[llvm-dev] Target Acceptance Policy

[Renato Golin via llvm-dev]

On 26 July 2016 at 01:54, Chandler Carruth <chandlerc at google.com> wrote:
> Every backend I can think of other than WebAssembly and the original AArch64
> port has come into existence out of tree, and only after proving useful to
> some folks decided to move into the tree. I'd personally advocate for
> keeping that door open.

Indeed. Minimum requirements, not expected target.


> All that said,I completely agree regarding coding conventions and other
> basic stuff. I don't think we need to wait for the official point in time to
> insist on all of the fundamental coding standards and such being followed.

Indeed. Expected target, not minimum requirements. :)

Bu having them separated, we're telling people that the latter is what
they should strive for, but we're happy to help them along the way.


>> > 7. The test coverage is broad and well written (small tests,
>> > documented).
>
> Yep, this seems like it should always be true regardless of the experimental
> nature.

I think this is the same as above. New targets are generally
introduced by people that don't know LLVM too well, or haven't written
a full back-end before.

Expecting *full* coverage and perfect tests from day one is as
unreasonable as expecting them to fit all the required code style and
other bits.

But in order for it to be relevant upstream (ie. no maintenance
burden), the tests need to be in good quality. After all, a lot of
unrelated buildbots will be running *their* tests.

It's the same thing: start small, grow up, become official. All in
all, we're here to help them grow. That's the message I'm trying to
pass along.

cheers,
--renato