SunPy Developer Discussion, 5 Jan 2015

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: SunPy Developer Discussion

Description

Topics:
PRs
0.6 sprint

A

Yes, we are live. Okay, yes, happy new year everyone, so we have a lot of PRS.

B

But the phenomena black adorable.

A

B

C

I would suggest that we spend the majority of this meeting discussing the sprint in two days. Yes, well, let's.

A

Start with a sprint, I think some of the PRS we did.

B

You really need to look at something else: cuz, it is scaring quiet. Second, there are 31 open pr's like an orphanage 15, a martyr's review. Most of those are passing Travis. Oh uh till Travis fell on its sword in a very graceful manner. Over Christmas.

B

Which totally may not be course it's annoying to have some people.

A

Discuss that he high school, we all start breaking, and then people have to fix guitars. There's, no reason that we can murder, which a murder yeah.

C

I think I'm a maximum will be able to accept one and then it'll break. All the others should choose that one wisely like all of them, not.

B

Quite a lot of them are pretty small, so.

A

Most of the recent ones.

A

What is a Stewart as on the side? What is the testing label? What do you mean by the testing label mean.

B

That is related to tests, although testing.

A

What are the pr itself is under casting it's that the PRS related to our tests.

B

What ya got you know.

A

Based acclaimed, interesting label.

B

Implies that it is related to the tests and the running their roles can.

C

We just call it tests, then, instead of testing the verb. That actually part of that is confusing.

B

Okay, you say so it is through. It trust us and we all happy now.

A

Yeah, okay, our jobs done for this meeting; okay! Well, that's a fantastic meeting: let's go home! Okay, all right! The sprigs! Let's talk about the sprint! The sprint is scheduled for Wednesday two days from now that correct.

C

D

C

B

We prepared for this way I'm always prepared Stephen I'm bad. I was born four pairs. Are you still junk? It's just been a really bad day.

E

That doesn't really answer my question. One realize I think.

C

That better, that answers it in the positive fashion.

A

Which I was drunk and Andy? You have not responded to whether you are doing the Sprint or 90 oops.

F

uh Yes, I'm doing a sprint all.

C

Right wow, is it re? I would highly suggest that we for the people I guttered. We go to the library for Earth at least. Do it all so that we're all together somewhere yeah.

A

Good idea, yep, let you know what the deal with a librarian is I just go in there. Yeah.

C

Yeah, you know, as for your pay, a prepared, is there anything just make you, sir, but it does only open it like 12.

C

F

Lunchtime some time outside.

E

Of opened it came or something like that. I know.

C

The word suggesting that people go into there because of a fumes, but out did they change the hours to accommodate that I. Don't know that they did.

C

C

So currently we have our the the Sprint is a sprint. Tag must be up to date, so the things that will be addressed in the sprint are just those things are tagged, sprint, 0.6 sprint, or is there something else we want to add or take away.

E

What were the state services status of the if I might that even close to being sprint able.

E

G

Is very very closely hey somewhere somewhere um yeah yeah.

H

C

You guys notice releasing.

I

On what there was some.

C

Discussion on on Astro pi about them doing, ephemerides, they're figuring it out and adding it to a store, buy something again. Anybody follow that conversation. Yes,.

B

C

Do you know what was the end result of that? Are they actually going to do it or there.

B

Isn't anything dude? There is some discussion about a future implementation of the jpo hemorrhoids. Oh.

C

Yeah, that's about it and they.

B

Have added the whatever it's called equatorial frame so that you can, as all one point oh you'll, be able to pull out the or as position of the Sun from a specific location, less.

E

B

And I don't particularly like, if you either use for it, I, don't know where that position got emotionally.

B

Yes, it did and.

C

So, does that mean that we should get rid of the FM rides from.

H

C

H

F

C

No, no, I mean, because you said, you're not ready well,.

H

I'm not ready, I'm just need time, so this pain to be at time for me to look into it and work on it.

C

Is it something you think you can finish in a day.

H

Maybe if I look just that yeah, if I forget about this picture,.

H

The units yeah.

C

H

If anyone volunteered for the spectra, you need stuff and please step.

