This is because hist use a Patch object to draw the stepped histogram, not a Line2D object.

I would suggest not using hist and instead use np.histogram + ax.step

vals, bins = np.histogram(..)
ax.step(bins[:-1], vals, where='post')

which should be aware of that rcparam.

Ah ok. So it does respect patch.linewidth.

I would like to suggest switching to using ax.step internally to plot such histograms, so that lines.linewidth gets respected (I think it makes sense e.g. for seaborn to set a relatively small patch.linewidth but a large lines.linewidth).

@anntzer No, because then the return typo of hist would depend on a kwarg which is really bad form.

Fair enough, reopened on seaborn's side as mwaskom/seaborn#704.

Reporting an idea that had been shot down on seaborn's side:
What about having hist locally patch rcParams to set patch.linewidth to lines.linewidth when making the plot? I still think the current behavior is unintuitive.

👎 on that from me as well.

If something like that were to happen, the right place would be for seaborn to wrap hist and then use the style context manager to locally over-ride the rcparams.