#501: A Nod to Sir-Mix-Alot

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

00:04 your earbuds.

00:05 This is episode 501, recorded October 20th, 2026.

00:10 I'm Calvin Hendryx-Parker.

00:11 And I'm Michael Kennedy.

00:12 Great to have you, Michael.

00:13 This show is sponsored through by us, our continual support through our own work.

00:18 Actually, check out Michael's great work over

00:21 at the Talk Python community.

00:23 He's got some great new courses up, some really cool stuff.

00:26 Some people who've been collaborating on the various courses that are on Talk Python.

00:31 And then if you want to support Six Feet Up, so I'm co-founder and Six Feet Up CTO,

00:35 and we do really fun projects with people who've got hard problems and need cool things solved.

00:42 So check out sixfeetup.com and then connect with us over on the socials.

00:45 You can find Michael and myself over on the various regulars, Mastodon,

00:49 Bluesky X, and LinkedIn.

00:51 And then the show itself is on Mastodon Bluesky X.

00:54 And then you can also join us for the live stream. So if you're here with us right now,

00:58 you're on the live stream.

00:59 We're at pythonbytes.fm/live.

01:02 And then if you want the handcrafted, not handcrafted,

01:04 but the bespoke show notes delivered to your inbox,

01:08 the artisanal show notes delivered to your inbox,

01:12 join our friends of the show list and we will never share your email.

01:17 But if we wanted to take a pause,

01:18 Michael, what would that look like? Maybe a little nod, a little nod,

01:22 A little nod, if you will.

01:24 Oh, yeah.

01:24 A little nod.

01:25 So let's talk about notebooks.

01:31 Now, I don't know how you work.

01:34 I could take a guess.

01:35 But I don't know how if you've got a problem and you're like debugging it.

01:38 And you're like, oh, I'm not really sure what I'm going to do here.

01:41 There's different things.

01:42 There's the time-tested print location 1.

01:47 Now I'm in location 1A.

01:49 Now I'm location, you know, the bisection method of like you just start printing out stuff all around the area you're trying to understand.

01:56 And, you know, there's a place for that.

01:58 But maybe we got better tools.

02:00 I don't know.

02:00 I would hope so.

02:01 I mean, I remember the day I discovered PDB and the magic.

02:05 Yeah, exactly.

02:05 Set a breakpoint.

02:06 Let's go in.

02:07 Yeah.

02:07 Yeah.

02:08 You could use, you know, VS Code,

02:10 PyCharm, one of them, and run it in a debugger potentially.

02:14 But there's this project called Nod, Notebooks on Demand.

02:17 So if your style of exploration is notebooks already and you're writing some code and you're like, oh,

02:23 this part is not working.

02:24 I need to kind of debug it. I need to understand it. I need to play with it.

02:28 Then you can use this nod or notebook on demand, which is super cool by Eric Rahn.

02:33 And the advantages that I see here are if I'm somehow able to explore my code with a notebook,

02:39 it's not just a debugger will let you sort of see what's there, step forward,

02:43 step backwards.

02:44 but you can actually explore different ideas.

02:46 Like, well, what if I had a function that did this?

02:49 Or what if I actually had a query that ran it differently?

02:51 Let me write that and see what happens.

02:53 So here's how this works.

02:54 What you do is you import nodpy and you get the notebook.

02:58 Then whenever you want to basically set a break point,

03:01 you just say notebook.

03:03 I don't know, break.

03:04 I might do import notebook as notebook break or something.

03:08 But anyway, you just call notebook and that will pause your code

03:11 And it fires up or attaches to a Jupyter notebook.

03:14 So you've got to install a Jupyter extension that basically lets it listen for these incoming events.

03:20 And when they do, it will break into a notebook instead of to a debugger.

03:24 And you're just in your function as a cell.

03:27 You can run your function.

03:28 You can explore it.

03:29 There's the other functions that are nearby that you can play with.

03:32 You can write other pieces, right?

03:35 Like a new cell.

03:35 You could write a new cell and then go play with that.

03:37 What do you think?

03:38 I love it.

03:38 I think folks who are used to notebooks but are a little intimidated by, say, the debugger,

03:43 this is way more of a natural feel.

03:46 It's weird because notebooks break your mind a little bit

03:49 because you can run the cells out of order.

03:51 You can jump back up and go to a previous thing

03:54 and play there and come back down and play here.

03:57 And so you can kind of lose yourself a little bit.

03:59 But this gives people that capability to express that creativity and debugging

04:05 by using various cells to play.

04:08 It's interesting, like when you're the most stressed out and things are the most not working,

04:13 being in a comfortable place makes a bigger difference than I think just looking at a calm time from the outside.

04:18 Why is this not working?

04:19 Like, oh, OK, but this thing, this thing, I'm really comfortable.

04:22 Let's go.

04:22 And now I understand, right?

04:24 Like a friend of mine had this saying about debugging and code,

04:29 about basically code complexity in general.

04:31 I'll share people like Jason said, debugging is always a little bit harder than writing code.

