Transcript #284: Spicy git for Engineers

00:00 Hello and welcome to Python Bytes, where we deliver Python news and headlines directly to

00:04 your earbuds. This is episode 284, recorded May 17th, 2022. I'm Michael Kennedy.

00:10 And I am Brian Okken.

00:11 And I'm Daniel Mulkey.

00:12 Daniel, great to have you here.

00:14 Thank you. It's an honor.

00:15 Yeah, it's an honor to have you. Now, before we get into our first topic that Brian's going to

00:20 tell us about, just give us a bit of your background.

00:22 Sure. I am an optical engineer in Southern California, but I have a significant amount of my time spent using Python for data analysis,

00:32 instrument control, and other things. So I've been doing it for a better part of the last five

00:37 years. And I've had a back and forth relationship with MATLAB and have finally married to Python,

00:42 so to speak.

00:43 Fantastic. You've finally been able to get out of your dysfunctional relationship with MATLAB.

00:48 Yes, exactly.

00:49 It sounds a little bit like you might live in a parallel universe to Brian.

00:53 Yeah, it sounds like it. We should definitely get you on testing code.

00:56 We can BS about that.

00:58 Sure. Yeah, I'd love to.

01:00 Brian, I would love to hear about our first topic. You want to talk about it?

01:05 It sounds very distinct, you know?

01:06 Distinctify. Yes, very distinct. So I ran across this. I can't remember how I ran across it. I guess

01:12 it doesn't matter. But one of the things I like, it's a Python package called Distinctify.

01:18 And it's very simple. It's a lightweight Python package to provide functions to generate colors

01:26 that are visually distinct from one another. So I was thinking like, you know, you got a chart,

01:33 like maybe you're taking user data or something, and you don't know how many lines you're going to plot,

01:37 but you're going to plot a whole bunch of lines. How do you pick the colors for what the lines are?

01:42 So this is a kind of a neat thing to just pick visually distinct colors. Pretty focused,

01:51 but it's pretty cool. And all you do is you kind of just give it, you give it like the number of

01:57 colors you want, and it gives you back the colors. And you can, it has display capabilities. So

02:02 you have to install extra stuff to make that happen. But you can display color swatches too

02:08 with it. And I was looking at some of the different colors that are available. Like one of the ones was

02:13 15 different colors. I think it's 15 colors for normal vision versus some color blindness. So if you

02:21 have colorblind people, you can pick based on some of that stuff. There's a whole bunch of examples in

02:27 the repo too, that it's kind of fun to look at. One of them was the like the normal colorblind one.

02:34 So was that it? No, that wasn't it. But there's some really cool examples of different colors. So if you just

02:42 give it a few, it just grabs a few, of course, but there's a whole bunch of neat ones, clusters and things.

02:49 So anyway, yeah, cool little library. It's great. Yeah, I like that they have, I noticed when I was looking

02:55 through it, they have a function for generating a color palette. And so you can generate a colorblind

02:58 friendly palette. So hypothetically, that works well for visual colorblind. And if it's in print and you're doing

03:04 black and white, so that was the most interesting thing to me. Oh, do you mean it has black and white?

03:07 That's interesting. Well, at least I think if you take a colorblind palette and you make it black and white, typically it's still a decent contrast. Oh, yeah. So you have to like worry about

03:15 printing things out. Oh, that's cool. Yeah, that's great. And one of its functions

03:20 is to take the color map that it generates and turn that into a matplotlib. Oh, yeah, yeah, yeah. It's cool.

03:26 Oh, that's what I was looking for.

03:28 Yeah.

03:29 Oh, wow.

03:31 And there's somebody in the audience who just found out they're colorblind.

03:34 Yeah, go ahead, Daniel.

03:35 And there's somebody in the audience who just found out they're colorblind.

03:38 They're like, is there a difference? What is this?

03:41 So my, I, yeah, one of my kids found out like in high school that they were colorblind.

03:48 So interesting.

03:50 Yeah. How would you know?

03:51 Yeah.

03:52 For a long time, you're just like, people tell me that's a color. I guess I'm not great at picking

03:55 out that color or something.

03:56 She had an art teacher said, I really love how you use both blues and greens in the sky.

04:01 And she was like, I intended to just use blue, but thanks.

04:05 I have a friend who went to art school and that was essentially his story that he always had really

04:11 vivid color choices because he didn't see the same as everybody else.

04:15 It was great. It was awesome.

04:17 That's pretty cool.

04:18 Yeah.

04:19 Cool. All right.

04:21 Ron, we ready for the next one?

04:22 Definitely.

04:24 Okay. So let's talk about SQL Soda or Soda SQL. So this is a open source CLI tool that if you're

04:33 doing like ETL, like ingest, transform loads type of stuff, doing other sort of analysis or

04:40 exploration of SQL data, it allows you to connect to your data source, like your database, and then

04:48 define tests for what invalid data looks like, right? Does this have to be a number? Can it,

04:53 does it just have to be not null? You know, what is it? So for an example, here, they're talking about,

05:00 here's the YAML file for a, like a warehouse, a data warehouse reporting type thing for Postgres.

