Sci Cloj Meetings, 7 Mar 2020

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Reproducible Notebooks with Guix-Jupyter - Ludovic Courtès - Scicloj meeting 10

Description

Ludovic Courtès discuses the Guix toolbox for Linux package management and Guix-Jupyter project for reproducible data science notebooks.

Gitlab project:
https://gitlab.inria.fr/guix-hpc/guix-kernel

Slides:
https://github.com/scicloj/scicloj/blob/master/resources/slides/scicloj-guix-jupyter.pdf

Recap:
https://scicloj.github.io/posts/2020-03-07-guix-jupyter/

A

Hello, everybody. This is a psychologic number 10. Psychology is a group of people working on open source solutions for data science, enclosure and actually everything about data enclosure. And today we have a special guest, Ludovic Cortez, who will be speaking about geeks Jupiter I'm, not sure how to introduce Ludovic, because he has been involved in so much so many projects, both in free software and in academic research and later Ludovic, will present himself more accurately.

A

But for now the reason we were meeting Ludovic is his involvement in developing no guile the scheme dialect and Agnew geeks the toolbox for deployment and amendment in Linux and geeks Jupiter. The people do civil notebooks that we will be hearing about today and I think if it is OK a lot of it. Would you like to present yourself a little more sure.

B

Yeah, thank you. Ok. Well, thank you for the introduction daniela. Everyone can hear me. Fine, yeah, yes, okay! So well, thanks a lot for having me today in this web meeting. I'm really pleased to be able to talk with all of you, people I'm, not a closure person that is not yet I'm, very interested in closure for lots of reasons. I'm a I'm, a schemer and there I'm pretty sure I'm aware of many lessons.

B

We can learn in scheme from closure and many good things that I've seen and sometimes I may be jealous of these things actually, but and I'm really happy to have the opportunity to talk about reproducibility and reproducible notebooks to Peter notebooks, because I think it's it's Hot Topic. So you wanted me to introduce myself I guess which you did pretty well I used to work on guide, while I, sometimes work on guide, I'm, even official eco maintainer of guy, but I'm, not so active.

B

These days, I spent most of my hacking time on gigs and all things geeks, which we'll talk about in a minute I've been doing that for a while now and over the last couple of years, I've been working on gigs in the context of reproducible science, specifically on high performance computing, so I work at inria, which is a French Research Institute, and there are quite a few people here, working on HPC high-performance computing and they have software deployment issues and they have reproducibility issues. So people are, you know, becoming aware of this problem.

B

They want to have tools to allow them to deploy the software more quickly on clusters and such and they want to be able to run their experiments. You know a month later so they're starting to get more and more concerned about reproducibility, which was I guess not so much the case until recently.

B

So that's it for the introduction. How should we proceed from now on Daniel? Let me know yeah.

A

Thank you so I guess it could be nice if we go round about around everybody who may wish to say anything about themselves, because I think several people who are relevant to the story, and so if that is okay, then could you please probably present yourself if you wish and tell just like a few words about who you are and why you may be interested in this talk so who've already an.

C

C

Yeah so I'm Martin I work on the next journal and yeah we're we're very interested in like providing better reproducibility for people and I think this is like the whole reproducibility problem is like it's. It's a lot being talked about in science, but it's actually like a problem that, like every computer user, constantly experiences when people just can't get software to work as it should, or to install and yeah I feel like that.

C

We've sort of reached like local maximum, with like old solutions to this problem, or even the newer solutions like like docker and yeah I'm, pretty excited about like these principled approaches that geeks and makes like applying these functional principles to to dependency management can bring like in terms of a much better user experience and caching and performance down the road and yeah. So pretty excited to to learn more about get stupider. Today,.

A

Wonderful Thank, You, Martin and who else may wish to present themselves and how they may utilize this toy, maybe I can campaign.

D

And chapo I have a background in computer science. I work as a consultant I have my own company and I work, mostly in pharma and regulated industries, I'm also quite interested in reproduce ability when you work with people with solutions where people's health is on the line, you need to deliver so-called systems and it's a big and very costly thing to understand exactly what systems to and make sure they continue to do so so I'm quite interested in that subject.

D

Overall, I'm also the maintain of Claude Jupiter, which is the one of the kernels for foreclosure and Jupiter. So I'm super interested to learn more about each.

A

Thank you so much class and anybody else wish to present themselves.

A

Okay, thank you. So everybody are very modest and so I guess we could begin. And so, if that is okay, then Ludovic will give a talk about geeks to Peter and then afterwards we will have a discussion. So a Ludovic Yatta all.

B

Right so yeah the way I thought about structuring this meeting is by starting with a with some sort of a talk. So I'll explain the the rationale and everything about geeks Jupiter and then also make some like demos. If everything goes well and then we can make that interactive. So if you have any question, you can also interrupt me at any time and I'll try to answer your questions that work for you all right. So this is. This is the first time for me. I mean this.

B

This whole web meeting format so I did my best I hope everything is gonna work as expected. Let's see so I will share my screen.

C

B

Let me see need to click on the right things. This one.

B

Okay, so you should be seeing slides instead of my face now, all right. So, let's see so how did I start looking at Jupiter, so actually I'm I'm, not really a Jupiter user, at least not initially I'm, starting to use it to be honest, I'm very much interested in Jupiter, because with the list background, as you people probably know it, I mean Jupiter is something we're quite used to under a different name.

B

We call it a rep all right, but that's that's pretty much the kind of tool that really Spurs like to use and so I'm I've been like I said before I've been following. You know all the work that's being done in terms of reproducible, science lately and I've heard more about Jupiter in the context of reproducible science. So, for example, there there are quite a few. There are lots of tutorials on the web on how to do reproducible science. There are MOOCs.

B

You know online courses, lots of things mentioned Jupiter as as a way to to work towards reproducible, science and I. Think that's very good because we used to have those. You know those static papers which were PDFs with figures and everything in there.

B

People start understand that as scientists as scientists, we need to be able to actually reproduce what's in the paper, and so we need not just the not just the narrative, but also the code that led to those conclusions, and so Jupiter is a great progress in that direction. So I'm very, very much enthusiastic about this particular use of Jupiter and again things like next on all go in the same direction and I think that's very good for reproducible science, but then, like I, said I've been working on on gifts for Ryan Giggs is about.