F

Ahead, that's 1079.

I

They're right Roley.

B

So what actually needs doing with that love? It.

H

D

Good spectral units stuff and this planet at all, it's it's.

H

Just that there is some money array, sparse one side to another and and so many units that they have need to be properly set, because you have the frequency on the side. You have the data units of the measure by the instrument and how their I mean all this magic from Florian how they get parts to one side. Another is kind of an interesting.

B

Yeah I don't know now what the test coverage like.

H

In we're in a spectra, yeah, yeah I think.

B

It's okay so thinks that makes it easy to just throw it on the wall until it stick.

H

Ya know, but there's many that are like plotting stuff. So probably that's not tested really.

B

Just so, what else can.

C

We add a priority tags to these, just so that yes, please yeah, so that you know I think a priority tag would be good anyway, but um for this so that we know which ones to attack first and that way you know the high party ones hopefully get finished and then other ones if they don't get to it. It's okay.

B

Because people agree with enemies, there are some prioritization to do on this list of things, but to be honest, most of it needs doing. If we're ever going to release note which sticks in the way we want to.

C

Yeah, but some of them I mean the ephemerides I don't know. Does it need to be no point six example.

B

So long ago we stop this. I don't even know where don't even know why we started doing this in the first. But what is.

C

B

C

Point of it well, we had pulled in the sunpass son Thunder pro from idea at some point and we.

B

C

Differential rotation, stuff yeah, some of that stuff I mean the Sun. The Sun stuff was basically so that you could get as I recall. So they get the distance to the from the earth to the Sun to add two maps to fit both didn't have it, for example, and the.

H

Angles, the beasts here and the P angle right.

C

E

Yeah, you need it. If you didn't do any kind of.

E

Tracking on the solar surface, you need some kind of notion of where the Sun is, as seen from along the son hath line right to thats all my case that we covered at the moment.

E

We don't even cover so and so who is you know, one percent coastal that was a correction animal and a one-percent closer to the Sun. The authors there's a correction and solace off the bat, but we don't even have that.

E

So we only have at the moment the soul of differential rotation stuff that I've been working on, has just the just a son and asked.

B

What is that our declination and today, right now, what is a apparent white ascendant in an apparent explanation? Oh.

C

What is it yeah.

H

There's something a wikipedia: it's is the one without.

H

For the synodic days simpler, what.

B

F

E

F

B

H

The one of the brain and an apprentice is just depending of the.

H

Something to do with this naughty bass, drummer I.

B

Don't feel so, this should be this complicated. The idiots.

H

I mean I I can I can work on that I have the book and I have the staff and it yes I just need time to do it. So this brain is good for me.

B

Yeah young, with that other units inspector you need to spectra, is going to be a problem, but we can sort well.

H

Yeah and and pay Florian through I'll blast with us.

B

C

So I still think would be useful to prioritize things as especially just prioritize things for the Sprint, because some things are better than you know, together with people and some things or not, I feel like.

A

How would you prioritize, uh how would you implement a prioritization check.

C

I yeah with tags yeah, maybe just tag high priority high, and that would be good enough.

C

You know like deprecated python 2.6. No, that's pretty much done anyways foolish right, but it still doesn't have to be high. Another right, for example,.

C

C

Unified unified downloader, I would say that I.

C

Fixing the sample data step so that we can get in the.

C

Removing the moving the data, the test data I, think that would be high coordinates, should also be high.

C

Achieve eighty percent testo coverage, I, don't know, I, don't know whether I would say hi for that one, but it would be great tab.

B

Achieve eighty percent test coverage to me is start testing plots if we really need to do that so hard yeah.

C

But that's pretty hot.

B

I know it's not that hard rajul started working on it. I heard, and he said he was going to come so hopefully we'll be done.

C

Okay, so that the testing for that is basically creating a some sort of output like a PNG or something then you're testing against that that right, yes, the challenge.

B

Becomes where to keep all the pain Jews? But yes, nice.

C

And then, presumably, if anything changes in at pot lib such that the pixels change, then when.

B

You regenerate angel right.