04:38 you know because there's all the complexities of what's going on so if you write code that's at the

04:43 limit of your capability of like understanding complexity or like the situation then you might

04:49 not be able to debug it right it's like yeah you should always leave a little bit of a buffer in

04:53 there of of that and i feel like this takes that down a notch right because it gives you

04:58 a better understanding during that harder debugging time yeah well it was like the difference between

05:03 going from PDB to, say, PyCharm for a debugger.

05:06 I remember in PDB, traversing up the call stack

05:10 and down into things just required you kind of holding a lot of that in your head. You didn't

05:14 get a good visual model of it, where if you use PyCharm, you could just kind of click through the

05:18 stack and you see the rest of the stack still sitting on the screen. There's the call stack right there.

05:21 Exactly.

05:21 You can just see it.

05:22 Yeah. And so this gives you that same comfort level and visual

05:27 anchoring when debugging. So I love the fact that they have built this for the notebook natives

05:33 I agree.

05:34 And I believe there's a couple other out there.

05:36 But one of the things that's interesting about this one is the markdown cells become comments,

05:42 cells become new lines, and the notebooks you can just throw away because they get saved into the source file if you want.

05:49 So if you run it again, it kind of recreates the exploration that you set up.

05:53 Oh, that's really nice.

05:54 It can persist from run to run.

05:57 Yeah.

05:58 So it's a pretty new project.

05:59 Don't expect it to be super polished.

06:01 But people well, check it out.

06:02 I think it's certainly worth a try.

06:04 Yeah, it looks really cool.

06:05 We should definitely check that out.

06:07 Definitely.

06:08 Over to you.

06:08 All right.

06:09 Well, are you ready for the new thing,

06:12 the hot new release of Polars?

06:14 Polars 2.0 has shipped.

06:16 And streaming engine is on by default.

06:19 There's a lot of new features.

06:21 That was the reason for the major bump in version

06:23 is because there's going to be some changes and some things you're going to need to do

06:28 as you upgrade to your new version of Polars.

06:31 But one other thing to mention here is really, really important is the performance.

06:36 I know we've mentioned in previous episodes, we've talked about Polars performance

06:39 versus things like DuckDB.

06:41 And in the blog post of the release here by Richie,

06:44 they've got some performance numbers here because SQL is a first class citizen now

06:50 in the Polars environment.

06:51 So if you are doing things like you would have done

06:53 in DuckDB, for example,

06:54 maybe you're more comfortable there because of the SQL environment,

06:58 and you love the speed of DuckDB, look at these numbers.

07:01 This is on some AWS instances here on a C7 4X large,

07:05 16 vCPUs and 32 giga RAM.

07:07 That is a large.

07:08 That is a very large.

07:09 They're running two different benchmarks here, the TPCH SF10 and the SF100.

07:15 And you can see here, Polars completes it in 3.2 seconds

07:19 compared to DuckDBDB, which is fast.

07:21 I mean, we all know compared to what we read in our,

07:24 went over in the last episode, DuckDB is at 4.44 seconds on that same benchmark.

07:30 And then they basically did the DS version of that benchmark as well and saw that we've got

07:34 Polars at 12.59 seconds where the current release of DuckDB is at 15.92.

07:40 The alpha of DuckDB 2.0 still slower at 14.3.

07:44 That's not saying don't use DuckDB but I'm just saying Polars is making a

07:47 real run for performance here with some of the new enhancements.

07:50 That streaming capability really minimizes memory usage.

07:55 And when you do use heavy memory on these various pullers examples,

07:59 you'll get the new, well,

08:01 new, basically the out of core or spill to disk is on by default.

08:05 So if you start hitting, for example,

08:08 by default, it's set to 80% of your RAM usage,

08:10 it will spill out to disk for these larger operations.

08:15 And it supports joins and group buys are not supported yet. They're planned but not shipped,

08:22 but it does support sort window functions and many expressions.

08:26 So those things will actually continue running even if you would have gotten an out of memory

08:30 or maybe the Ohm killer would have come and killed your process.

08:33 You can actually make these bigger datasets work

08:37 by leveraging that spill to disk feature of it.

08:40 What a neat idea because the examples you got on the screen

08:44 do have a lot of memory, but at the same time,

08:46 memory is expensive even before all this crazy stuff, right?

08:49 Relative to how much you can have.

08:51 Like memory on a server is more expensive than an NVMe disk that is not as fast, but still really fast.

08:59 But still pretty darn fast.

09:00 I mean, you're basically kind of going back a couple of years in memory technology, but you're gaining the ability to finish the work.

09:07 So I think that may be more important.

09:10 Yeah, exactly.

09:10 A little detail.

09:12 Oh, all of a sudden I went from 16 gigs of RAM to 500 and I don't know,

09:16 30 or whatever, right?

09:17 Yep.

09:17 Yeah, so one of the things here, again, I mentioned the streaming engine is on by default,

09:21 but that has a side effect.

09:22 So one of the things you need to consider when you're upgrading your code from 1.0

09:25 or the one version to the two version is that order of operations

09:30 or order of the data is not guaranteed in the streaming version

09:33 unless you set maintaining order to true.