05:07 So you just set up like your connection and your host and, and all that kind of stuff. And then

05:12 off it goes. So pretty neat. And then you can scan your data set to run tests against your data.

05:19 Isn't that cool?

05:19 That's right. It's Soda cool.

05:22 It's Soda cool. It is Soda cool.

05:26 Yeah. So you just say Soda scan and you give it the YAML file for the connection information and then a YAML

05:34 file for the types of things you want to test. So they've got this example of how you're talking to

05:39 one of the data warehouses and it's going and pulling in these config files. And it basically,

05:44 this example, it's testing 54 different conditions. Three tests were executed. Everything's good to go.

05:51 So, you know, if you're getting kind of data dropped on you or you're scanning, you know,

05:56 scraping data from other places on some kind of background job and you want to bring it in,

06:01 you know, if it's all automated, how do you know when it goes wrong? Right. So here's a nice,

06:04 simple way to express that.

06:05 Yeah. That's neat.

06:06 Yeah. And Brandon out in the audience says, I think we're looking at great expectations for this same

06:12 thing. And yeah, this is kind of a, I guess my, my first impression is this is a less code

06:18 way of doing what great expectations does. Right. So like you can just put together some YAML files

06:23 that define, you know, what you want to test for. Right. So for example, in this YAML file,

06:28 I can say the metrics are row count, missing count and missing percentage. And then I can test that the

06:33 row count is greater than zero. Right. And then the, another one is for the column, for the ID,

06:39 it's a UUID that it's, I'm allowing 0% of the UUID format to be invalid. Right. You know,

06:46 that's got like a certain structure to it. Right. It's like a, either a straight UUID or a string

06:50 that looks, that can be parsable over to one, I'm guessing something like that. So pretty cool. I

06:55 think that's probably the biggest difference. So if you just want to define kind of like declaratively,

06:59 like here are the conditions of which I want it to test. And then you want to just set it up to

07:04 continuously scan it. Looks good. The invalid percentage looks interesting because,

07:09 it's, it's an interesting addition of like, you know, there can be some bad rows, but we don't

07:16 want more than like 20% bad rows or something like that. Right. Right. Maybe you can't have zero

07:21 errors, right? Like you just, sometimes the data is just not there, but if it's a hundred percent not

07:27 there, then something's gotten terribly wrong or the data formats change and it's not called that

07:31 anymore or whatever in JSON, who knows? Daniel, what do you think? My, my data is always in CSV

07:36 files. So I have, I guess there are pros and cons to never having touch SQL, as I've heard

07:41 much, much easier to version control. Just put the CSV. Yeah. Anyway, I think, this one's pretty

07:51 neat. People can check it out if they're doing relational data stuff and they, especially if you're

07:56 doing a lot of like on demand, you know, not like you ask for it, but it's just on demand

08:01 processing or you're given a database and you want to check it out to see how it's, how it's doing.

08:05 So I won't go on anymore on that because I've got a ton of other extras. So kick it over to you,

08:10 Daniel. Cool. So let's see. There was a review article back in 2020 published in the research

08:18 journal nature, for anyone not in the research articles world, nature is one of the top level ones

08:27 for referencing grad school. We had some fancy work we did with quantum entanglement and we got rejected

08:32 by a sub journal of nature. So to get anything into nature is highly non-trivial. I will add the,

08:39 it's like the JAMA, the journal of American medical association of science. Basically it's absolutely one of

08:46 the top ones. I will say it's a review article. so it's easier typically to get a review article

08:51 than to say, Hey, this is bleeding edge research. It's going to change the world, but still the big

08:55 news is two things. One that there's a article by Travis elephant and others on array programming with

09:02 NumPy in nature. it's a big enough deal that they chose to publish this and it got through.

09:07 And it's, I think very significant that that software was something that was good enough to

09:12 publish. the other, and you know, they go through and they talk about kind of the fundamentals

09:17 of it all. There's one diagram I really like that sort of shows how the whole ecosystem stacks up.

09:22 You've got numbers. Oh yeah. That's a cool visualization. Yeah. And then you got scipy and

09:27 matplotlib and you know, the other plotting libraries. So there's the foundation. Yeah.

09:31 I was just going to say for people who are listening, it's like the tree of life for scientific libraries.

09:36 Sorry. Go on, Daniel. Yeah, that's absolutely right. So from that foundation, as far as algorithms

09:42 and plots, you go up to like specific method you're using, are you doing image processing?

09:47 Are you doing machine learning or something else? And off to domain specifics like AstroPy. And I think

09:52 you've had those guys on Python. So you've gotten to talk to them and then down to very

09:55 application specific. So it, you know, NumPy serving almost everybody who does anything numerical down

10:00 to like Q-tip, which is used for people working on quantum computers. So very large breadth being

10:07 discussed here. Q-tip. That's so cute. I like it. and yeah, so it's notable that Python got into

10:17 nature. And if you go search for Python, there are a lot of other articles, but it's also interesting

10:21 to see that they're willing to publish software. You guys have talked in the past about how you can't

10:26 always publish the software package and any research journal. So how do you get credit for that if you're

10:31 in academia? but this is an interesting take to see that nature goes to publish it.