B

Let me see still juggling with windows. Kicks is about reproducibility right, it's about being able to reproduce computational environment to software environments. I started wondering to what extent is Jupiter really about reproducibility right. So, of course you can have cold, you can run it and you can produce figures and all that and that's great, but how to what extent that is, that code actually reproducible. So, for example, when I see a Jupiter cell like that one I hope you can read it.

B

So this is a biotin cell in Jupiter, in Jupiter notebook and when I see that you know with my Giggs background. My first reaction is that you know.

C

B

That really reproducible am I really going to get the same result as you on your computer. If I run it on my machine like, for example, the first line of that cell is metal tube right.

B

So the assumption of that cell is that I have not only Python installed the right version of Python, so probably Python, 3 I, guess. I also need to have matplotlib installed and again the right version of MATLAB, so that I can actually use the use this API.

B

Otherwise, it's not going to work or I'm gonna get a runtime error or I'm gonna get different results. You know that kind of thing to me. With my geeks background: it's it's really! It's really a problem. You know you want to be able to make sure that you can actually run that code and get the result that the author intended to produce, and so that's the problem I and my colleague P Antoine, who be set out to address through geeks.

B

So again we we've been working on geeks, so you know when you have a nail, you see hammers everywhere or something like that, and so we wanted to see how we could take advantage of geeks to introduce reproducible software deployments to a tool like Jupiter and that's what we've been so this this example before was just a simple one, but turns out that when I started researching this, this issue I realized that people would actually do things like when people install directly in the Jupiter notebook so that they could sort of make sure to some extent, that they would get the right software environment for their notebook, which is indeed a bit scary, because you're in the running basically arbitrary code on your machine.

B

As you execute the notebook and you well, you actually download the arbitrary code from the internet. That's that doesn't sound great. It doesn't sound very much reproducible either because you don't really know what what you're going to install, because that depends on the time at which you ran that PP install would command, for example, so that that sounds problematic.

D

Sorry to all the users, we are well aware of the problems of mutation, so yeah I think most of us are always be there.

B

Right, most of you are already familiar with the issue. You mean right.

B

Alright, so yeah again, so people would use pip install or they would use Condor inside their notebook. Things like that and I thought. You know this is not really gonna work in the long term. Right Condor is not really good at reproducing software environments in the long term, either so that that doesn't quite work and then the the solution that many people have started using is binder.

B

So binder is both both a bunch of software packages and service, a centralized service at my binder dot org and basically you can give it a git repo with either like a setup, dot, PI file or things like that, and it will create a docker image based on your setup block by file or on your docker file, and then it will be able to bring your notebook in inside that docker image, which is pretty cool.

B

But again it's not pretty addressing the the core problem, which is that the notebook itself does not contain software deployment information.

B

So if I, if I write a notebook and I share it with you, so that you can actually read it on your laptop while you still don't know what to do to run it in in the right sort of environment, that's kind of a problem, so instead people share complete binder images roughly so that they can actually, you know, make sure they run the notebook in the right software environment right. So this is the list of instructions recognized by by binder and by Red Bull to Tucker, which is the software that underlies binder roughly so.

B

So the question we ask is what, if notebooks were self-contained, what if they were deployment to where what? If we could directly introduce software deployment information inside the notebook, so that, if I share my notebook with you, while it's self-contained, you can run it and you'll be sure you're running software in the exact same environment as myself.

B

Unless that's how we started gig, stupider, so I just say a few words about kicks.

B

Right so gigs for those not familiar with it easily is you could view it as a package manager or you could view it as the provisioning tool, for you know, container images or as a distribution. You know there are several ways to approach it, but basically it provides a bunch of command lines that allow you to create software environments or to create images of software environments. So the the geeks environment command that we see here creates on the fly, an environment that contains precisely the package that you gave on on the command line.

B

So here I'm, saying I want an environment that contains Python numpy scifi and in that environment I want to run Python 3. The pythons recommend alright.

B

So that's pretty much the kind of software environment control that we want to have for jupiter notebooks as well right, and so we thought. How could we somehow add that capability to jupiter notebooks and that's that's the whole story. The whole theme which led us to geeks Jupiter, so I wrote a blog post, a while back, which maybe some of you saw about the story behind geeks Jupiter, which is roughly what I am explaining here today.

B

So how did that work? So maybe let me do a couple of quick demos, so I need to share screen again.

B

Do we have any questions so far.

B

B

So just quickly preparing the demo.

B

And sharing screen again.

B

B

So can you see something here yeah? This is my Emacs shell, all right, I hope it's going to be readable. So what exactly is gig? So just very briefly so so we get to shared understanding of what geeks is about. So, like I said you can view it as a package manager, so you can run gigs install, let's say closure all right, and then it behaves like a regular package manager. Oh it's telling me I, don't have much space space left on my laptop, but now I've installed closure right.

B

So if I do kicks package list installed, then I see closure. All right, I also actually need to install the the JVM. If I want to be able to run closure so and it installed the ice-t package, which is basically open, JDK and from there I can do Java our.

B

Kicks profile, / share, /, Java, slash closure, not sure I, guess.

B

And I should be getting my closure ripple from here.

B

Sticking a while okay here we go.

B

And I have my ripple great. So from that point of view, it's not so different from a regular package manager, except that if we look at what I've done so I've run two geeks in sole command and now I can run geeks package list generations, and here we see two generations listed right. So the first one has only closure installed and the second one installs iced tea on top of closure, and that's why we see that plus sign at the beginning of the line here.

B

Ok, so that means I can roll back to any previous generations and that's kind of thing, but, more importantly, what's very different from mutual package manager except Nick. So I don't know if people here in this meeting are familiar with Nick's, but I should say that geeks basically implements the the functional deployment model of Nick's right. So some of the things I'm describing here obviously apply to Nick's as well or there's. Not so much I mean some of the API samples are difference, for example, but yeah. This is the benefits.

B

I'm talking about here are really about the functional software deployment model that was pioneered by Nick's, and so one thing, that's very different from regular package managers is that there is very good provenance tracking right. So if I run the geeks describe command, for example, it tells me which geeks I'm currently using so geeks itself is developed using git.

B

So here, geeks described is telling me that I have one channel what we call channels, which is the geeks. The official geeks channel is telling me precisely which commits I'm running alright. So if somebody else want to run the exact same geeks to install, then the exact same cluster and the exact same JVM, the tikkun start from that commit, and that saw the information the need basically to be able to reproduce the exact same environment bit for bit.