09:35 So there's certain row level or row order default operations

09:39 that might want that.

09:40 Be sure you check out the upgrading guide before you switch up to 2.0.

09:45 And then there's also the new map data type called map type.

09:48 So it's now the arrow map type supported in the Polars map data type. So if you're in the arrows game,

09:56 which is actually kind of cool because arrow gives you that in-memory transportable data type that you can move between

10:04 various processes, Polars now supports the map type there, which is basically like a dictionary

10:09 that's super powered in the Polars world.

10:12 That's kind of the big, big news here.

10:15 There's last words, I don't know, coming out in months, but I think the benchmarks are the thing

10:21 speaks for itself on this release.

10:23 It is incredibly fast performance.

10:26 This is another thing I just I love about the Python community is people may claim Python slow and maybe should be using

10:33 Go or Rust. I don't know. I think you can there's lots of great work going on to make sure that

10:38 Python is still a first class citizen and maintains that performance that is needed to do great

10:43 production work with these tools.

10:45 Yeah.

10:45 If you've got a large scale application so frequently it's

10:49 There's a little part of it that is performance critical and 90% of it that's not as well.

10:55 And you can bring different tools to bear on that 10%.

10:58 Yep.

10:59 Yep.

10:59 There's an upgrade guide.

11:00 So if you go to the Polars documentation under the releases upgrade slash two.

11:05 So make sure you check that out before performing any kind of an update or point your LLM agent

11:10 at that upgrade guide and say, make it happen.

11:12 You probably get better results by doing that.

11:14 Yes.

11:15 A hundred percent you will.

11:16 Yeah.

11:16 Yeah.

11:17 I'm sure you will.

11:17 So to your performance comment, I don't think that this had anything to do with free threaded Python or any concurrency on the consumer side.

11:26 No, I don't think so.

11:27 Nope, not yet.

11:28 But it certainly looks like Holders itself internally is using multiple threads.

11:34 Otherwise, why would you get a much better result on the much larger server,

11:38 right?

11:38 Unless it's just pure memory.

11:39 So you're still in a sense, this is like...

11:41 Yeah, number cores wouldn't matter unless you're using that.

11:43 Yeah, exactly.

11:44 Exactly.

11:44 So it's like, hey, we don't have this great threading story, or do we?

11:48 We don't have this great threading story on Python, but still, here's the part where you actually care.

11:53 Yep.

11:53 You know, jamming on it anyway.

11:55 All right.

11:55 So pick the right libraries.

11:57 And that'd be good.

11:57 Speaking of performance, Michael, what have we cached more?

12:01 Would you say if you could cache a lot?

12:03 I love that name of this project.

12:06 Django cache a lot.

12:08 Yeah.

12:09 A little bit of CERM makes a lot.

12:11 A real throwback, right?

12:12 I'm here for it all day long.

12:13 I am.

12:13 So am I.

12:14 Oh my gosh, what a music genre that was back in the day.

12:18 That was fun.

12:19 So this project, 1.4 thousand stars,

12:24 called Django Cash-A-Lot.

12:26 I think we've got to just throw it out there, the requisite joke.

12:28 Three things, two, I already broke it.

12:31 There's two things that are hard in computer science.

12:33 Right.

12:34 Naming things, cache and validation and off by one errors,

12:37 right?

12:37 Yeah.

12:37 But the fact that cache and validation is like one of the first two before the joke even goes off the rails,

12:43 is it's interesting because it absolutely is not easy.

12:47 So there's different types of caching that you might want to do.

12:51 One way to cache is just I'm going to put some kind of front end

12:55 or pre-compute or something in front of a page.

12:58 So like on this page, it's got the issues,

13:02 counts, it's got the number of watchers,

13:04 the forks, all that kind of stuff.

13:06 GitHub could just say, well, we're just going to pre-render all of that stuff.

13:09 And when you ask, here are your bytes back, right?

13:12 That's probably the most effective, but it also has a lot of challenges.

13:15 For example, what if a new issue comes up while I'm sitting on this page?

13:19 Should that refresh?

13:20 Maybe.

13:20 What about my little picture up here?

13:22 If you just straight up cache the whole thing, Calvin logs in.

13:26 He doesn't want to see me, and I certainly don't want him accessing my stuff.

13:29 You know what I mean?

13:30 So there's just a lot of issues with caching that much.

13:33 So one way you can cache that's really interesting is just say,

13:36 well, what if everything I asked for the database was already,

13:40 the queries were ran, the results were given back, they were parsed, and they were turned into ORM objects,

13:46 and then those things were held on to, right?

13:48 That could be a much better option where you say, give me the details of this page, that part is cached,

13:54 give me the details of the user, that part is not cached,

13:57 and then render it, right?

13:59 So that's kind of where this thing lives.

14:01 It basically caches your ORM queries, and then here's the joke part.

14:05 Hopefully this part is not the joke, but there's a part that is challenging,

14:09 the invalidation. So this automatically invalidates them,

14:13 which I think is pretty interesting.

14:15 Notices when your tables change or something like that.

14:17 And hey, this cache is now invalid.

14:19 That's really, really nice.