10:35 Yeah, this is super interesting. And I think it's, it's very valuable to just raise awareness,

10:40 right? It's, you know, this is the water that we swim in, but not everyone. Everyone is immersed in

10:45 the Python data science tooling, right? Yeah. There's a lot of authors on here.

10:49 Yeah. I was trying to understand. I'm guessing those are the maintainers of the packages that

10:54 were included, but I mean, you, you don't have 20 people write one paper, so I don't know how,

10:59 I think it's, it's kind of like the LIGO papers or like the gravitational wave interferometer ones

11:03 where like this crazy list. It's like the first page of the article is almost all authors just

11:08 because there's so many people that worked on this for so long. So I'm guessing that's

11:12 yeah. And you can access it. Some, some, articles, some journals, you can't actually read

11:19 it unless you have a subscription, but this one's available. So.

11:22 Indeed. Yeah. A very cool pick before we move on. Maybe, you know, Daniel Alvaro and audience

11:30 has have any of you come across a way to validate pandas data frames against a schema, much like SQL,

11:36 soda, soda, soda, my scope. I feel like we have, but I don't remember, but yeah, I don't

11:44 remember either. Sorry. Maybe something we should seek out for the next one. And I think we might

11:51 get some answers in the audience. So we'll, we'll let them, inform us as we move on. So Brian,

11:56 what's next? well, this isn't Python specific, but I think a lot of Python people are using

12:03 GitHub actions. so, GitHub announced, I guess recently, supercharging GitHub

12:10 actions with job summaries. That's an article that we'll link to. And, basically the, it's pretty,

12:17 it's pretty cool. I can't wait to try this. I'm using GitHub actions. And, the gist is you can

12:23 now have Markdown go directly into your GitHub job summary sort of thing with like this,

12:30 this crazy, global variable called GitHub step summary. but it, it's got marked out to it.

12:39 And I'm like, well, what can you do with this though? But, Simon Willison, released,

12:44 was, was tweeting about it. And, and then said, and then Ned Batchelder said, Hey, I'm using it too.

12:51 So Ned, has a little example on his, on coverage.py that shows, what does it show?

12:58 It shows, you, you get this nice total coverage percentage. If you want to put that

13:03 in your, in the coverage for your, your, your repo, you can do that. Interesting that

13:10 coverage.py is not a hundred percent covered. I don't know why I find that funny.

13:16 The irony. I love it. but, and then, so Simon also listed, data set is, has an example on data set, you doing, adding some extra stuff, to, to,

13:30 what is he adding changed files? Oh, he's got a, a tool that does, looks for,

13:36 how many files have changed and, and recently. And he actually just wrote, he just wrote a

13:42 write write up for that. So we're linked to that as well. So GitHub action job summaries,

13:47 and he shows how it, how it works. You can pop, pop out stuff. And I love Markdown.

13:52 So even little code fences and all sorts of stuff. That's very cool. If you want to structure something

13:56 real nice like that. Yeah. It even has, so, so supposedly it's got a whole bunch of stuff.

14:02 It's got like, you can do tables even. So that's neat. And emojis. Why not? So,

14:08 Oh yeah. Pretty cool. Put a little fire emoji in there. Yes. Do it. Does anybody get images?

14:13 Like if you create an image during the action, can you reference it? I do. I don't know.

14:18 Didn't doesn't mention images, but maybe you could base 64 and code it and embed it as a data.

14:23 Oh, wow. It even does a mermaid, which is a way to do diagrams, within it. That's pretty neat.

14:31 Very nice. Like flow charts. Yeah. Fantastic. This is a good one. I need to learn to do more

14:36 with GitHub actions. I don't do very much with them. I love them. They're like, it was, I used to use

14:41 Travis back in the day, but, and, I think these are way easier. So, you know, do you do any of

14:48 those sorts of things? Any, CI automation type stuff? at a previous company, we used Azure DevOps

14:54 and set up some stuff to build packages and build applications, but, not at the moment. It's just,

15:00 it doesn't, it doesn't happen to be any code bases I have, but I need that.

15:04 Yeah. Very cool. All right. Well, I've got an interesting one here. I want to dive into it,

15:09 you guys. So this one, let me give some attribution here. This one was sent over by

15:15 Antomar, from meta. And then this is a writeup by Alex way good. And what it is, is it's the,

15:24 basically the, the notes for all of us who are not there for the 2022 Python language summit.

15:30 So that's pretty cool. There were around 30 core developers, triagers, and special guests gathered

15:36 the day before PyCon. And so they, they had a bunch of different talks and ideas they discussed

15:43 quick summary. Really it's about so much of this is about performance and parallelism right now.