B

So I like to say that this provides the same. You know the same advantages of docker images containers because you have bit for bit reproducibility, but at the same time you still get the advantages of traditional package managers and in particular you have very detailed provenance tracking information. So you can always know how to reproduce a given software artifacts right, that's a significant difference, all right and then yeah again, like I, should before. If I want to create a Python, Python, 3 environment I can run its environment like this, which creates an on-the-fly environment.

B

We spied on three right and now I have Python three. You can do things like that.

B

All right so so much about geeks itself and.

B

All right, then, going back to kick stupider again. Do interrupt me if you, if you have any questions, although we can also discuss discuss afterwards.

B

Okay, so going to kicks Jupiter now.

B

Okay, so the window you are seeing here is my browser connected to to a jupiter notebook running on my machine. Okay,.

B

So let me know if I should do in a bit more I, don't know to what extent that's readable or pixelized, so in Jupiter notebook. We have this new button here which allows you to select a kernel and essentially when you're using geeks Jupiter, you have a new kernel. So normally you know you have a Python kernel. You have the closure closure, Peter kernel, for example, you have the the our kernel and so on and so forth and Geeks jupiter is actually also a kernel right.

B

So you can say new notebook with the geeks kernel and this is it to get a new notebook with the geeks kernel.

B

So as many of you I guess, no jubilee notebook is associated with one language. So when you create a new notebook, it's normally associated with one specific kernel, so one language, but when you select the geeks kernel instead, you get some sort of a meta kernel which will allow you to first choose what environment you want to use, what software environment, then, what colonel you want to use in that environment. So let me give an exam an example.

B

So this is my you know my two peter notebook cell I can introduce a special geeks, granola notation and then I can say geeks environment.

B

Let's assume I want to create a Python environment. I call it Python, but the name is does not really matter. It's only used internally in the notebook. Oh actually I should say that I wanted to use the two clothes repeater kernel and I realized. It wasn't packaged yet in geek. So I started looking at it yesterday and made some progress, but it's not there yet. So maybe maybe we need to discuss afterwards to actually have it. That would be great yeah.

D

I can just add if, if you need anything, let me know No yeah.

B

All right so anyway, for today I'm sorry, I start with the Python example so I'm creating that environment, which I call Python, but then I need to specify which packages I want to use in that environment. So we will need the the Python Carano locusts. What is it called again? I, always don't forget its I PI kernel.

B

Then we need probably let's say: matplotlib right and so from there I'm, creating that environment, so I entered the cell and then, in the background, the geeks kernel is creating the environment right and then instilling me right. I've chosen the I PI kernel package, which is quarantine geeks, so the the five one three, the version and my multiplicative as well and from there I should be able to type in Python code, I think or maybe it's to creating the environment.

B

So what the kernel needs to do in the background is to actually you know, actually ensure that these packages are available so that they are built and available locally and then to actually start the debate on kernel itself. That's coming from that I buy kernel package, alright, so now I can actually run Python code. So these two plus two is Python code. I can do things like import matplotlib right, even with compression and then so I'm, not very, very good at Python. So I cannot really show you a whole lot of code in there.

B

I know about this. For example, this is my UID, while the UID of my kernel running Python, that sort of stuff all right, so the the key takeaway here that we've created the the Python environment right at the beginning of the notebook space firing rich packages we wanted to have in that environment, and then we get precisely those packages, and then we run called with the kernel that we asked for and that's very different from the normal way of using Jupiter notebooks, because this time the notebook is pretty much self-contained or almost safe, self-contained.

D

If the environment is a throwaway thing that discarded after the session or cached locally at the machine or house, how would that work next time you open this document.

B

So the the environment is it's a thing in a way, but it's it's cached so in in gigs and actually that's inherited again from the NYX package manager. So every time you build something like you build a packaged. The result of the build is pretty much cached in a special directory, which we call the store so for geeks. That's under / news, / store and in in terms of in functional programming terms, it's basically memorization right.

B

So I guess I talk again a bit about it later, but basically we are viewing build processes functions.

B

So what that means is that concretely, the first time you create an environment with let's say by race, for you know, Python my plot live and so on and so forth, but normally the next time it's going to be pretty fast, because it's already it's already so these geeks environment, a notation in notebooks, is already an improvement, in my opinion, compared to normal notebooks, but we could still do better.

B

So I, don't know if you're familiar with that movie, it's from its from my time, maybe not from yours. Maybe people are too young to know about back from the future back to the future. Sorry, but I like to say that geeks you to traval both in space and time, so it allows you to travel in space in the sense that you can deploy the exact same software environment, on different machines like on different clusters or on your laptop or on someone else's laptop.

B

But it also allows you to travel in time, which means that you can actually say alright. I wanna recreate the software environment that was current like six months ago and that that's where it becomes interesting.

B

So that that's the second notation that we've implemented in incubator, which we called geek spin, so let me give a quick demo.

B

Need to select the right window.

B

All right so I'm back to my browser.

B

This time, I'm switching to a a notebook that already contains all the the relevant pieces of information.

B

So, for example, I have my notebook, but if, if I really want to be complete to really completely describe the software environment, that I want to have it's not sufficient to have you know that kicks environmental notation that I showed before right, because, for example, today's Python package in gigs is Python, three point something. But in a year from now the default version of Python will be I, don't know Python for maybe Oh No yeah.

B

Fight on many versions, our hot topic in Python and I hear anyway, so you know the default version. That's packaged by geeks. Obviously changes over time right so that that geeks environment, a notation is, is not sufficient right. So if I really want to fully describe my environment, I also need to say which geeks revision I'm using for that notebook and that's where the spin annotation comes in.

B

So if you remember just before, I showed you that geeks describe commands so geeks describe when you're using geeks on the command line allows you to to see exactly which revision of geeks you're using and here geeks P needs all of the of the do all of Kings described because geek spin allows you to say alright I want to use a specific revision of geeks. So here this is my experimentation. Let's assume I found out that this particular commit.

B

Id is working for me and I want to make sure that my notebook will, you always be run with geeks from that commit. Then I can enter that geek spin annotation and in the background, geeks Jupiter instantiate geeks from that comment. Alright, so then it said that it succeeded and if I follow this link, it's basically pointing to the web view of the committee I'm using which is a comet from January last year, all right.

B

So this time all the geeks annotations I'm going to have from now on in my notebook, are actually going to use this specific revision of weeks. But.

E

When you say this revision of geeks, does it mean only the package manager or also that packages themselves, as they were seen by the package manager? Alright,.