14:21 Yeah.

14:22 Because otherwise you're spending all the time and

14:24 mechanics to figure that out.

14:25 Yeah. You know, sometimes you've got to subscribe to changes

14:29 if your database server gives you those kinds of things.

14:32 I don't know.

14:33 It's pretty tricky.

14:39 So the edge cases are here though there's probably still some not funness i believe it plugs into

14:44 the django um just caching built-in caching yeah stuff right for the orm because it points out hey

14:51 if you have multiple servers doing the caching i know multiple web servers and i bet servers also

14:57 would just mean web workers as in web garden anyway with a common cache server uh server

15:02 make sure they have their clock synchronized because they're gonna have to use uh each other's

15:07 numbers and so on.

15:08 But that seems like a solved problem these days.

15:10 If you don't have clock synchronization, you've got lots of other problems happening too.

15:14 Exactly.

15:15 Exactly.

15:16 SSL or TLS won't work.

15:17 Details.

15:18 Details.

15:19 Also, if you run a script that somehow changes the database outside of that,

15:24 so it needs to run manage.py.

15:26 So I imagine that what's happening is there's a last modified

15:29 on the cache server.

15:31 And every time you run a query, it goes, when was this last modified?

15:33 And if that changes, then it's just going to do a full query.

15:36 Otherwise, it'll just keep pulling.

15:37 I'll bet you that that's it.

15:38 Because otherwise, if you just use the ORM directly through this other script, it would

15:43 have changed it.

15:43 But you've got to run this manage pi command on it.

15:46 That's what I'm guessing.

15:47 You've got to be in the Django world, the Django machinery for it.

15:51 Exactly.

15:52 I've had this problem before.

15:53 If you run Redis caches in Redis command, like Redis through Django, like accessing it,

16:00 not just as a cache, but using it directly.

16:02 and you try and do something outside of that Redis,

16:04 there's some subtleties there that you've got to be really, really careful of.

16:07 So it sounds like this is similar.

16:08 Yeah, yeah.

16:09 It's using Django ORM technology, not the database technology or the cache technology,

16:16 right?

16:17 Like I said, like subscribing to changes to a table or whatever.

16:20 So one of the things that Warren's about is if you have a high rate of database modifications,

16:24 then that'll keep invalidating your cache.

16:26 So maybe it won't help that much.

16:29 The win, maybe a lose.

16:31 Exactly.

16:32 You're just adding complexity, but maybe every read effectively is going to just run anyway.

16:38 So there's a couple of limits.

16:39 People can check them out, different setups and so on.

16:42 But I think it's a cool project.

16:43 So yeah.

16:45 If you're looking to kind of gain a performance gain real quick there, that's an easy one to try and see if it fits your use case.

16:53 Caching is such a bit of magic tixie dust that you can sprinkle on.

16:57 But the challenge is it's the complexity and the invalidation.

17:00 Yeah.

17:01 Yeah, because you don't want to serve the stale data like you mentioned at the beginning.

17:04 Yeah.

17:05 That's interesting.

17:06 Performance is tricky.

17:07 Cash it.

17:08 Yeah.

17:08 Yeah, performance is tricky.

17:10 And speaking of performance, some people may think asynchronous is the answer to all things in performance.

17:17 But Carlton Gibson actually had a good bit of work done at the Django on the Med sprint.

17:23 So if you're not familiar with code sprints, people get together from time to time and will work on the open source projects.

17:29 And one of the outcomes from the latest sprint was a lot of work toward making async

17:34 with transaction atomic come to Django, hopefully in 6.2,

17:38 which would be pretty incredible because that's the next release of Django.

17:41 And this is one of the pieces that's been missing

17:44 for a while in the Django ecosystem is the ability to use the ORM in an asynchronous fashion.

17:49 Because currently ORM, you have to still wrap it

17:52 with async to async version of that.

17:55 And so he's got some example code here showing basically what this is all about.

18:00 But this is also a very, very tricky problem.

18:03 Asynchronous operations and concurrency, where multiple things can happen in any order,

18:08 don't combine well typically with database transactions,

18:10 which are necessarily well ordered, as he mentions here in the blog post.

18:15 There's a PR open as of, I think, yesterday and today

18:19 with no approvals and no milestones yet.

18:21 So if you're excited about transactions and the Django ecosystem and ORMs,

18:26 you may want to go check out the current work that Carlton did based on this sprint.

18:30 He made quite a bit of forward progress.

18:33 And again, I think that this is not an easy problem

18:36 or else it would have been solved long ago.

18:38 We've had asynchronous views in Django now for years and years.

18:42 And I think that, like I mentioned here,

18:44 Django has required using the ORM's sync API for transactions specifically.

18:49 If you aren't using a transaction, if you didn't wrap it in a begin and a commit,

18:52 which is what you kind of put a wrap a transaction into

18:54 because you're running multiple SQL commands at one time

18:58 or multiple ORM operations at one time, you have to use this sync to async wrapper

19:03 to make that all go correctly.

19:05 But now if we get this async with transaction atomic,

19:09 Django, it could actually make this work.

