Re: [gentoo-soc] Weekly Report: Portage Powered Android

[Gunwant Jain]

Thanks for your query.

On 20/06/23 12:45PM, Pengcheng Xu wrote:
> Have you figured out _why_ Android is not booting?  I can think of a bunch of reasons, such as missing dependencies for a dynamically-linked executable, crashing on execution, etc.  Simply saying "would not boot Android" does not convey any information.

The binary exits after execution in the container. Also, it is a static
binary.

> Besides, I don't think you need NDK for something as low-level as an init; you shouldn’t need one as long as the code does not use Android-specific functions or call into Android service (e.g. vold).  You should be able to get away with something standard, i.e. newlib/musl/glibc.  I haven't read the source code though, and I might be wrong here; do your survey.

I had the same thoughts/arguments regarding using NDK for init. Also as
far as I can tell, it does not depend any Android-specific functions
either. But even then, using Gentoo's libc does not produce a working
binary. Couple of things to keep in mind regarding the NDK setup:
    - It uses clang instead of gcc
    - Uses a different sysroot, which is provided/bundled in NDK itself
      (it is bundled according to the API level).
    - It uses a different STL (again, bundled with NDK). 

> You may try with bionic toolchain, but just as I said, so long as no Android-specific APIs are involved, libc should not be the problem.

I would argue the same actually, but using Bionic is the closest we can
get to have a NDK like environment without actually using NDK.


If required, I could also have this week devoted entirely in rewriting
`bootstrap-init`, but this would be counter-productive since the same
code-base produces working binaries using a different build-system.

I am open to any new advices on this topic.

Regards,
Gunwant

>
> Regards,
> --
> Pengcheng Xu
> https://jsteward.moe