B

It's it's really both! So it's it's! It uses this specific revision of the package manager, but the package manager is what is what allows you to actually install the software packages themselves right so, for example, back in January geeks would provide I, don't know guy to point to point for maybe, and so if I asked for that revision of geeks and then I asked for the guy'll package I'm going to get guide to point to point for then it's all done see graph from January last year and it.

E

Cannot give you the guy'll, which is, for example, later after, like if it's tree and it can't it can't give you the girl three, because at that time there was no country.

B

For exact right, exactly but I could have several cells using different revisions of gigs and there's different revisions of guide. For example, okay,.

E

C

Could you also specify, like the version of Kyle, you want to use like two point, one.

B

Not directly so usually in geeks in like in many districts, there is only one version of a given package available at a given point in time, so that there are some exceptions. So, for example, for see we have several major versions available for Python. We have a Python, 2 and Python 3 currently, but usually there's only one version of a given package so for guy like there's a need, the latest stable version and that's why you actually need to go back to the previous revision of geeks.

B

If you want to use older software, that's yeah, I, guess that's different and also a very different way of approaching things compared to some other tools that try to provide. You know to to leave some degree of Liberty as to which version of a package you're using yeah they did you're approaching. Some is functional. Package management is that you declare the the whole package graph of a given package. So you probably don't want to maintain ten version of a given package all the time, because that's I mean there's a lot of QA work.

B

That would need to be done right so that that's how it works.

C

Or yeah, or do you want to? Is it okay to dig deeper into this now or do you want to continue I'm sure having.

B

C

Questions yep um yeah constant. What I see from a user perspective.

C

Could other people try to mute themselves on hearing.

C

Like users usually specify or like yeah, when you work from with geeks from the get-go, this just might be might be fine. This way, but yeah normally when you're trying to reproduce already existent worked- and you have all these declarations of like what people are using like in requirements, dot, txt files or in depth even files in enclosure, and would it be possible, like yeah, I I, get that this is like it. It depends a lot on the library like a Python library that depends system.

C

Libraries is obviously much harder to to get different versions going whereas like for for a closure dependency, that's just a jar. This might actually be pretty simple right to to build something, to bring in a different version of versions of Maidan packages and and support all declarations that there are.

C

B

That yeah I think it's Indian. It's really a matter of policy in a way like for, for the core geeks packages, I mean for maintenance reasons.

B

We want to focus on one version, but then geeks allows you to have third-party channels what we call channels where you provide additional packages, and so, for example, we have someone working with the the proof assistant who wrote a channel with many different versions of because they want to be able to test their software against all these versions, so that that's something you can always do as a separate channel and then gigs also captures the the specific revision of that channel that you're using so if you're, using gigs with several channels, then gigs described, is going to show you the revision of each channel that you're using and then you can reproduce again the exact same gigs instance with the same channels.

B

So that that's one way you could address it. If you have specific needs like where you you want to make sure you always have several versions available for a given set of packages.

C

Okay- and this version won't have to be in a package name then, like you need some kind of mapping them.

B

Right so so gigs.

C

B

Have several packages with the same name and different versions and then on the command line? You can say I want I want that package at that version, so that that's also supported in extra Peter. You can say Python at two, for example, things like that.

C

Alright and lastly, under like specifying like what people are used to their sticks, import that kind of Maps to maps from these other declarations right.

B

Yes exactly so, it's only knows about packages available as well as geeks packages. So if you have software available as a via, for example, peep for Python or that kind of stuff, then you first have to convert it so to speak to a geeks package and the geeks import tool that you mentioned allows you to do that quite easily.

B

So it's semi-automatic! Thank you.

E

So you can import libraries from different programming ecosystems. Yes,.

B

So we we're basically in geeks, trying to subsume every language specific package manager. That's something visual.

F

B

It's obviously nice and not really doable this way, but we're really trying to you know it's provided it's all set, that's consistent, and that gives enough of an incentive to to be used instead of language. Specific package managers, because you're never gonna use a single language in your whole stack right. So you need you need tools that can that can handle software in different languages.

B

That's my point of view and that's why we have this geeks import command that supports different common represent or is like open for oh camel, pipe I for Python, create forests, sheep and cron. You know all these things makes it quite easy in general to to convert existing packages to gigs.

B

There is still a bit of added additional work depending on on the tool or the report that you're using. But that's that's, yeah, that's already an improvement. That's less work to do right.

B

Okay, so let me perhaps continue.

B

I'm going to save you a few words about the implementation, I think I'm hearing myself a little bit.

B

All right so I already mentioned the implementation a little bit.

B

Maybe everyone could you make sure that the mic is muted, because I hear a little bit of an eco I? Think it's okay. Now, alright, so I talked a bit about the implementation or you probably guessed already Peter is so geeks. Jupiter is a colonel in Jupiter balance.

B

So basically, what that means is that Jupiter itself Jupiter notebook talks to Giggs Jupiter using the the kernel product call, which is JSON JSON over z1q roughly, we have Jupiter kernel implementer in this meeting, so someone was probably very familiar with it and then geeks Jupiter essentially behaves as a proxy to the actual kernel that you want to talk to. So every time notebook send the message to the geeks Jupiter kernel, then the geeks Jupiter kernel then relays that message to the right, colonel.

B

So in the example I gave before I had the cell, with the with the Python kernel and Geeks Jupiter, had studied that Python process separately and then was essentially forwarding messages from Jupiter to the Python kernel and back and forth right.

B

The same happens with all the other kernels that you might want to use.

B

Yes, so this is where I should talk about the execution environment of the kernel itself,.

B

So one thing that kicks Jupiter does is obviously to set up the execution environment of the kernel, but it also run runs the kernel in a nice lated environment. Let me give an example.

B

So because this is about reproducibility, we want to make sure that you know everything. You know that your notebook is not going to interact with random things on your file system, for example, or with the network things like that, and so that meant running the kernel in the container. So in linux terms it means using separate namespaces if you're familiar with that, like it's pretty much like change routes, chroot thing so, for example, if I, if I go back to my Titan example or maybe maybe that one has more.

B

Right that one is probably more complete.

B

Alright, so I'm recreating an environment with that I PI kernel package, though I'm I, guess I'm, not using the same revision of kick. So, maybe that's why it's taking some more time? I! Guess! Ok!

B

So now, I have my Python environment. I should be able to run Python cells, one by the way, this cell, that defines the hello function here. Had it has again the the geeks environment, notation, meaning that I'm explicitly asking geeks to take the my ipython environment that I created above so it's in this case it's redundant, but it's a facility that allows you to be explicit for each cell about the environment that you want to use. In case you have several environments in the same notebook.

