[pyrale]

> If it becomes bootstrappable, then of course all the above critique is immediately retracted.

So basically you're saying you're just trying to get people to carry water for your project?

> because there is still literally no way to compile it from source

https://gitlab.haskell.org/ghc/ghc/-/wikis/building/#buildin...

I cannot comprehend how you can get to the conclusion that a compiler that was litterally made so that people could hack into it and learn from that has no build documentation.

[lrvick]

My project has no need of Haskell, but if anyone puts in the work to make haskell compileable from only public source code my team and I will put in the work to reproduce, package, and maintain it for the community for free as we do most other languages.

Your link details building GHC with an existing non reproducible GHC compiler binary compiled by a single individual that must be blindly trusted.

Full source bootstrapping means no binary blobs or trust in anyone else needed, which makes supply chain integrity possible. This is a bare minimum for any language to be considered for production use in any environment where security matters.

To me it -is- crazy when a major language compiler skips something so basic, but Haskell did.

To be fair rust team skipped this too, but thankfully rust is popular enough that a community member cared enough about high security applications to write mrustc, a bootstrap rust compiler written in C++. If not for that Rust would be in the same boat as Haskell.

Meanwhile Go and Zig did it right, and have both provided full source bootstrapping instructions from a C compiler since day 1.

[pyrale]

> Your link details building GHC with an existing non reproducible GHC compiler binary compiled by a single individual that must be blindly trusted.

You mean Hadrian? Its source is shipped with GHC.

Even if you were not to trust Hadrian, the doc also has info about building GHC using make.

> since day 1.

Could it be that languages made around 2010 have learned a thing or two from previous languages?

[lrvick]

Building GHC regardless of using hadrian or make still requires an existing GHC binary. That is the core trust problem.

GHC has a recursive dependency on itself with no way to go back before that loop.

[fuhsnn]

Interested in why you consider Zig's precompiled WASM blob full source bootstrapping, is it because it's shipped by first party?

[lrvick]

I did not package Zig myself. Does that blob make it into final artifacts? If so that is a bug we should swiftly correct.