15:50 And then there's a lot of maintainability back channels, back flows here. All right. So

15:58 coming to these first, Sam Gill made a huge splash last year when he talked and he introduced the no

16:04 GIL work that they had done for, I thought I'm three, eight, I believe. I can't remember three,

16:09 three, nine. No, it was three, nine for them. Cinder was three. So for three, nine, and there's a lot of

16:17 interesting optimizations and whatnot in that talk. So the idea is, could we live without a

16:23 global interpreter lock? Larry Hastings tried the galectomy, sort of said, you know, it's too much

16:30 of a penalty to try to live without it. But this no GIL work that Sam Gross did actually had very small

16:37 overhead in terms of what it added, but potentially removed some of the GIL things. So there's a lot of

16:43 analysis of that. People were excited, but they, how is it written? It says robust, there was robust

16:51 questioning. One, I guess one of the biggest parts that they discussed was maybe this should be a fork

16:59 of CPython. There should be a no GIL version of Python. And, but Sam is like, I really don't want to

17:07 have just another separate version of Python. I really want this to just help everyone. So pretty

17:13 interesting. I think originally it was maybe going to be a runtime flag you could pass to Python,

17:18 but it's looking like it more likely is going to turn out to be a compiler flag. So you'd have to have

17:23 a no GIL build, even though it's from the same source code. So yeah, a bunch of interesting things,

17:29 concerns about how it's going to work with like C libraries and so on. But that's, there's,

17:35 all these are pretty interesting read-ups, reads, write-ups. So Eric Snow did a presentation on his

17:43 per interpreter Gill, which is interesting in how it approaches a slightly different problem

17:49 than say Sam Gross. So Sam is trying to get it out of Python. Eric is saying, well, if we could just

17:55 have a sub interpreter, like a little mini in-process interpreter that runs per thread, then they can all

18:00 GIL to their heart's content. It doesn't matter because it's all single threaded, right?

18:04 But what's interesting is if you go look at this one in here, we've got this one. It says something

18:12 like way back in 1997, this idea of multiple sub interpreters was added by Guido, but it really

18:21 hasn't, nothing has been done with it. And when somebody tries to do stuff with it, there were

18:27 thousands of global variables. And if you're going to have per interpreters, you have to somehow have

18:32 those not shared because then you're going to have the GIL back on them, right? You have that locking.

18:36 So due partly to the deprecation of some of the old libraries and stuff, it's gotten a little simpler, but

18:44 no, that was it for the next write-up. But anyway, they reduced this to almost

18:49 1,000, to 1,200 remaining globals.

18:52 So needless to say, it is not totally solved here, right?

18:58 So again, one of the possible worries of all this stuff is, well, how are the C extensions going to deal with this?

19:03 Like they don't know about multiple sub interpreters.

19:06 Yeah. So anyway, that's another one of the main threads going on there.

19:10 Let's see.

19:11 Then this is probably the biggest deal.

19:14 This is Faster CPython 3.12 and Beyond by Mark Shannon and Guido Van Rossum.

19:20 So stepping back, a release.

19:22 Python 3.11, if you haven't heard, is fast. It's supposed to be 1.25 times faster than 3.10.

19:29 How about that?

19:30 Yikes.

19:31 This blows me away. In one year, they were able to make Python 1.25x faster, and it's been out for

19:38 30 years. It's not like, oh, well, we released it last year, and now we've learned some things.

19:42 It's really, really, really solidified in the way that it is, and then still, there's a lot of work.

19:48 And this apparently is just the beginning.

19:50 This is like a five-year plan to add all sorts of optimizing JIT compilers and all

19:56 sorts of things.

19:57 How did they quantify that, or what subset of the language was I tested on?

20:02 That's the tricky thing to say.

20:04 Python is 25% faster.

20:06 Doesn't matter what you do. Even if you're just waiting on a database, it's still 25% faster.

20:10 Does it just overclock your computer in the background?

20:12 It liquid cools it.

20:14 I believe that number comes from the unit tests, like all the tests for CPython.

20:22 I'm not 100% sure, but I believe that was the conversation.

20:26 And so one of the big things coming is possibly a JIT, an optimizing JIT compiler. So right now, they've

20:32 found a way to optimize individual bytecode instructions to make the runtime

20:38 smarter and go, oh, I see what you're trying to do.

20:40 We could have a specialized version of that. But that's on a per line basis. Like, how about

20:46 inlining this method? Because I only see it called in two places or something like that, right? So you need something that

20:52 can look more broadly at the code. So that's this idea of the JIT compiler and

20:56 so on. So yeah, this is really good. But all three of these things I've talked about

21:00 are like, both, they might help each other, but they also might inhibit each other, right? So like the

21:06 no-gil work might interfere with some of the optimizations that they're doing

21:09 over here and the multiple sub-

21:11 interpreters also might be some interplay that they've got to be got to be worked out. So I'll just summarize the rest.

21:18 WebAssembly. And so we've talked about PyScript last time and Pyodide.

21:24 Pyodide. This is the official CPython build target for just CPython. So this is really

21:34 interesting. That is sort of a more from the core devs rather than somebody coercing

21:38 CPython into a different build on their own. So that's pretty neat.

21:41 F-strings. Apparently the F-string parser is kind of this weird side parser thing that's not actually

21:48 part of the Python code parser. But now we have peg, the peg parser. It can support more of this.

21:53 and sort of unify that.

21:55 So yeah, there's something like 1,400 lines of customized C code for parsing F-string.

22:02 Well, the people who wrote it knew. They did a lot of work.

