Jeff Bezanson - What's the deal with Julia binary sizes?

Abstract:
If you have tried to produce a compiled library or application with PackageCompiler.jl, you may have been disappointed by the size of the resulting artifact. In this talk I will explain where all the bytes come from, and what we have done and continue to do to remove as many as possible. Fortunately sizes have been going down over time, and we have added some tools you can use now to reduce sizes further. I will briefly show how to use them.
Contents
00:00 Introduction
00:21 The function we'll compile with package compiler
01:23 Fora function that just does addition, the size is 901 MB (with Julia 1.6 based PkgCompiler)
02:43 Quick aside, sometimes binaries are just big cause they include a lot of things (hint, our binary has a lot too)
04:00 Specialization means big binaries
05:23 Exactly why is our package (sys image) so big?
06:42 Shrink ray time
11:22 Summary of our journey to get to 50MB from 900
16:14 The guide on how to shrink your thingamajig the same way
17:21 What's left in our sys image?
18:46 The Julia base library is kinda big
20:45 Reflection creates obstacles for reducing size of the binary
21:50 Oh, c'mon!
23:23 Roadmap
28:10 That's it for the slides
28:30 Q&A starts
S/O to @conradwiebe7919 for the video timestamps!
Want to help add timestamps to our UA-cam videos to help with discoverability? Find out more here: github.com/JuliaCommunity/You...
Interested in improving the auto generated captions? Get involved here: github.com/JuliaCommunity/You... you. Timestamps were added to the video, we will close this issue, when we get access rights to this GitHub repository.

[@kamilziemian995]

I always enjoy Jeff Bezanson's talks.

[@conradwiebe7919]

00:00 - intro
00:21 - The function we'll compile with package compiler
01:23 - Fora function that just does addition, the size is 901 MB (with Julia 1.6 based PkgCompiler)
02:43 - Quick aside, sometimes binaries are just big cause they include a lot of things (hint hint, our binary has a lot too)
04:00 - Specialization means big binaries
05:23 - Exactly why is our package (sys image) so big?
06:42 - Shrink ray time
11:22 - Summary of our journey to get to 50MB from 900
16:14 - The guide on how to shrink your thingamajig the same way
17:21 - What's left in our sys image?
18:46 - The Julia base library is kinda big
20:45 - Nobody like reflection (ok, actually, this segment is that reflection makes inspection difficult)
23:23 - Roadmap
28:10 - That's it for the slides
28:30 Q&A starts and I'm too lazy to segment them

[@kamilziemian995]

Thank you. We tweak your timestamps a bit and add the to the video description.

[@kamilziemian995]

@@gregandark8571 To qute Jeff Bezanson "Yikes!".

[@piratepartyftw]

This is the most compelling reason I see for Julia 2.0. Shrink Base. Refactor it to minimize the core language, and put other features in packages. It would also help with maintainability and technical debt. I hope eventually we get there. May take several years, though.

[@gregandark8571]

+/- how much?

[@gregandark8571]

I mean - In how many years do you think we'll get Julia 2.0 with all the native compilation candy's?

[@LiorAssouline]

Great talk! Can you please detail how to remove LLVM from the geneerated lib ?

[@LiorAssouline]

OK, I just realized it's just deleting it from the lib directory... 🤦

[@davidgustavsson4000]

If nothing useful is exported from the toy example, shouldn't the compiler just be able to make it all a noop?

[@Ptr-NG]

Merci d'avoir fourni un, Matlab like, riche outil gratuit..

[@androth1502]

so you went from "omg, that's bananas!" to "oh nelly, that's bloated!"

[@kamilziemian995]

21:50 😂