Use numpy.linspace to generate grids built with reference point and spacings

[gcaria]

Close #3559

A very simple fix taken straight from the issue's only comment.

[rcomer]

Hi @gcaria, thanks for this. We are currently having multiple problems with our Cirrus CI testing, so the test output will be showing a whole raft of failures that are not linked to your change. To get everything working again, we need to merge #4108, #4104, #4112 and probably something to address #3993. So please bear with us while we sort all that out, then you can rebase and someone can have a proper look at this.

Again, thanks for your contribution!

[rcomer]

Hi @gcaria, we have now got the tests all passing on master, so please rebase at your convenience 🙂

[gcaria]

Thanks for the feedback @rcomer !

It seems that the test failures are then caused by this commit, although it's hard to believe such a simple edit can break anything. Maybe this has to do with some expected reference data that doesn't match the actual one anymore?

I admit I have some difficulties running the test suite on my own, since I get a lot of AssertionError: CML do not match failures (I believe I have followed the instructions on the "contributing" page)

[rcomer]

Thanks for the update @gcaria. Yes, it looks like this is causing very small changes all over the place, and in some cases it does look for the worse. E.g. here the latitude points change from round multiples of 0.5 to ?.?999.

Some of the test input files are in pp-format, and the pp file headers only specify latitude/longitude by their "zeroth" value, step and number of rows/columns. So that probably explains why so many unrelated tests are affected.

I'm not sure what the right thing to do here is. It seems like a computer science question to me, and I am very much not a computer scientist! Perhaps @alastair-gemmell might have some ideas, since he opened the original issue.

[wjbenfold]

I tried a close and reopen to nudge the CI, but I think it's mostly just looking at quite old tests. A rebase would fix that.

When I run the tests locally with this change, I get the CML match failures and a couple of other errors from test_multiple_unordered_lbprocs and test_multiple_unordered_rotated_lbprocs. It think it's worth investigating those errors as something's not quite coming out with the right length.

In terms of whether the CML failures are a bad thing, I think they'd all/mostly pass an isclose check, but CML is looking for an exact match. The changes seem fine to me, though @bjlittle might also have an opinion as he looked at similar changes caused by a numpy update fairly recently.

Before merging this change, I think that we should update the lines just below it that use arange too. We could also consider whether we can choose when to check for regularity with iris.util.points_step based on step size as quite a lot of pp file load time is devoted to that check currently, though that's probably outside of scope of this PR.

[ESadek-MO]

We have discussed this @SciTools/peloton and have decided to close this; please feel free to reopen! Thanks for your work on this!