22:06 There's like 600 of the global variables right there.

22:10 Exactly.

22:11 The most important 1,400 lines in all of Python right now. They have string functionality.

22:19 Then two of the big optimizations from sender.

22:24 That's the Python 3.8 specialization from meta.

22:28 One is, this is a presentation by Itmar Osterreicher.

22:32 So this is the person who sent this in actually.

22:36 This is looking at async methods.

22:40 And if you can be sure it's not actually going to await, treat it like a regular method.

22:45 So, you know, if you have an async method, you might say, do this, do this,

22:49 do this.

22:49 If I already have the value in the cache, return, else await database call,

22:54 right?

22:55 If you already have it in the cache, why do you need to create a co-routine,

22:59 schedule it on the back, on the loop, wait for the loop to get to it, and then return?

23:04 Just boot.

23:04 Just call it.

23:05 Like a regular method, just give us the answer.

23:07 That's the idea.

23:09 There's some interesting ideas that it might change runtime ordering, although I don't know

23:14 there was any promises of runtime ordering, but yeah.

23:17 So that one's interesting.

23:18 Also, the issue and PR backlog, now that we've moved to GitHub, apparently,

23:24 there are issues that are still 20 years old that are still open.

23:29 And traditionally, the core devs and the triagers and so on have approached

23:36 these things like, well, should we close this or probably we need to keep it open

23:40 because it's important for historical reasons.

23:42 And they're starting to talk about like, this is not helpful for anyone.

23:45 Maybe our first question is like, why should we keep this open?

23:49 And if the answer is not clear, just close it.

23:51 There's a lot of talk about, well, this historical stuff and maybe someone

23:55 wants to pick it up.

23:55 If it were me, if I got to pick and obviously I don't, so it doesn't really matter,

23:59 I would just go, if it's older than two years, just close it.

24:02 Like there's a script that just says, over in two years, select all, close.

24:05 Now let's go through and figure it out because at some point, you know, if you've got 20 years

24:10 of you should make this change, maybe even, maybe these, these things aren't even

24:14 relevant anymore, you know, or things have moved beyond it or it doesn't make sense

24:17 in 2022.

24:19 I don't know.

24:19 But I'm just, mostly what I got out of the articles, I'm thankful that I don't

24:23 have to deal with 20 years of issues and PRs.

24:25 But also, they don't go away if you close them.

24:30 They're still there if people really want to see them.

24:32 You can, so I think they should be, maybe two years might be a little extreme,

24:37 but at the very least five or three or something like that.

24:40 There should be a number where that's true and that number should be less than 30.

24:43 And it's, and it's a smaller number than 20, right?

24:47 So, yeah.

24:48 All right.

24:49 This is a long section.

24:50 Last thing, I'll close it out with this.

24:52 Immortal objects, the path forward for immortal objects.

24:56 So let me ask you guys this.

24:58 Can you change none or true or false?

25:00 No, right?

25:01 Does, do you think it's ever going to go away?

25:04 Like, are we done using true and then it's just going to get garbage collected

25:08 or reference counted out of memory?

25:09 Nope.

25:10 But you know what?

25:11 Every time you interact with true and false, it's still incrementing its ref count.

25:15 Interesting.

25:17 And none and stuff because it's an object, right?

25:20 Oh, yeah.

25:21 And so, this discussion is like, isn't there some that just shouldn't be participating

25:26 in reference counting because they're, they're just fundamental to, you know,

25:32 like the idea of a class, like the, the structure of a thing that defines

25:36 what a class is, true, false, the numbers, like the low numbers, like there should be

25:40 some that are not consuming that memory because they don't need to keep track

25:45 of that section and so on.

25:47 Right.

25:48 So anyway, this was the proposal.

25:49 Again, it's complicated is the story, but yeah, I do something a little bit like this

25:57 on Talk Python, the training site.

25:59 So I've done a lot to tweak the garbage collection around there and really change

26:04 the, the defaults of like, what are the triggers for garbage collection?

26:09 So if I've got this many allocations and so on, and one of the things you can do

26:13 is you can tell it from here on, like what has existed up until now, freeze that

26:19 and don't, don't look at it when you have to look for cycles.

26:22 Right.

26:22 So I just, in my app startup when it's a, it's kind of imported the things

26:27 and it's about to start, it just says, okay, everything that you've done

26:30 to come to life, just don't, don't trick that anymore.

26:33 Anything else I make from here on out, please clean that up.

26:35 And it, it seems to, it's kind of a super cheap, cheap overs, version, but you still get

26:39 reference counting, right?

26:40 Yeah, that's definitely an optimization that I think is worth it for some of these

26:45 immortal objects.

26:46 Why not?

26:47 Yeah.

26:47 I mean, we shouldn't be reference counting on none.

26:49 That's kind of weird.

26:49 Unless, unless it slows things down by having like some.

26:53 It does.

26:54 That's the thing that's crazy.

26:55 So over here, they're like, all right, here's the deal.

26:57 we, we shouldn't, we shouldn't have to worry about this.

27:01 And so, where was it?

27:03 Except it adds an if statement to everything, right?