19:11 I just think that he puts a little ad in here for the Django on the Med sprint,

19:16 which I highly recommend people going to sprints.

19:17 It's a really great way to get involved in the community.

19:19 But if you wanna be able to use this version of the code,

19:23 There's again a PR.

19:24 You can check out the thread.

19:26 You can see how it works for your specific workloads.

19:28 I think a lot of this is going to be workload specific

19:31 and you're heading into kind of extreme performance area here

19:35 when you're talking about asynchronous operations on transactions.

19:39 This could be a big win for your specific workloads.

19:42 It could also be a big headache if it doesn't operate as expected

19:45 or you're not paying close enough attention to the order of operations

19:48 for some of these things, which it goes over with in the post.

19:51 It's a really good post.

19:52 They had a fun time there on the Mediterranean.

19:55 Not terrible going for bicycle rides along the sea.

19:57 That looks all right.

19:58 They got quite a bit of work done.

20:00 He actually had, it looks like he collaborated with Caleb Haddington,

20:03 who wrote a book on that.

20:05 So you actually click over here, the Using Async in Python.

20:09 So someone who knew quite a bit about the async story in Python.

20:13 And so they spent literally three days working through this and ended up with a pull request

20:19 against the ASGIE ref and on the Django bits to make this all go.

20:23 That's pretty optimistic.

20:24 It seems like the right brains got together at the right time,

20:26 which is again the reason why we do things like sprints where people get together in person in the same place,

20:32 and usually they're in pretty nice places,

20:33 to work on awesome code like this.

20:36 So you can see here, again, pretty extensive,

20:39 a couple poll requests for people to go review,

20:42 and lots of thanks, and lots of great things

20:44 happened at Django on the Met.

20:45 So I don't know, how do you feel about asynchronous transactions

20:48 to a database?

20:49 I feel okay with it, I think.

20:51 I think I feel okay with it.

20:53 I had Calvin, I had Carlton, sorry, Carlton. I had Carlton on.

20:58 He and I look the same.

21:00 Yeah, I know.

21:01 Well, it's the C.

21:03 So I had him on to talk about updates to Django's async story.

21:08 And it sounds like Django's async story is really strong these days.

21:11 It's just down to a couple of little hard problems to solve,

21:16 like transactions.

21:17 I'm a big fan of async.

21:19 I think it unlocks a bunch of different deployment stories and operational stories that are really powerful.

21:26 It doesn't always solve a ton of problems, but it can.

21:30 Yeah, well, I feel like this keeps Django more than relevant in today's day and age of things like FastAPI and everyone basically touting full asynchronous stories.

21:42 Django is still a very, very valid option when it comes to a web framework.

21:45 And Frank Wiles and I did a presentation at this year's DjangoCon comparing FastAPI with Django and Ninja,

21:52 which Ninja is an API layer on top of Django that does async really, really well.

21:56 and the performance is obviously FastAPI still faster,

21:59 but the performance is really, really good for Django

22:01 if you're using things like Ninja because it leverages this asynchronous capabilities

22:06 that folks like Carlton have worked really, really hard

22:09 to make sure that Django has a good story for.

22:11 Yeah.

22:11 You were talking about Django Ninja, right?

22:13 Yeah, yeah, Django Ninja.

22:14 Yeah, which is super neat.

22:15 Yeah, the analogous to FastAPI.

22:18 So I think it's really powerful.

22:21 One thing I would throw out there though, this is not even necessarily related to async,

22:26 but it could be in the mix here.

22:28 So if you're worried about transactions and performance,

22:31 and this is a big caveat, but if it's true,

22:34 and you're using ORMs, you could use a design pattern called optimistic concurrency,

22:39 which itself is already a pretty interesting thing.

22:42 Like ORMs don't necessarily always work that well with transactions.

22:47 Right.

22:47 Because I'm going to do a query, I'm going to get a user object.

22:49 I'll return that user object to some other part of the code.

22:52 If it makes a change to it, it needs to save it.

22:54 Like you don't want to let the entire request run within a single transaction,

22:58 right?

22:58 These need to be like, I'm doing the data stuff and then I'm done.

23:01 But if you say like, well, we're going to enforce concurrency by just like the beginning

23:04 of a web request and the end of the web request becomes a transaction.

23:07 You're just saying I'm having a single threaded process end to end, which is not great.

23:11 So optimistic concurrency puts a last modified hash on the result.

23:16 And when you try to save it, it's going to say update the username to name where the modified

23:22 hash is equal to the one that I have.

23:23 And if that changes, like, oh, it was changed behind you,

23:26 you know, you could catch that, requery it, make the, you know what I mean?

23:29 Like that can get you some better concurrency and some better safety if you're using an ORM, right?

23:34 Like that's a pretty big story.

23:36 But I don't know.

23:36 There's a lot going on here.

23:37 I don't know how I feel about ASync transactions because I try to avoid using too many transactions already.

23:44 Usually you don't.

23:45 But, you know, if you're running a bank,

23:46 you want them.

23:47 Right.

23:47 things can get out of order and, and then you could end up with a strange right history based on performance.

