[OE-core] [PATCH] go: Allow bootstrapping using host go compiler

[richard.purdie at linuxfoundation.org]

On Fri, 2019-04-12 at 14:16 +0100, Daniel Thompson wrote:
> On Fri, Apr 12, 2019 at 01:39:52PM +0100, Richard Purdie wrote:
> > On Fri, 2019-04-12 at 13:30 +0100, Daniel Thompson wrote:
> > Whilst this is quite a creative approach it isn't deterministic and
> > I can see this introducing bugs which are hard to track down :(
> 
> I guess I can't disagree too strongly... the version check and
> bbplain message about the go toolchain we adopted were intended to
> mitigate this by at least making the selected go toolchain
> conspicious (and thus a git blame would probably draw flames towards
> me).

Once something makes it into sstate, the message is lost and you don't
know how it was created. The problems invariably end up on the mailing
lists/bugzilla with others being on the "front line" for this. Rightly
or wrongly its just the way it works out.

> We could make things fully deterministic by downloading go compiler
> binaries (rather than source) if the host architecture is not
> supported by go-1.4. Is this feasible or should I continue trying to
> adopt the host go compiler for these architectures?

Is the go compiler a standalone binary that can cope with this? I think
I prefer that to requiring people install host dependencies.


> > I think we need to make this a specific configuration option.
> 
> I guess we could wrap this up in a config option although I hope
> it is ok to have per-architecture defaults so the config option is
> always sane out-of-the-box.

Yes, per-arch is fine and we definitely like "out the box"! Not sure
any of us are classed as sane anymore but I digress! :)

Cheers,

Richard