C

E

Whoa very different versions of map college I'm, gonna support like 1.4 1.0, 1.0, 0,.

B

They can seriously support I think probably need to limit ourselves for any testing on one, or else it takes seven years for our Travis building. You.

C

Know what one thing I've been wondering is, so we should have tests.

C

This is something I've been doing myself, and maybe we should consider having it more general, which is like we can have tests that ship with some pie, which tell you whether some pie works, but then we can have some more extensive tests that we don't ship with some pie. This is something we do ourselves.

C

For example, I have a test that goes through and downloads all the goes theta, because I want to make sure that, because it has to span a few different times and the file names changes, and it's also testing whether the files available you know, but that's not something that I would that you would have in some pie tests well.

B

I think it should be cooped up in some photos. We could have a special set of tests that are marked like long-running or something that I've run by default.

C

But I mean that does download all the data, which is oh yeah.

A

Love doesn't matter, though, seems like that's something that we should ship and I, don't see why we should have shipped something like that. We just make sure it doesn't run by the ball. Yeah.

C

Okay, sure sure, um okay, so that means really would want I mean.

B

That way to change the levels.

C

Of tests like one, what.

B

We already kinda do in that. We have online ones and offline ones exactly. There was any other way.

C

I'd be able to what.

B

C

Do so, then, for the for this testing the plotting we could just like, we have a sample data down a little bit or we would have a test plot downloader that would first download all the pots and then compare them. Maybe.

B

C

Know that would be pretty easy to implement, basically just cover you just copy either I've been having this.

B

Conversation on on Astro pydev about doing this, yeah.

C

B

Probably wait, favorite solution so far is have a separate get module, which is a soft module where you can not keep the history.

C

What do you mean ocupe history, the.

B

Problem is their tests. They should be able to scroll back through the history mm-hmm. The problem is, if you put all of the test, plots in the main some point: repo, the repos in a balloon right.

C

B

That's one thing:.

C

You should you should now keep them there. Just like we move the sample data off. It should just download them when they're needed. Although.

A

I mean I think Stuart's points, unlike the sample data, the tests it'd be nice. If we could you.

B

A

B

On an older version, yeah you should do. Tesla should be up to run note version. If you check out any point of the gate history, the whole test suite should wrong and hopefully succeed.

A

C

I mean we don't we don't have to use, get to do. Versioning.

C

Especially for binary files, you could just make sure that it goes up it. Like knows, you know, checks the my plot live version that you have, and then it goes off and downloads the proper files, because every we would have to recreate the files. You know it would be a one-time event. You recreate all the files, presumably put them in a new directory, and then you still have the old ones. It would be a of a set event when that would happen.

B

Mm-Hmm well, I mean I. Suppose the bed did actually the best solution is get annex, but that brings a long, a whole novel level of pain because people have to have get up, get annex you stall to use it, and then we have to have somewhere to host the date. I don't know it's not it's not a problem with an easy solution that anything.

C

Well, if we don't, if we can't agree or come up with a good solution, then by Wednesday and it will it be very impossible to achieve the eighty percent Tufts coverage on 1 175 with that plotting test. Yeah.

B

It's probably possible.

E

You said what percentage of our tests involve plotting.

B

127, the basement, probably a good five percent of the whole code bases plotting code in lines of code. So it's.

E

We've got bigger fish to fry other than plotting on smell, ready, we're at 76 or something ray so forget, we're doing any purchase at all. Yeah.

C

I mean I actually think the coverage is not so much important. I think well. Coverage is good, but I also have looked through the test that there's a lot of things that we don't test for, even though the coverage says that oh sure.

B

Yeah, there is definitely a difference. Yeah.

C

So uh I would think that a besides looking at the percentage, I think it would be good for us to just look through all the tests and add the things we need think we need- or at least stubs for things, that we need and then and not worry about this percentage coverage thing so much.

C

That would be my recommendation.

C

B

Angie's coverage thing is a good indicator. You can definitely look at it and see where the big holes are yeah.

C

Yeah yeah, yeah, I, agree, I, agree, but I think the big holes right now are so big that they're, like pretty obvious when you just look through things.