23:53 If some slow query opens the first transaction and there's a fast one that

23:56 comes behind it, making a right,

23:59 you've got to make sure you handle that.

24:00 So it looks like we're working on it.

24:02 Yeah.

24:03 Yeah.

24:03 And you got to start talking like transaction, uh,

24:06 concurrency levels within trans, like read committed,

24:08 read uncommitted, all the different things you can set on a database.

24:11 It gets, all right.

24:12 I always, that part's still hard.

24:14 That part's still hard.

24:14 I can't do it.

24:15 We just got to talk.

24:16 Let's just talk about something.

24:16 Let's go talk about our extras.

24:18 How about some extras, Michael?

24:19 What do you got?

24:19 Well, let's see here.

24:20 I already got them all queued up.

24:22 So first extra, this comes to us from Christian Klaus,

24:26 as you can see here, right now here, at Mike C. Kennedy,

24:29 FYI update on Calvin's first topic, free threaded wheels for polars.

24:36 Oh, wow.

24:37 Very quick.

24:37 So we've already got, well, there's a very interesting story and people out there may

24:43 know a little better than I do.

24:44 So I only have like very sparse GitHub comments to go on.

24:47 So please give me some leeway to get it right here.

24:50 So this is an open issue to add free-threaded wheels,

24:53 like add the A-B-I-T wheels to PyPI.

24:57 So it says currently Polars builds on the free-threaded Python build.

25:01 So you can use it, but there's no binary wheels.

25:04 It's going to be like a slow install build or rust and so on.

25:09 But to be clear, you can use Polars 2 on free-threaded Python.

25:13 Just the installing of it is a bit tricky.

25:16 So this issue is like, can we make that better?

25:19 It's open.

25:19 That's the part that's interesting, that it's not closed.

25:22 But if you go down here somewhere, if I look for PyPI,

25:27 it says, however, before we can do it, we still have to wait on getting more space from PyPI

25:32 because PyPI has said that you're using too much space

25:36 and we're not going to allow you to upload anything anymore.

25:39 That's a new one.

25:40 So you can see the request here.

25:41 If you jump over to it and you wait for GitHub, because GitHub has become slow the last few years,

25:46 not on a red sentence fault.

25:48 But over on PyPI, it says, project limit request, pullers runtime compatibility,

25:53 50 gigabytes, please.

25:56 And it's asking for, this is by Richie Vink,

25:59 the creator, as you mentioned before.

26:02 And it said, help, we can't upload anymore.

26:04 So this is an interesting problem because...

26:07 Yeah, obviously they're trying to avoid abuse of PyPI.

26:10 Right. And just trying to pay for it. You know, they're asking for 50 gigs. That's a lot of space.

26:16 At the same time, if you have a popular project that has a big binary built component like the rest part of this,

26:23 and you want to support all the different things, right?

26:26 I want to support free threaded wheels on ARM, you know, like all the different things that you might build for that starts to explode out into a bunch of combinatorial options that take a lot of space.

26:36 right so i'm sympathetic to it somebody it's just i don't know i can't remember where it was um so

26:41 it's like do do we really need pi pi yes you really need it to be or do do we really no do

26:47 we really need old releases on pi pi i'm like yes yes yes you do please anyway thank you christian

26:53 klaus for sending this in this is um it's about polars it ties into your items but it's also just

26:59 an interesting difficulty of the day you know there's still things going on behind the scenes

27:04 absolutely okay two more things i want to cover real quick let's do this one next so i've i started

27:11 working on these write-ups that are basically telling like guides visual guides to talk python

27:17 episodes that tell this the technical story of what the episode was about so for example we've got

27:24 let's see the probably the funnest one is this eve because it's a game so it's got maybe 10 to 15

27:31 graphics.

27:31 Got a nice little reader view.

27:32 You can do like full screen or you can, you know,

27:36 a little, it's like a little Instapaper-ish because it's,

27:39 it's a decent amount of reading. So it goes through and it says,

27:41 you know, okay, what is this? It talks about like the problem with nice pictures,

27:46 tells the story of, of what all the, what's going on. And just,

27:49 it has this like visual storytelling component of each,

27:53 um, of each episode.

27:55 And so I've been working on this. I don't have that many

27:57 out.

27:57 I think I've got six out, but I just, I was fascinated with like,

28:01 could I make a visual storytelling thing?

28:04 That's just not, it's not just narration,

28:06 right? It's, it's, let's look at the

28:07 different things discussed in the episode and kind of make a technical article out of them.

28:12 Yeah.

28:12 Well, and people learn differently.

28:14 Some people want to hear the podcast.

28:15 Some people want to read the blog posts and some people would love this visual representation,

28:19 which is, I like how it themed or how you theme the graphics,

28:23 you know, in line with Eve, you know, as opposed to. Yeah.

28:25 Thanks. There's a bunch of really cool graphics in here.

28:29 It's very sci-fi looking.

28:31 Yeah.

28:31 And each one's a little bit of a different theme.

28:34 I'll go to the one that we just did together, the state of the art in Python.

28:40 And it's got some really interesting pictures.