B

B

I, don't know what it's doing. Maybe it's downloading something whatever. Maybe I can go back to that to that one okay, so this is also a Python environment, so the one we saw before and like I showed before. If I look at the UID of my jupiter notebook process, it's 1,000. Okay and the good thing is that because the kernel despite on kernel, is running in a container in separate PID namespaces, separate user namespaces, then you you're always going to get UID 1000 all right, regardless of your whole system.

B

So this is entirely deterministic and likewise, if I try to list what's available on on my in my container well, I have only these directories available. So /mu is the is where the store the store directory is available. And then, if I, look at slash home.

B

Then I have only the Jupiter directory and again, if I look inside that directory up home, slash Jupiter, then it's basically empty alright.

B

So what that means is that you know remaining in a container separate namespace is a different file system view and that ensures that my notebook cannot use things that happen to be on my laptop right, so it it ends it ensured the notebook is going to be self-contained and is not going to refer to local files or even to the the internet. So it cannot actually connect to a remote machine here, because it's running in a separate networking namespace, and so it basically likes network access entirely.

B

So that's pretty good in terms of reproducibility. That's that's what we want to have I think we want to make sure we we only have access to you know, what's computed and what's specified in the notebook itself, then.

C

B

Hours I was waiting for that question. Actually, so, alright, so not having access to the network is obviously a problem. Usually you'll want to download data from somewhere so that you can process it. For example, things like that. So how do we do that? Well, we need additional facilities right. We need to be able to download stuff so for that we introduced a an additional annotation, which is these geeks download, a notation that you can see here in this cell.

B

So it's pretty much like a normal downloads right, except that you have to specify the expected hash at the end right, so the sha-256 hash, and so that ensures that when you run the notebook you're going to download this file from this URL, but it has to have this hash and if, if you get a different hash, then it's an error. You cannot go any further right, so I can evaluate these cell.

B

Well, it's going to download it. I, don't know how fast the network is in this room, but yeah. It's insuring again that you're really getting what you initially intended to get here and.

C

B

C

B

Me cached, honest or right. This will also be cached in this store. So the store is all about. Caching right, it's all about memorization,.

B

So I'm guessing it's being downloaded.

B

Demo effect right.

B

Right so anyway, this is. This is a basic, where the basic approach that we have for four gigs Jupiter. So the thing here is first to have people be able to describe precisely the software environment that they need and then also make sure that they describe precisely the external resources that the notebook needs.

B

B

Processing, a girl and we'll check later I guess.

B

Okay, so this is actually, if you've ever read about Knicks or geeks, before what we're doing it is essentially an extension of the work that had been done initially for for Knicks and also implemented by by geeks. So this is a very good paper. I really recommend it by L code Ostra and his former colleagues about the inception of Knicks, essentially and the functional software deployment model and essentially we're here implementing the same model roughly, but in the context of an interactive environment which is Jupiter notebooks right, but the ideas are pretty much the same.

B

Only translated to the context of the interactive notebook, yeah I think that's that's pretty pretty cool so from the viewpoint of a functional programmer. You know this is thing, that's quite easy to understand. We want to view notebooks as pure functions, so they take input and they produce output, but there are no side effects, no undeclared inputs. So with these kicks to be either kernel with the additional annotations. The idea is that the notebook can really be viewed as a pure function, because everything all the inputs are declared in the notebook itself.

B

That's the idea, so I'm gonna wrap up for the the talk and then we can. We can go in discussion mode, I guess so. There are obviously still some open issues. So this is pretty much a proof of concept. I mean it's working, it's usable, but there are things that are. That could be improved like the the user interface. So, for example, we he saw the geek spin interface before it's great.

B

It does the job, but as a user you'd like to be able to say rights, for example, I want to use the the first commit where the Python version was number of three points, X, something like that. So we I'm sure we could improve the way you specify, for example, the geeks revision that you want to use, because you know you think that sha-1 that git commit ID is is convenient if you're already using geeks, maybe but if you're not into geeks. That's not the best user interface.

B

Another question, so we implemented it as a Jupiter kernel. The question is: should deployment be built into Jupiter itself, rather than be a separate thing that kernels or tools like binder implements? That's I, guess that's an open question their arguments in favor of keeping it separate from Jupiter in itself. It provides more flexibility, you're not tied to a specific deployment tool, which is good, but on the other hand, software deployment should really be a central concern for Jupiter notebooks.

B

If we are to use them for reproducible science, so yeah, it's you know their pros and cons. I guess another question that people ask is about interoperability, so this is. This is great. These annotations allow you to have a reproducible, notebook and that's pretty cool, but it assumes that people are running Jupiter notebook on on the machine that already runs gigs, which you know as a geeks developer. I. Think it's a great thing to do, but still that's that's.

B

That could be problematic from an interoperability viewpoint. So previously someone suggested that we could add a geeks notation to automatically generate a docker image from your notebook. So, like you would have a semicolon, geeks pack in your notebook that could generate a darker image of your notebook, something like that and I think that that could be an improvement in that regard and it's actually your low-hanging fruit, because Geeks has a geek spec command that can generate darker images or singularity images or plane doubles so again, future work, I, guess and so in a nutshell.

B

A geeks Jupiter is about making self-contained notebooks that can actually be reproducible with automatic and reproducible software deployments and also called running in isolated environment again, to make sure that it's as reproducible as it could all right. This is it for the the talk part of this meeting for myself, but I'm happy to answer questions now.

B

Questions left thank.

E

You very much for the talk yeah. Thank you.

C

Right may I go with.

B

Sure go ahead. Nothing.

C

All right thanks a lot so yeah. So if I may, like my outline like the dream, I would have about dependencies, and maybe you can speak a bit to like if, if that's ever attainable and like how it would fit within the goal of gigs so like as I said, I think it's it's quite critical like in order for people to like. Currently this requires quite a bit of buy-in from people and they need to change like yeah all of the commands and tools.

C

They know don't work anymore right and also all the description files so like in my ideal world, like people could continue to use the descriptions that that they've used in the past, like requirements, dot txt like Depp's, Eden and and yeah other descriptions, I. Think people care like more about like exact versions of libraries that they're using them like things further down the stack like GCC and stuff I. Think for for just like the distribution like yeah. What you're currently doing is is fine but yeah when we go further down and also like spot.