27:06 Yeah.

27:07 It says the naive implementation of this makes it 6% slower, not faster.

27:13 Like, oh no.

27:14 It makes sense.

27:16 Yeah.

27:17 And we think we can make it only 2% slower.

27:19 It's going to be slower though.

27:22 Yeah.

27:22 Well, the thing is, normally you would just reference count it.

27:26 You just go none plus equals one, right?

27:29 Or plus plus, minus minus.

27:31 But here you're like, you have to have a test.

27:33 Like, if it's an immortal object, do this, else do that.

27:36 And it's just like that bit in the hot loop of the runtime is just apparently overhead,

27:41 you know?

27:41 Yeah, for everything.

27:42 So everything you reference has to check to see whether or not it's an immortal object.

27:46 before it does the reference counting.

27:48 So.

27:49 Yeah.

27:49 Maybe it has a NOAA method on it.

27:52 I don't know.

27:53 I think it probably works straight on the field though.

27:55 All right.

27:56 Much like Highlander, Alvaro says, there can only be one none.

28:01 All right.

28:04 Well.

28:04 Straight off.

28:05 Yeah.

28:06 Yeah.

28:06 This is definitely an interesting trade off.

28:07 All right.

28:08 Well, I think that's more than enough for the language writer.

28:11 But it was really cool that Alex wrote that up and NMR sent it in because that's a good insight

28:16 to what's next.

28:17 Cool.

28:19 So, it's my turn, right?

28:22 Given that.

28:22 Sorry.

28:23 So, yes, you're good.

28:24 Go, Daniel.

28:24 Cool.

28:25 So, people in the software community are blessed with many options for doing source control.

28:31 You know, you've got get SVN, Mercurial, other historical ones that maybe aren't as well used,

28:36 but optical engineers, mechanical engineers, electrical engineers, everybody else doesn't have it

28:41 nearly as good as the software community.

28:43 So, anytime I see an option for that, it definitely sticks out in my mind.

28:47 So, I don't remember how I found this, but came upon Allspice fairly recently,

28:52 which is Git for people who are doing circuits.

28:55 This is cool.

28:56 And so, it has, it looks exactly like Git.

28:59 You've got version control, you've got all the things you expect to have.

29:02 It's compatible with some of the common electrical design programs, but it really just

29:08 gives you the ability to do all these sorts of things that you take for granted

29:12 if you're in a sophisticated workflow like software, but that you wish dearly

29:16 you had for any other discipline.

29:18 So, when you put something in a source control and you diff it, what do you get?

29:22 They have...

29:24 Are you diffing graphics?

29:26 Are you diffing some sort of definition file that defines the circuit?

29:30 So, they have, you know, one of the first things they have is a diff tool

29:34 because they know that that's kind of one of the big questions, right?

29:36 Is how do you compare the schematics?

29:37 So, they have a way to do it visually and you can look at all the changes

29:42 and it looks like, you know, they're highlighting each commit to whatever change

29:45 was made on the schematic.

29:46 Oh, that's cool.

29:47 Yeah.

29:48 Oh, that is very cool.

29:50 Yeah.

29:50 Yeah.

29:51 So, one potential question would be, well, great.

29:54 You know, it's nice that you can do that on the internet, but I work at a commercial

29:57 company that doesn't want to do that.

29:58 But they do have both, what they have a, they have self-hosting and they have a

30:05 government cloud version if you're subject to things like ITAR, EAR.

30:09 So, you can, in the same sense, the Git has an enterprise option, Allspice also has an enterprise

30:14 option.

30:15 like an on-prem, self-hosted version.

30:17 Yeah.

30:17 So, you don't have to give away your secrets.

30:19 Yes.

30:20 But I have no personal experience with it, but it's very promising and exciting to see

30:26 somebody trying to come up with better ways to do engineering work besides just software.

30:31 You can even configure it to integrate with Tortoise Git, like the Windows Explorer

30:36 right-click type of Git.

30:37 Yeah.

30:38 So, exciting stuff.

30:40 hopefully somebody helps out the Mechies and the optical engineers as well one day.

30:44 Yeah, I mean, there's always large file support, but the diff is terrible,

30:49 right?

30:51 So, usually.

30:52 Yeah, you're looking at binary files or stuff that's, yeah, humans are so good

30:58 at processing images that if you have a visual comparison that that's orders of magnitude

31:01 better than trying to look at lines of your, even if it is a plain text file that you can read

31:07 through.

31:07 Yeah, definitely.

31:09 Yeah.

31:11 Here's your XML with its namespaces.

31:15 Good luck.

31:17 What?

31:18 What does this mean?

31:19 Yeah.

31:19 Well, cool.

31:20 I like it.

31:21 All right.

31:22 I do too.

31:22 Brian, you got any extras for us?

31:24 I, yeah, actually.

31:26 So, I've been busy.

31:27 I've been, kind of like this back stream of test and code episodes.

31:30 So, the most recent one is that I, put out was with Will McCoogan.

31:35 We're talking about rich and textual and textualized.

31:38 It's really fun, really fun one.

