Go-boot | Not Hacker News!

Discussion (15 comments)

Showing 36 comments

16 days ago

4 replies

I really like this idea but can anyone please summarize what it does for me. To me it feels very fascinating (bare metal golang in general) but I am not sure I truly understand its usecase and I would love to know more.

pjmlp

16 days ago

6 replies

The use cases is not writing unsafe C in first place, and proving the point Go is usable in such scenarios, regardless of naysayers.

The creators of USB Armory also created TamaGo, instead of using Rust, exactly for the same reasons, to prove a point.

https://github.com/usbarmory/tamago

https://reversec.com/usb-armory/

Because in IT, seeing is believing.

bradfitz

15 days ago

1 reply

I've been idly following this stuff on & off for years, but I never saw proving a point "instead of using Rust" as one of the motivations of the project. Was that ever stated anywhere?

pjmlp

15 days ago

Yes,

> Languages like Rust have already proven they role in bare metal world, Go on the other hand needs to … and it really can!

From https://fiif.fi/wp-content/uploads/sites/9/2021/06/TamaGo.pd...

schmuckonwheels

10 days ago

3 replies

If one can't write safe C code, then maybe stick to web development and leave the bootloaders and UEFI stuff to people who can.

Training wheels are merely a race to the bottom for barely-literate programmers.

throwaway894345

10 days ago

1 reply

There are only people who think they can write safe C code and those who know they can’t.

pjmlp

10 days ago

1 reply

Including the language authors, let that sink in.

throwaway894345

9 days ago

Yeah, it’s very much like the meme showing the bell curve with the novice and the wizard/expert both saying “I can’t write safe C code” and the guy in the middle bragging that he can.

pjmlp

10 days ago

Nah, people ignore on purpose that C creators are the first to acknowledge C's flaws, hence why Alef, Limbo and Go were created by them, and Plan 9/Inferno as improvements on UNIX.

Too many focus on where the journey started instead of where it ended.

monocasa

10 days ago

The number of memory safety CVEs written in C by people who ostensibly 'didn't need training wheels' point strongly to the antithesis of your argument.

And I say that as someone who's been a kernel engineer for 20 years.

flanked-evergl

10 days ago

1 reply

No amount of proven points will give Go null safety, though.

pjmlp

10 days ago

1 reply

Yet the whole Docker, Kubernetes, CNCF ecosystem is powered by Go, doesn't seem to have been hindered by lack of null safety.

Same applies to GCP, AWS and Azure, powered mostly by Java, C# and C++.

People should stop being so obsessed with one specific language feature, when there is so much C and C++ code being produced every day.

mrsmrtss

10 days ago

1 reply

And Linux kernel is written in C etc, so by this logic you don't even need memory safety. There is no good excuse for designing a language in modern times (this century) with every object nullable by default. C# at least mostly has solved this design mistake later by introducing nullable reference types (https://learn.microsoft.com/en-us/dotnet/csharp/nullable-ref...). Then again, Go designers insisted that generics were also unnecessary, until they changed their mind.

pjmlp

9 days ago

1 reply

On the contrary, because there we have 40 years of security exploits to prove otherwise, and Linux kernel has plenty of CVEs.

C# solution doesn't work, most projects never adopted it, because it is a mess to use with third party libraries that never bothered to add the required annotations, hence why it is still a warning and optional to this day.

mrsmrtss

9 days ago

I’m not sure which .NET libraries you are referring to, but all the ones we use have nullable reference types enabled. If you configure warnings as errors (as you should), then it works exceptionally well. Even if you were to use a library where nullable reference types are not enabled, you only need to check for null once during the library call, rather than everywhere in your codebase.

xyse53

15 days ago

It's also a good way to learn about UEFI for people most familiar with go.

guywithahat

10 days ago

That's a shame, I was hoping it would be so I could boot thousands of kernels in parallel at once

qhwudbebd

16 days ago

Quite apart from that, an EFI shell that's less awful than the standard UEFI one is an interesting project in its own right...

tomcam

15 days ago

2 replies

When you turn on a computer, it transfers code to software required to get the machine up and running reliably--the boot process. That used start in a chip called the BIOS. It's a 40-year old holdover from the early days of the IBM PC. UEFI is a more complex and feature-rich protocol. Due to its default memory management Go hasn't been considered the first choice for such purposes but this proof of concept uses Go for the very low level code needed for UEFI.

reactordev

10 days ago

1 reply

[delayed]

tomcam

5d ago

Was trying to be concise. Also, stop accusing me of having any pride. I'm married and a father!

pjmlp

10 days ago

GC has never been an impediment for Xerox PARC.

typical182

10 days ago

1 reply

There’s a proposal from the folks behind this project to upstream the Go runtime hooks that enable this into Go proper.

From what I can tell, the core Go team seems generally favorable to it, so seems like a decent chance it will happen.

From:

#73608 proposal: all: add bare metal support

https://github.com/golang/go/issues/73608

> Go applications built with GOOS=none would run on bare metal, without any underlying OS. All required support is provided by the Go runtime and external driver packages, also written in Go.

And:

> These hooks act as a "Rosetta Stone" for integration of a freestanding Go runtime within an arbitrary environment, whether bare metal or OS supported.

clktmr

10 days ago

This would also benefit the Embedded Go project, which uses similar modifications to the runtime.

https://embeddedgo.github.io/

techgnosis

15 days ago

There aren't that many UEFI shells and the ones that exist are certainly not modern. Anything new is helpful, especially if its written in a popular language like Go.

hulitu

13 days ago

1 reply

> Go-boot: bare metal Go UEFI boot manager

The bare metal list is quiet thin.

Why is so HW focused ? I use refind and it seems to be HW independent.

danudey

9 days ago

That's the list of hardware they've explicitly tested on. Always bear in mind that, for any given standard, no matter how straightforward, there are going to be dozens of vendors who screw it up for no real reason other than incompetence or malice.

The older a piece of software is, the more workarounds it will have accrued for various hardware bugs or vendor misdeeds, so it's reasonable for the project to disclaim that it's only been tested on a small number of physical hardware devices even if, in theory, it should work out of the box on all of them.

seanw444

10 days ago

3 replies

As much as I appreciate Go, putting it on bare metal makes me cringe a little.

gethly

10 days ago

If that makes you cringe, I cannot even begin to imagine what this https://tinygo.org will do to you.

pjmlp

10 days ago

Why? Xerox PARC used to do this.

As did all machines that booted into a Lisp or BASIC REPL.

reactordev

10 days ago

[delayed]

i-con

10 days ago

1 reply

I'm confused, is it bare metal or is it an EFI application? (bare metal used to mean that something can run without services, like those that UEFI provides)

danudey

9 days ago

I think what it means is:

1. It's an EFI application

2. It doesn't require any external runtimes, any setup, etc. (i.e. your UEFI system can boot straight into it without anything in between).

At least, that seems to be the case.

tuananh

10 days ago

There's also Sprout by Edera https://github.com/edera-dev/sprout

> Sprout: UEFI Bootloader in Rust

pizlonator

10 days ago