28:43 Check it out. If you just go to Talk Python, hover over episodes,

28:47 it drops into the visual guides and you can go and see all of them.

28:50 Like I said, there's a handful of them.

28:51 It's not just a podcast anymore.

28:53 No, it's so much more.

28:55 So much more.

28:57 And you've got so much more.

28:59 Then, remember how I lamented?

29:01 And I wish I knew this.

29:02 Two people sent this in.

29:04 Nick Moua.

29:04 Hey, Nick.

29:05 And Raphael Bosen and said, Michael,

29:07 you kind of went off on PEP 824 about like, why is it just so focused?

29:12 Couldn't it have this other thing that you wanted?

29:14 And you're like, no, I don't really want it.

29:16 What are you talking about?

29:16 You're insane.

29:17 Well, if I had just gone back at one in PEP numbers,

29:22 8pep823 is in fact that null coalescing operator that I was looking for right like a.b question mark

29:29 dot c d is the complicated sort of deal here that if I'm trying to access something in a chain

29:37 like x.y.z.whatever if any of those is none I need to just stop and say it's none or stop and

29:43 raise an exception like that that that is not none type has no attribute whatever this is what that's

29:50 about. So that's pretty interesting.

29:51 And I think I'm here for it.

29:53 Nice.

29:53 Yeah.

29:54 Last thing. And this one is, it's a big one.

29:56 This is really fun.

29:58 This is really, yeah, this is a big one.

29:59 So I'm launching Talk Python membership,

30:03 and I'm hoping that we can create a really nice community over

30:06 there that has a couple of aspects.

30:08 So I've been thinking for a long time, like,

30:10 couldn't I offer something like a little bit more for people,

30:13 listeners of the podcast and this podcast too, that's one thing I could offer that seems super obvious, like, hey, you get ad free episodes,

30:19 or you get episodes ahead of time.

30:21 It's like, well, you know, that's great,

30:23 but that's not that great of it.

30:24 Like that's not a huge benefit.

30:25 So sure, Talk Python membership does come with ads,

30:28 but I wanted to see what else I could offer.

30:30 So it comes with a nice Discord community.

30:33 A bunch of the Talk Python guests are already in the community.

30:36 So you can come hang out with many of the people that you've heard on the show.

30:40 Yeah.

30:40 Calvin, for example.

30:41 It's been one of the reasons I've now keeping Discord open

30:44 more regularly.

30:46 That's awesome.

30:46 There's enough gravitational pull because of the work you're doing there

30:49 I've keep it open now.

30:50 Thank you.

30:51 That's awesome.

30:51 Yeah, I do too.

30:52 And it's really, really, it's a neat area to be in.

30:54 And if you're not a gamer, just Discord is easier.

30:57 It's easier than Slack.

30:58 It's not just for gamers.

30:59 It's actually just an easier Slack.

31:01 I agree.

31:02 And one of the main things I wanted to focus on is live events.

31:06 So it comes with a meetup.

31:09 If you would like to go to a Python meetup or a regional conference or something like that,

31:13 but your area doesn't have one, you can't make it, or you just want to go to more,

31:17 You could go and I'm running a monthly tech session where it's like a conference style presentation,

31:24 but then sort of a mix and mingle, just hang out on Discord video channels with whoever gave the presentation and everyone else,

31:31 me included.

31:31 So all those things.

31:33 And I'm also going to add a preview, like a summary type of thing to the episode.

31:39 So a separate sidecar podcast feed.

31:42 Like here's an hour and 20 minute podcast episode, but here's a four minute.

31:46 What's it about?

31:47 kind of the cliff notes.

31:48 And if you find that interesting, then you can go listen to the full one,

31:51 right?

31:51 Kind of let you, you know, protect your time a little bit,

31:53 but keep up on the ones that, I don't know, you don't have the time to listen.

31:56 I love it.

31:57 I'm here for it.

31:57 I'm all about community.

31:58 And so this is a great, great move.

32:00 I'm glad you're doing this.

32:01 Thanks.

32:02 Thank you so much.

32:02 So folks, support all the work.

32:05 Talkpython.fm.

32:06 Click on membership right at the top.

32:08 Check it out.

32:08 Be part of it.

32:09 Hope you're there.

32:09 Cool.

32:10 Well, thank you for all that.

32:11 That's an amazing extra to end on there for you.

32:13 I've got a couple extras that we want

32:17 is the fact that Jazz Band is shutting down.

32:19 That was announced back in March, but there was a renewed discussion a week ago here on Reddit.

32:25 And one of the things that came out of that is,

32:26 by the way, there's Django Commons

32:28 for things like what Jazz Band used to do.

32:30 So if you wanted to come out and have a place where you can find

32:34 kind of more of the Commons supported projects for Django,

32:37 make sure you check out django-commons.org.

32:40 You can join the Commons, you can see the projects,

32:42 you can see here, you know, pretty heavy hitter projects,

32:45 Django Valkey, so support for Valkey, things like Tailwind CLI,

32:50 DRF stuff.

32:52 One of my favorites, Django FSM 2

32:55 is now part of the Commons projects.