31:41 But, actually, so, since we talked last Tuesday, I've got four extra episodes that have

31:45 come out.

31:45 So, we've got teaching, including testing with web front-end stuff, which was,

31:53 it was kind of an interesting story about, like, basically, if you're college-level

31:56 students, but they're new to coding, when do you include testing?

32:01 And, and, the Carl says, right away, why not?

32:05 So, also, developer and productivity episode, I think that's, oh, yeah, and a Python,

32:12 Django, rich and testing article, so, or episode, so, lots of goodness over on testing

32:18 code.

32:19 they have a Django rich package, apparently.

32:22 Yeah, that, that, that was just for other, like, the CLI, the Django, CLI stuff,

32:29 including rich with that, which was great, but they've incorporated, incorporated the,

32:35 that into the test runner, so the, the Django test runner can do rich tracebacks,

32:40 which is pretty cool.

32:42 Perfect.

32:42 So, how do you know you got anything else you want to give a quick shout?

32:46 Sorry.

32:47 Sure.

32:47 So, you know, Adafruit's a well-known company for doing maker electronics

32:53 and, oh yeah, I don't have the links up, sorry.

32:55 But, you know, Adafruit's well-known and they do a good job of focusing at

33:00 the first five-minute experience of getting you up to speed with something on electronics.

33:03 But there are other companies that do the same thing as well, so I was going to

33:06 shout out SparkFun, Seed Studio, and then other companies like OpenMV, who has a focus on

33:12 machine vision.

33:13 And they're less geared more for the people at the entry level, so maybe if you're a

33:17 little more comfortable with certain things or a little more comfortable,

33:20 you know, exploring those based on your own, they could be good options.

33:23 Right.

33:23 More specialized, maybe, for people who are trying to actually build.

33:26 Or if you go to, yeah, if you go to Adafruit and what you want is out of stock,

33:29 you can check some other places too.

33:31 Which, unfortunately, happens a lot these days.

33:33 It's, yeah, those things come and go.

33:35 A lot of demand.

33:36 Awesome.

33:37 All right, I do have some.

33:39 Cool.

33:39 Yeah, that's right.

33:40 I do have some extra ones, but I kind of got a lot, so.

33:43 All right, let's see.

33:44 I'll go last.

33:46 All right, the first one is, I always love a good documentary on tech stuff,

33:49 and sometimes these are super cheesy, but there's a documentary called Power On,

33:53 The Story of Xbox, which is a four-hour video, which you can watch on YouTube,

33:58 which I'll link directly to the YouTube video.

34:00 And it's really good.

34:01 It's really interesting.

34:02 Whether you love or hate the Xbox, I honestly don't care that much one way or the other,

34:07 but it's just an interesting sort of view of like the last 20 years of technology

34:11 from the sort of the gaming side.

34:13 So if people are looking for something to watch and they want to spend four

34:16 hours doing it or spread it out, you know, they can check this out.

34:19 All right, speaking of videos, not that one.

34:23 This one, I took, so recently I released my Git course on sort of a pragmatic

34:29 introduction to Git and I decided I wanted to share one part of it with a broader

34:34 world.

34:34 So I released a video called The Four Reasons to Branch with Git and I put that on

34:38 YouTube and people can check that out.

34:39 So it's like an hour long video I posted this week.

34:42 And then this one comes to us from Jason Perkore saying, how cool is it to

34:48 see Python showing up like right on the front page of various places?

34:51 So there's this place called EasyPost, easypost.com, which allows you to like do labels

34:57 and track your labels and stuff.

34:58 But if you just scroll down just a little bit, it says, you know what, why don't you

35:01 just either buy labels or you can just use this Python API right here.

35:07 And it doesn't even sort of, if your developers click to reveal the secret,

35:11 you know, it's just like, no, here's your Python code for our company.

35:16 So just kind of a cool little thing for that.

35:20 Let's see.

35:21 Brian Skin pointed out that the Stack Overflow 2022 developer survey is open for accepting

35:30 comments.

35:31 comments, which is cool.

35:32 And I'm going to put this up here on the screen first.

35:35 So Brian, do you see this?

35:36 It has all of this stuff.

35:37 I can't, if I click it, it'll just go away and like, this is an image, right?

35:40 Right here.

35:40 Yeah.

35:40 Yeah.

35:41 What if I wanted that as text?

35:43 What if I wanted to somehow grab that?

35:46 So I've got this app, which I'm going to tell you all about next, called Text

35:52 Sniper.

35:52 Watch this.

35:53 So you can't quite see if I just drag over that, just like you would a screenshot.

35:57 And then let's see, I need somewhere I can paste this anywhere there.

36:03 So what I got out of that is check this out.

36:05 Oh, wow.

36:06 Isn't that cool?

36:07 Yeah.

36:07 I just control seed from like the picture on my, on my screen.

36:14 and it can do PDFs.

36:15 It can do screenshots.

36:17 Like, so for example, if you're watching a video presentation and you see a

36:21 slide, you're like, oh, I want to capture those bullet points or that grab it.

36:24 You got it.

36:25 So that is called Text Sniper, which is super neat.

36:29 All it does, it's just like the select region for screenshot.