C

So I mean I would just think that it would be more worthwhile to keep that in mind, rather than slavishly try to achieve this. Eighty percent, if we were to achieve it, that would be great, but well.

A

I guess I know what you're saying to you, but you know having that eighty percent to go for I think will serve that purpose at their parallel purposes. Right right.

C

But then don't necessarily overlap right. You may you may write a test when you know that there's another test that needs to be done, even though it covers the other test, is important and covers the code. It's already covered technically, but they'll decide to write the test for something that's uncovered and ignore the other thing that needs to be done as well.

B

Well, the simple solution to that is: don't.

A

Know what Steve saying so that seventy-six percent or something six percent, that isn't actually testing the right thing. But.

B

Sure there are bits that are missing still, but in some respects having some test coverage, at least if we get to the point where we're executing every line of code they're going to reduce our bugs by a like reduced, where the box behind by an amount. Obviously you can still have light. You can still not test edge cases and execute every line of code, but we're not even a queer, not even importing all of the different modules when we run our tests.

B

Cesses test cases, women while I'm- definitely not disagreeing with Steve I- think it's more important to actually do something for Allah a percentage of some part than it is to get oh yeah. Obviously, both are important. The late I don't think that there, by when we hit like ninety percent test companies, and yet we can start making issues that go through and go. You know this needs more tests missing stuff, yeah.

A

Well, I'm curious, like I'm, definitely siding with Stewart on this sentiment, but Steve. What in your mind, was like the largest holes in our current testing covered that wouldn't be obvious from a coverage kind of fun of you, hey.

C

That's a hard question answer: I'd have to look through I. Just remember, I mean if you look at a lot of our test. Suites they're fairly, sparse and you know a lot of them, for example, don't even really test all the input input possibilities, for example crew yeah, so it it looks like you know it's well tested, but in reality you know it has some holes, so I just thought I'd bring that up, but I think it. We can follow this and and I mean you're talking about test quality right I mean yeah yeah.

A

Exactly cool I just was back to success. That is fine. The coverage quickly.

C

A

C

Barrett I would just like: let's just keep that in mind when we go through that, that's all I would say on the.

A

You might as well just comment on the you know the eighty percent issue different 1170, whatever that was and stressed that yeah.

C

A

Are we currently at you said 76 Stuart, thanks I'm, afraid some of these I mean I'd to saying that that pisses me off the most when it comes to the test, is because the is the part where I had been handcuffed to the strongest the goddamn map rotate.

A

It certainly would be nice to have plot based testing there, because that would have been.

B

A

B

Need prophase testing, it is one of our most beautiful eating. Bits of API is are playing API and the fact that it breaks racially bike goes peak goes peak, is broken or not. Point 5, I'm a public one point for its map. What they've changed? Some things you broke, we didn't know about it until I got a message of Larry Manning on hangouts, a few easy for going here, broken and like.

B

We don't know when some of our most user facing code just explodes because of various reasons. It's not good yeah I mean especially plotting. We need to find.

C

A solution, a very visible problem, yeah exactly and.

B

You know doing it will, as I say massively increase RM about 74 and massively a code is my tower. They do the same thing store a store, a PNG test against it. A lot.

E

B

Stopped changing that to be fair, probably in the repo, so.

E

Those what we need to do, then we.

A

Have that listed as little sister.

I

A

For plot based testing of puff pastry things, I.

F

A

So and I will test the blood with the gallery input or met affected to test plots for 10 open, marched. Fifth 2013.

A

That's an old one.

H

Yeah, you know we use the gallery to test blocks. Oh.

B

But then we've got when we go run the gallery. Somehow, like we run our test suite.

C

C

C

There is an interesting solution, though, that you would run the gallery running. The test would run the gallery, so you have the gallery there. I, don't.

B

Know not well of notebooks doing stuff out notebooks in version control is too fraught with problems. Well,.

C

It its proper problems, because it's including binary, exactly, which is what we would have anyone.

B

B

Yeah not the best thing in the world to put in version control system, you're.

H

Very careful I, don't think, there's the gallery sale, but this usually alkyl of the gallery of the plots to test the plotting.

