I assume, if the fist y is at 50, there would be no vertical line y=0 to y=50? E.g. something like (not tested with your code)

Is that intentional? I could imagine many people would want that (optionally or by default). Otherwise this is not similar to a histogram.

Closes #5855.

Yes, that's intentional based on the consensus in @6669. I would be ok with having both options, as we might want the above implementation to signify that we have no info about the rest of the distribution, but other might want the same results as with ax.hist(data, bins=x, histtype='step') just with preprocessed data.

Naively I would expect "edges" to actually refer to the second option with closed lines, so it would make more sense to call the above "bins" or just descriptively "between" (which is even a naturally sounding answer to where).

As a reminder, there's also #5855 where I suggested allowing both len(x) == len(y)-1 (the case here) and len(y) == len(x)+1 (with "edges", or altenatively this can be seen as a histogram flipped by 90° -- indeed, my intent was to make it easy to switch between one orientation and the other).

Tbh not sure how to implement that, but also I don't really like the idea of implicitly flipping x and y.

It does not have to be implicit, but let's say I would prefer it not to be much harder to do a horizontal histogram rather than a vertical one.

On that note, there seems to be some inconsistency on how np.inf is handled.

fig, ax = plt.subplots()
ax.step(np.r_[x, x[-1]],[-np.inf,*y, -np.inf], where='pre');
ax.set_ylim(1,50)
ax.set_yscale('log')

and

fig, ax = plt.subplots()
ax.step([*x, x[-1]],[-np.inf,*y, -np.inf], where='pre');
ax.set_ylim(1,50);

Either show or don't show the last edge.

The issues with having the closing edges is that the only way to make it work seems if one could specify transform point by point. Right now it looks like it's possible to have x in data coords and y in axes coords, but I don't see how to do it for an line with coords [data, data] [data, axes]. If there's no better solution I'll make it default to a small nonzero number.

you can (should be able to... no guarantees) be able to make it zero and the log-scale handler will clip it itself to a small nonzero number.
but note that I am not suggesting to add closing edges yourself, instead they should come from the user (as the first/last y-values).

@anntzer I added a flipxy bool to do what you wanted in #5855, could you take a look if that implementation works for you?

I guess a variant would be to have orientation="horizontal"/"vertical" which is consistent with the rest of matplotlib (whereas flipxy exists nowhere else). Probably needs some discussion...

I had that at first, but apart from being a handful to write I got confused which way was ''vertical" and which way was "horizontal". I don't have a strong feeling either way. Would be great to get a consensus, so I can start putting in tests.

The orientation would be the same as for hist(). I agree it's not clear this API would be best, just suggesting it.

@dstansby I've added tests, could you review and tell me what's missing?

In general I like the idea, but I won't have the time to properly review this in the coming days at least. I'm not claiming exclusive rights to this PR either :)

Although a quick thought is that after all, it may not be such a great idea to have an 'orientation' kwarg which only has an effect when 'where' has a certain value ("edges"), so perhaps instead having where="edges-horizontal"/"edges-vertical" (or some better name, but basically folding this into 'where') may be better.
Not really sure here, just some thoughts.

Well it works for both new symmetric where options. I didn't modify the rest of them because its not clear to me what the behavior of pre and post should look like.

Tbh I'd be curious to know what fraction of step calls is not plotting histograms.

At least when I first proposed the idea in #5855 the goal was not histograms (see #5855 (comment) and #5855 (comment)).

Apart from the obvious I fail to see what's different about this comment and as implemented here where='between' (or that but horizontal). Feel free to put up an image of what you mean and I can try to work it in.

As for this comment we're planning to make something at least for mpl-hep of the signature histplot(values, bin_edges, histtype=<outline/filled/errors>) which will under the hood call step()/fill_between()/errorbar() or their combinations. If interested we'd be happy to work it into matplotlib itself. But this PR already should simplify our life quite a bit.

I left my problem with this implementation as a comment to #5855.

See #15065