36:34 That's great.

36:34 And then boom, what it doesn't matter what's under it.

36:37 It's just, if it's texted, OCR is it, and then you got it.

36:39 Yeah.

36:40 So often like a small restaurant will put their address or their phone number

36:44 like in an image.

36:48 Like, come on, I gotta click on that sucker.

36:50 I want to put, just drop this, paste it into maps or something.

36:54 That's right.

36:54 So I don't know.

36:56 I think for doing research, if you're like watching videos, you want to get

36:58 something out of something that's on the screen, like a slide or whatever.

37:01 This is, this is pretty awesome.

37:02 And it, it costs something like $11 once.

37:05 So it's, you know, if it's useful to you, it'll be worth it.

37:08 If not, then, you know, it's not, it's gotta be worth $11 or zero to you.

37:12 That was like a good OCR app.

37:15 Yeah.

37:16 Yeah.

37:16 And it's just the, the ease of use, right?

37:18 Not take a screenshot and go find your app.

37:20 It's just like slap, slap drop.

37:22 Okay.

37:24 so last one of my extras, Sam Lowe and Philip Guo sent over, allow Sam Lowe,

37:31 sorry.

37:31 And then sent over that.

37:33 I had them on to talk about pandas tutor and they were talking about the

37:37 challenges of running pandas tutor on the server side and letting people run

37:42 code, but it's pretty limited because you don't want them to hack the various

37:46 things.

37:46 You don't want to keep it pretty limited.

37:47 So they don't take advantage of like your compute resources.

37:50 So now they just posted a message saying pandas tutor.

37:54 If you go over here and say, visualize your code, it'll go and do all these cool

37:59 visualizations.

37:59 I know we've spoken about this before, but notice what it says here.

38:03 I can scroll a little.

38:04 It says initializing high iodide on WASM download pandas running boom.

38:11 And so all of this is running in client side Python, which is just, yeah.

38:17 So we talked about that being one of the topics of the language summit, the WASM

38:21 support.

38:22 And here you have it in action.

38:23 So I said on the show, like, hey, if you guys consider this, like, yeah,

38:26 maybe we should.

38:27 And then like, this turns out to be a great idea.

38:29 That's pretty cool.

38:32 Like the messages run code on the server.

38:34 That's slower.

38:34 We just recommend you run it here.

38:36 Nice.

38:38 All right.

38:39 That's pretty neat.

38:39 Well done, you guys.

38:41 And that's it for my extras as well.

38:43 That's a lot more real than I thought.

38:45 I guess I thought pyro died and Web assembly were a little bit further off,

38:47 but that's like, hey, there's an application right now doing that today.

38:50 Yeah.

38:51 Yeah.

38:52 The Brian, the anti-gravity high script thing you showed last week was so cool.

38:57 Yeah.

38:58 I didn't even know it was doing that before we showed it, but it's pretty neat.

39:02 Yeah.

39:02 Yeah.

39:02 Yeah.

39:02 a lot of the interactions are super.

39:05 They're getting starting to be real hour.

39:07 We're getting there, Daniel.

39:07 We're getting there.

39:08 All right.

39:10 How about a joke to wrap it up?

39:12 Definitely.

39:13 So we've all been in, well, maybe we haven't all been, we can all imagine being

39:19 in awkward situations, maybe on a weird date.

39:22 So I don't go on dates really being married for a long time, but imagine,

39:26 imagine that you had, here's a graphic of a woman who's on a date, like maybe

39:33 just woke up in the morning after the first day, first time together sort

39:38 of thing.

39:38 And the guy who's like sculpted, right?

39:41 He's like clearly like a super fake, probably good looking guy, whatever.

39:45 But he's in the shower and she's like flipping through his phone and says,

39:49 when she looks through your phone, but all she can find is fork a child and kill it.

39:54 Google search for a kill child and fork parent, kill parent with fork, kill parent

39:59 without killing child.

40:00 kill child without killing grandchild.

40:04 And she's got this face of like, oh, what's that?

40:07 Sorry.

40:07 Those are great.

40:09 Yeah, she's got this look like I thought it was going so well and he's a murderer.

40:13 I can't believe it.

40:14 No, he's just trying to figure out Linux.

40:17 Don't, don't hold it again.

40:18 Kill child without killing grandchild.

40:24 It's so bad.

40:24 Can you do that?

40:27 Well, I don't know.

40:30 I haven't searched it, but I don't want to have to explain that search if I

40:33 did.

40:33 Search it in stealth.

40:35 That's what incognito most.

40:37 This is totally benign, but if somebody sees it out of context, maybe they

40:43 won't feel that way about it.

40:44 There's what will get you on the FBI list and then software engineers.

40:50 There's like Vendrygram of that.

40:53 there's probably a small intersection there.

40:54 It's probably pretty big actually.

40:57 Yeah, it's probably pretty big.

40:59 Anyway, well, thanks everybody for a fun show again.

41:03 Yeah, you bet.

41:05 Thanks, Brian and Daniel.

41:06 It's good to have you here.

41:07 Thanks.

41:08 Thanks for coming.

41:09 Bye.