B

And the way we could help with the setup phase is if it's easy to generate a docker image, for example, for your whole jupiter notebook that includes geeks, like you generates, you will have a better and roughly to generate an image that includes your that includes jupiter, your notebook geeks itself, so that you can easily pass that image to someone who does not run gigs and still it's easy to use it right that that's one one approach.

B

I guess you also mentioned proprietary software, so new accessor project is obviously not about proprietary software, so it does not provide any non free software package, but you can have your own channels providing non-free software packages. That's something you can do some people do it. I would actually argue that from a reproducible science viewpoint using black boxes is not a great idea.

B

I mean you want to be able to know what software you're running, what it is that you're doing and to me that's a problem to rely on black boxes, for that, but I do understand that some people do it and yeah, because in their field it's very common to use I mean if you want to use GPUs like you were saying you, you have to use that Nvidia thing, for example. Unfortunately, I hope it's going to change, but yeah.

C

B

That's one case where and video is a situation of in the monopolistic situation, essentially so yeah.

B

What tells you submissions, Mac, OS and Windows I guess in the same vein or.

C

I mean maybe even or easier way, maybe to also like provide docker images to get started with gigs. That yeah that and also run on on these platforms would be a probably a much easier first step right right.

B

So it's actually ironic because the project does not provide any official darker image for geeks, for example, but we do have tools that make it very easy to generate docker images, so I'm sure we could. We could do better and provide an official docker image yeah, that's something we need to work on and it would.

C

Make it easier.

B

To get started, I guess, and so so indeed you could use it on Mac OS and on Windows, probably I'm, not very familiar with Windows I guess. Windows nowadays implements the Linux Cisco ABI with the windows subsystem for Linux. So it may be that you can just take the the geeks binary tar ball from the download page of the geeks website and unpack it and run it directly on Windows I, don't know, I haven't tried, but from what I understood of the windows subsystem for Linux, it's supposed to work out of the box and.

C

Lastly, like what we tried, what I think yeah an external we we try to mount the like, keep a global store around and like mount that into like via NFS or some shared file system like mounted into into every notebook and then like it's. Currently, it's quite a bit big problem like with docker images like when you want to run them and the cloud that you have to download them. First and these environments don't really compose right.

C

So, whatever the install the order you install in like totally changes your your layers and then you have to download gigabytes of stuff with just some minor changes. Right and- and this is something that's yeah- would work much nicer in with gigs like mounting the thing you don't need to download anything and things just all of it like easily of the network and everything that's once been installed does never has to be built.

C

Somebody else again, I think that's yeah that can provide down the road, quite a amazing user experience, and but maybe yeah like like also the the build server that you provide like. Would it be possible or how how many resources are there and like how long do you keep stuff around there, or is that something that yeah we're we're? You would also yeah we're more support. Could help like to to speed up the build times for for old stuff and yeah keep a bigger shared cache around.

C

Is that something where you can see making improvements to usability as well, just by adding Hardware, basically,.

B

So, regarding the the official build servers for geeks, so they're, the one we are currently using, our sponsor by a Research Institute in Germany and the MDC max delbruck Center, and they they provide quite quite a number of bid machines to us like we have around.

B

Maybe 40 build machines now which is depending on how you look at it quite a lot, and not necessarily so many but I mean that's. That's already, I think a good number of big machines for for what we have to build, and we also have tens of terabytes of storage available starting from maybe last year, I think so. That's still quite recent. That's a new year year and a few months, I guess. But that's we have quite a lot of storage already to store things from the past.

B

That's not enough to store things from you know forever. Obviously, but then yeah I mean, as suppose, I know. The next project, for example, is using Amazon s3 to store things for a long time. That's one option, but in the end also we're working on reproducibility reproducible builds, and so our goal is to make sure everything remains reproducible, even if you actually have to build it yourself right we're also working with search.

C

B

C

Sorry, I'm or yeah reproducibility is the one thing, but yeah. Also the convenience of yeah like running things quickly. Right I mean it's great, that it works eventually. Yeah yeah.

B

I mean sure in terms of user experience it has to be fast and the binaries have to be available I just so so. Currently, all I can say that we can keep binaries for a pretty long time, but it's only since earlier and a few months ago, all.

C

Right all right, unfortunately, I have to run already yeah thanks so much everybody like yeah for putting this on and yeah for taking the time. Thank you. Thank you. Bye, bye, bye,.

A

Anybody else wishes to ask anything.

D

Yeah I'm wondering if you could maybe talk a little bit about how you package foot fatigue and just you said earlier that you looked at packaging co2 pizza. That's not a problem we're solving now, but what what does it entail?.

B

All right, so what does it take to package for geeks I can show a quick example to make that more concrete.

B

Alright, so this is my Emacs window, so let me see.

B

I can look up the package definition for closure, for example, and so you lease furs. It should be pretty nice and pleasant to see because it's full of parentheses.

B

So essentially, a package definition in gigs is purely declarative and it's pretty much like you know what you would write with other packaging tools. So you declare the name version: the URL for the source, the the commit you're targeting, for example, and then so this one is not the simplest package. I should probably start with the the simple package just to get a feel of what it looks like. So this is Agnew hello package and yeah. You get all the the usual pieces of information.

B

So one piece of information here is the build system which tells you how you're going to build the software. So connubial system means that you're using the the standard configure make make install sequence and yeah.

C

B

Then there are other bit systems like there's a see. Make build system for software using see make there is a closure build system for closure things, I, guess I'm, not sure exactly what it uses under the hood, and this is all so. This is all in scheme. So at the same time, you have declarative package definitions, but it's all schemes, so you can use you know in Emacs you can use gazer, which is a mode for guile and racket and chicken, and you can use it to navigate your code, for example.

B

So all those things that really spurs are used to, which is very good, that I should yeah.

C

B

If I, this is an example of a package with dependencies, so this one is already a bit more complex, so we have here, we have patches okay and then here we have dependencies, so we have native inputs and inputs. So native inputs is about dependencies needed that build time and inputs. These dependencies needed at runtime, essentially.

B

Then a package can provide arguments tweet-tweet build system. So, for example, if you want to modify the the sequence of build phases, you can say alright I'm using the standard build phases, but after the install phase, I want to have a custom, install phase and that's the lender, where you can do arbitrary things. So in this example, we're just patching the the file name that appears in the Egret and F grep scripts. That's for the grab package.

B

Things like that.