B

But you know that is the PNG's. You feel more easy, yeah.

C

For for the peak fix stuff that I did, I did I mean that that is how I tested it. I had I have an notebook that does all the plots and then, whenever I make changes, I just rerun the entire notebook and then check all the plots.

C

Visually is a rare, a visual yeah, but the plots are in the notebook. So it is an interesting solution.

C

Yeah but PNG's are probably the way to go, are.

E

Those some independent of operating system that is matplotlib independent and we want.

B

Someday it's if you use, if you use the included a GG back-end to save the files, it is in theory operating system dependent because the whole staff is included in that bother. You.

C

But I mean / for most people they most people would not be running this. Let's I would say what we need to do is come up with a solution that we can run it's not as a part of the perfect solution, but it's something that you know we have available to us and then we can consider how we would make it possible for anyone and to check and run, but.

B

It's at the same problem: we have a testing infrastructure. The easiest way for us to run tests is using our testing infrastructure. We don't want to build another, their.

C

Life is going to be more work.

B

C

But like there are lots of issues that are complicated because we're trying to think of this as a general solution for everyone which is like.

C

Your computer Oh or you know how do we handle multiple versions and things like that and weak to that did.

B

The plotting stuff, with multiple opportunities that.

C

We have right now the.

B

Plotting multiple injections of Mary motley is a challenge, but at the same time in worry.

A

B

All right the multiple versions map eclipsing. We can force our test suite to use the most recent one because at least then it works it's better than where we are at the moment.

B

Like and yes different versions of our fingers and get away, but like you can't test, we cannot test against a possible version of our entire dependency stack, because I'll build matrix would be for normal.

C

Yeah, it's me. Yes, it's not coming up here.

C

B

C

Can you guys hear me? Yes, sorry, I missed all that.

B

Basically, we need to get one person right. We need to get. We need to include some plot tests, even if they just not even testing the output of the plot, even just executing the code would be a good start.

B

Even if we just test that we get flipping matplotlib object back out of executing plot would be an improvement. Yeah.

F

Okay, less then.

B

If something breaks- and it comes up with an arrow- we know that they come up with an error when you try and plot it yeah well, they'll get there is a phenomenal improvement or where we are the moment. Even if we're not testing the output from the damn thing, yeah in fact, actually I opened an issue to facilitate that change. Yeah in fact, I over the pull request to facilitate that train key of the day where's he gone 1232.

B

A

Stuart, when we're talking about comparing the PNG you're talking about a pixel-by-pixel, compare right, it has two PS. I had.

E

A

Why can't we just rehashes that? Why do we need you? What are you gonna hash just? Do it get a 256-bit hash of it? Just compare the hashes of the PNG and save the hash yeah. Are we saving the whole PNG I was looking at this yeah no work, yeah.

B

C

The way just say that I'm.

A

Somewhere, yeah and we're version, control of hashes and we're happy store over I mean it's going to hit problems when, like packages update underneath our feet, yeah.

B

That's that's yeah! That's an interesting idea. I like that, like Cooper work, is there a decent hashing library included in the Python standard? My break I have no idea, but the huh sure I.

A

B

Anyway, we should merge 1232, int, int, pull requests in.

A

My computer crashed by Universal.

C

A

Yeah I was just looking at that one of my computer crashed so or does that only affect tests that are run on the whole suite if you ran it on, you ran to know so I think it's any time.

B

You actually keep I test from within the directory from within the test directory, not from within some pie. Oh any tests, oh I, can remember, hang on I'll check. If you run the tests using spite and setup reply test, then you will definitely use.

A

It right right, I'm, just wondering I usually just run this small set of the test whenever I'm testing software, obviously yeah cuz well,.

B

Then the setup to apply thing supports dash capital p for the name of a subject that you want to run. So you can do part and set up don't pie. Yes, they be caught that.

A

We're feeling we'll use a snap light lip line, but if I go in and definitely use it, I just go in and try to act as a particular test file and, of course, all this file, but don't know it sweet I'm done.

B

A

A

Mean I'm fine with merging this, but I just just curious how far it sends.

B

B