32:57 So tons more that are joining in there, all one home,

33:00 which means they have a nice, safe place to live.

33:02 So check out the Django Commons.

33:04 The other thing, this is not necessarily Python related,

33:07 but boy, is it an awesome, fun little tool I found.

33:10 Since I'm doing more and more of my agentic work

33:13 on remote machines and a lot of the output maybe markdown files,

33:17 I would love to have a very rich way to view markdown in the terminal.

33:22 So I went and searched for a TUI slash CLI markdown navigator and I found TreeMD.

33:27 I believe it is in Rust.

33:29 Rust? Yeah, 99.2% Rust.

33:32 But it is a really nice markdown viewer, which gives you obviously mouse usage. A lot of the TUI apps,

33:39 the big benefit is you can use the mouse in your terminal.

33:42 It gives you an outline.

33:43 It gives you obviously a rich rendering.

33:46 If your terminal supports images,

33:48 it will see inline images inside the markdown files.

33:51 Wow.

33:52 Yeah, that's cool.

33:52 See by their screenshot doesn't show that off because that's actually a really fancy

33:56 feature of this.

33:57 You can jump in the outline.

33:59 I'll make that bigger maybe.

34:00 Yeah, there you go. You can jump in the outline to various sections.

34:03 You can also pre-filter a markdown file from the

34:06 command line to say, show me all the Rust snippets or Python code snippets in there.

34:11 And then it'll basically filter the document down to just those snippets and give you a navigator of the various

34:16 snippets. Cause you're like, I'm looking for a code sample in this markdown.

34:19 Just show me only those pieces.

34:21 And you can do that with a tree MD. So I know again, it's not a, not a Python tool,

34:24 but boy, is it a part of my new productivity toolkit of great things that I have installed

34:30 to basically remote work remotely with my agents and the outputs that are coming from them.

34:36 So another cool thing that I'll mention, remote documents.

34:39 You can type treeMD and like a GitHub repo reference to owner slash repo,

34:44 and it will get to open the readme immediately.

34:47 Or you can just put in any URL to a raw markdown file on the web,

34:51 and it's now open in your local viewer in your command line.

34:54 So you can be remote.

34:55 You know, I've been doubling down on my usage of herdr

34:59 and my framework desktop that's over here so that I can run my Pi sessions on my CloudCo sessions.

35:03 And now I can view all my markdown files.

35:06 so that was a cool one i think it's a great a great find it looks really nice so two things first one

35:11 thing i've learned there are so many projects by running this show there are so many projects out

35:16 there that are a UI focused sort of thing and they sometimes don't even have one screenshot what are

35:20 you doing you're literally the thing you're about this you are about pictures why do you have no

35:27 picture right there's only words here and it's got themes too oh yeah i love it like nine different

35:32 themes.

35:34 Yeah.

35:34 And the other, I think, is our mission is to bring tools to Python developers,

35:38 not necessarily.

35:39 I don't think they don't have to be Python tools.

35:42 It doesn't have to be written in, compiled in or whatever,

35:45 Python, in order for it to be something awesome that we can use. So

35:48 I think it's great.

35:49 Cool.

35:50 All right, Michael.

35:51 Would you say that it looks good for you?

35:53 It does look good. Oh, no, it started playing.

35:57 All right, so the joke this week comes to us from this guy named Calvin sent this in to me.

36:04 No, Calvin shared this with me earlier, and it's, boy, this thing captures the zeitgeist so well, doesn't it?

36:10 It's called Looks Good to Me, LGTM,

36:13 Claude Opus 5.5.

36:16 It starts on a Friday.

36:18 It's this kinetic typography, a lot of moving words sort of interesting.

36:23 Really well done, I think, on the graphics.

36:25 And it talks about basically the story of the manager who gets super psyched about AI and just makes everyone use it.

36:34 Yeah.

36:34 So go give it a listen.

36:35 It's a fun little music video.

36:37 I believe it was generated by AI itself, too,

36:40 which is kind of fun.

36:40 Oh, it's got to be.

36:41 It's got to be.

36:42 Very meta.

36:44 You've got to be careful because once you go down this path, I believe it will send you down a path of other Opus 5 videos.

36:51 I'll set the stage.

36:52 So there's a Slack.

36:53 I mean, it's not actually Slack, but you know it's Slack.

36:55 and Brad, VP of Velocity, Vice President of Velocity.

37:00 Hey team, quick one.

37:01 Before the weekend, leadership is super excited about AI.

37:04 Off it goes.

37:05 It's like, ooh.

37:07 Give it a listen.

37:09 You'll definitely laugh.

37:10 It's very fun.

37:11 You'll get a giggle.

37:12 Yeah, you will.

37:13 You know what?

37:13 I think this is a perfect bookend as well, Calvin.

37:16 We started with Sir Mix-a-Lot and we end with LGTM.

37:22 Let's go.

37:22 Looks good to me.

37:24 All right, Michael.

37:25 This episode looks good to me.

37:27 Let's give it a wrap.

37:28 Let's give it a wrap.

37:29 Oh, my gosh.

37:30 All right.

37:30 Bye, all.

37:31 See y'all later.