B

Previously someone mentioned Geeks import all right, so Geeks import looks like this. For example, if I want to import the package, the type I package, that's called its gen dangerous, it actually exists. I can run gigs import by PI, it's dangerous and then I hope you can read it. It downloaded metadata from Wi-Fi and then it automatically generated that thing. That's package definition, usually for pi PI, for example, it works out of the box, like you can copy/paste that package definition and it just works.

B

Then you know, depending on the upstream repo, you may have to make additional modifications to make it work, or it really depends, but for pi PI is quite reliable, for example, and there there are quite a few importers that are supported. What's it called yeah, so there are so called importers for new packages, fornix type IC package package, helper Crown, crates and so on and so forth.

B

Does that answer your question.

D

E

No Java packages packages- oh that.

B

There are actually Java packages, so we have quite a number of Java packages. Well, let me see I can give you precise, that's yeah. What do you import them like yeah, I, think so I think we don't have a maven importer, so I suppose people wrote them by hand.

C

B

Like we have 312 Java packages, which is probably not so much actually from a Java programmer through points, but we do have Java packages. That's what I meant to say well,.

C

B

Only a handful of closure packages I checked yesterday and we don't have an importer also for closure packages. So that's definitely something we need, because doing them by hand is not fun. So it's it's great I'm, not very familiar with line in Ghana I think it's called the closure package Hollinger, but.

E

It's the most used one, but it's not the only one yeah, especially.

C

Now there is a bit.

E

Of convergence with the package- it's like is some tooling that came out which is called from from cognitive, the people from closure. It's called tools, deaths and it changes the landscape of package manager package managing in enclosure, but we still amidst the whole kind of evolution. It was a clearing at that.

D

Now, I think that for co2, laser to be usable and along with kicks, it would need to be able to import from me even include Josh.

E

Hey yeah I would love this.

A

Everything feels unstable.

E

Yeah, so it can't my.

A

God mask ass, oh yeah, you can see they will hear all the time then yeah. So I, guess by the official time of the meeting to end, is in ten minutes and for now we can chat and it might not be on the YouTube edited version that we'll upload later. But if there is anything else so that people here will do this pass until we possibly find to the week, then.

A

And yes, I I was curious about about guide to and nothing you do. Yeah. Oh.

F

Here's Laura is.

E

That I yeah yeah, ah hey.

B

Oh hello, I'm, really sorry I ran out of battery, which normally doesn't happen.

C

B

You know it's usually lasts for four hours at least but I don't know. What's happened, I'm really. Sorry.

B

So what were we saying again, I.

D

Think I was remarking that I I mean I, can't help think about because I I think using which I was closer would probably mean Klobuchar, and there are several ways you can use it today and you can.

D

You can install it using Conda and it will give you a clean, closer image with Jupiter in it, but nothing else and one model to use it is to dynamically load libraries in your session as you go, you can also build your own crow to guitar and include anything you like and deploy that, but but I think most people will use the other model, and- and so so not only would I think as I see it for it to be meaningful.

D

We would not only need a way to get culture in there, but also need somehow to be able to get access to closure, libraries and probably ideally also Java, libraries.

B

D

I'm curious to hear if anybody has other ideas, but that's where.

B

How I see it so I think what you're saying essentially that we have a packaging problem right I mean I mean at least that's that's the way. I look at it, so I think it would be great to have a bridge from Java and closure or packages to geeks. That's how we usually address that, because the the benefits done it that you can.

B

You can actually take care of all your packages with with geeks with a single tool which is quite convenient and you get the the properties that you have with geeks, which include you know, provenance tracking reproducibility. So that's pretty nice. So, in my opinion, it's nicer than mixing. You know Condor with docker. We speak that sort of thing, but obviously.

D

Completely I'm not arguing in favor that I'm just trying to state how I see what we would need to make it work in a suitable manner right, because we don't want to put Conda in the editor mechanisms.

B

Right so I mean so for most languages that have language specific package managers. We've we've written an importer like I, showed before and I think that's what we would need to do for closure as well, which is complicated by the fact that, like Daniel, was explaining that there are several tools, so lining gun, plus tool steps. That's the name of the other one right. Yes, yes, which is a.

E

Power, it yes, but I, don't see. I just realized that I, don't think you need to support any of the closure tools needs everything you need to support the memory, because this is how we work with packages in closure. Every package I mean every dependency is the job. So what you need to sell far is what jars you need and if you need to do anything for closure specifically what what you need is is support for the JVM.

E

F

E

So this is this: is what I suddenly realized, why I mean importing? What? What? What is that early reporting yeah? So if we import jars and then that's what quick gigs needs to import as well, I mean if, if I, if my real name is, is correct right.

B

So obviously for Java you can use Charles directly, you don't have to go through Nick's or even Condor, or whatever I mean you can. Yeah, like you say, can directly use jars and that's one of the big advantages of Java for for the geeks project itself, really try hard to build everything from source, which means not reusing pre-built binaries such as jars and that's I, think that's important from many different viewpoints.

B

It could be security, it could be transparency, reproducibility, that's very important, but obviously it's also challenging and especially engine in the Java world, because, like you say, people are used to sharing jars because it's so convenient. That's one of the big selling points of Java.

B

But then you know it leaves a bit to be desired because you can never really be sure where that binary jar came from what its corresponding source is. That's not ideal to me, the mail, the ministries.

E

It you know that the one one specific like one version is is is always the same thing. It cannot be modified.

B

Right you mean the jar itself. That's available in the Maven. Rifle cannot be modified. It's like yeah yeah,.

E

Because they're I mean it's an immutable store. Jars are specified by.

E

You know three components is a group ID'd as a name and there's a version together. They form a identifier, you cannot mix up and you cannot modify after the fact jar right once once they are published. That's it. That's that's so you know, you know what you're getting I mean in terms of in terms of version artifact.

A

Another detail worth mentioning here is that typically, typically closure jars contain the closure source code and they're not and not Java bytecode compilation of it, so whether we could consider it consider it as actually source code could be an interesting question.

A

On the other hand, you know taking a jar and changing the code he needs to create. Another job is not an obvious task, at least it and a tool chain of closure. So maybe we need a way to depend on actual closure project sauce, not a job. Why?

A

B

Yeah I think so so, for example, for the closure packages that we have in geeks, we actually build them from source I, don't remember the specifics, because I'm, not the the author of that part, but I- think it's important in general to be able to you know to rebuild by yourself and verify that you get the same binary as the official binary, for example. We're very much involved with the reproducible builds effort, which is that available attributes.

B

Ball builds dot, org, yeah I think it's very important security, wise in terms of transparency, also in terms of reproducible science, also, because you want to make sure you're running the code that you think you run.

