But if someone wanted to work on it, I don't think there is any reason not to add an external margin.