B

But yes, oh I didn't know that closure journals usually embed source code as well. So that's, obviously an improvement. It makes it easier to verify the dot class files. I guess.

A

A

Now is like the official time of end of the meeting and I guess it depends on you if you wish to stay a little bit, if you can, then that would be lovely and otherwise we can take it as the moment to end the official video and then anybody who wishes to stay and keep chatting. So what do you think? Would you like to continue a little bit more or maybe you need to go to another thing, I.

B

Guess I need to go within ten minutes or something so maybe I. If there's one last thing you people want to discuss I'm happy to discuss it and in ten minutes I be going I guess why.

F

Are you good, you know yeah.

A

So anybody close yeah.

E

Minutes but I was wondering how the project is evolving because I'm following the projects from a distance and I accessed as they you know, as as someone who appreciates both guile and and and what gives to us, I I, wonder how, for example, since last time, I heard you in Nava and Japan, it was too useful or no actually more I think maybe very for you. I, don't know three years. When was this as I. Remember anyway,.

E

Okay, so so how what what is changing? How how is the projects evolving do you? Do you manage to get people on board? Do you manage to get the resources you need to you? You have a clear direction. The are you happy with how the project is being adopted. The adoption rate being honest, I mean I. Would I would love to hear you about talking about that.

B

Yeah well so there there are many many things happening every day. So for me, as one of the maintainer zits, it's a bit challenging actually to follow everything. Well, I don't have to follow everything but I'm interested in many of these tasks.

B

So we've definitely seen more people adopt it. We've actually released a one point or release last year in May, and you know it had been seven years of work. So I'm really happy. We reached that my son, and now we keep going. We still have a lot to do on our plate.

B

So, for example, over the last few years, we've been more and more involved with the with reproducible bills and also with bootstrapping ball bills and actually other other guile and new hacker starting working on non new hackers to started working on bootstrap herbal pills, which is about making sure that we build everything from source down to the compilers compilers.

B

So if we go deep down to the wound up the dependency graph being able to build that from source to some extent or another word put, it is being able to use as few binaries as possible as sources. You know, I think that's, that's again, very important work from security and transparency and user freedom viewpoints and I'm really happy people started working on it yeah. So that's that's one example. Then we've had lots of improvements in many areas for the stand alone, distribution that we called geeks system yeah.

B

So we've had people work on writing new services.

B

Writing a graphical installer. So like pretty much like that of Debian thing, we've had many improvements to the UI to the command line interface. There's a new max interface. Also, that's been evolving.

C

B

Than there are side projects like this geeks HPC thing, where we focused more specifically on reproducible science and.

B

Yeah high performance computing well, we've had also changes like channels. You know core changes, I would say well, yeah.

C

B

Of things, we get more feedback, which probably means that the option is rising, I guess.

E

B

So I guess there's they're. Mostly there are many individuals like good new Linux, savvy people, I would say, or people interested in user freedom things like that, and then there are. There are more people interested in reproducibility, so people who were familiar with other package managers, for example, and probably got tired of up their shortcomings in terms of reproducibility, and so we we get actually lots of people who are not familiar with functional programming, but just wanted people to use. You know to have those with the facilities like transactional upgrades and wallbox.

B

You know that kind of thing and.

C

It's interesting.

B

Because we, even all the people who write new package definitions, most of them are not schemers or even mispers, which I think is interesting. You know sorry, you know initially, they start like bye-bye writing simple package definitions, because you know it's purely declarative, so you don't have to be an expert in scheme to understand what's going on and then they start writing actual scheme code without really you know noticing I would say, and that.

F

B

That's the whole plan actually yeah.

E

Nice and how is the I mean in this space of and you know, you're there with Nicks and how do you fight among yourselves, I.

B

Mean Nick's versus geeks yeah, oh, we don't I, don't think we fight among ourselves. I mean we. We regularly meet actually.

B

E

B

Definitely not a fight. We have so that the two projects are I mean are similar in the sense that they both implement the functional software deployment model and that sets them apart from pretty much everything else right. But then we have different approaches to many things like the the UI is very different and, as you know, the the internals are very different because we chose to to provide a unified scheme interface to basically everything rural snicks. Has you know this specific domain-specific language? That's called the next language plus.

C

B

Bash scripts plus sequel, you know lots of things, so it's yeah, I guess I mean these are two different projects right, so they're. The common thing here is functional deployment and reproducibility, but then everything else is different. We take different focuses also so.

E

In reality, what kind of people will end up with Nixon? What kind of people will end up with gigs that work.

B

Well, so for me, one of the golden I started geeks was to to make Knicks usable by normal people. So.

C

B

For example, in terms of the of the CLI, the command-line interface I think it's easier to use geeks and Knicks nowadays, although the Knicks CLI has also vastly improved I guess about the last few years, but I really want it to be simple to use so we're trying how to wear them on the user interface and that I think that makes a difference if you've tried both you I hope you do see a difference. So that's one.

C

B

Also, if you're for tinkerers and I would hope that you know most people would at times be tinker. As you know, you don't have to be a tinker all of your time, but in gigs we really make it easy to to see what's going on in your system, because everything is in one language. It's it's very easy to jump from one place of your system to a different place.

B

It's sort of I mean twist lispers, I, wouldn't say it's a list machine when you run geek system, because obviously we're still running the typical new user land, which is lots of C code. But still there is. There is quite a bit of skin code in there, like the init system, isn't scheme everything to build the system itself isn't scheme, so that makes it really hack a ball and, to me very pleasant, I think to to Lee spirits. It should be very pleasant.

E

Nice, thank you.

A

Yeah so I guess it is now 10 past the official time and not week. Any last comments you would like to say well.

B

Just thank you for organizing this meeting. It's been a bit bumpy at times due to technical issues, but I'm I'm really happy that we were able to chat together.

E

Me too, thank you so much yeah.

A

Thanks a lot very interesting, yeah I'm, not sure little dick. If you know how refreshing it was for us to like have that glimpse of another another ecosystem and those inspiring ways, you use functional programming, and now you know it is for us the time to learn, to go and learn, and at least myself I must I must now try to do some guile and geeks, and thank you so much for this. It was.

B

Your so I hope.

A

We can screen.

B

In touch and discuss things like close repeater packaging or writing an importer for closure packages, things like that I would be interested in discussing it further great.

E

D

Yeah yeah good.

B

Bye. Thank you. Thank you.

B