Re: [edk2-devel] Adding Bhyve support into upstream EDK2

["Ard Biesheuvel" <ard.biesheuvel@linaro.org>]

On Sat, 7 Mar 2020 at 08:39, Laszlo Ersek <lersek@redhat.com> wrote:
>
> Hi Jiewen,
>
> On 03/07/20 02:43, Yao, Jiewen wrote:
> > Just saw Laszlo's email. Similar feedback. Especially, I like the regression test part.
>
> Thanks.
>
> > I am not sure how many virtual platforms we will have eventually.
> > If there are more and more, maybe we can create a new edk2-virt-platform repo, and put them together there. (Similar to edk2-platform repo for the physical platform)
>
> Regarding the last part ("move them together here") -- I'm 100% opposed
> to removing OvmfPkg and ArmVirtPkg from edk2. They *must* remain in the
> exact same git repository where the core (MdePkg, MdeModulePkg,
> CryptoPkg, SecurityPkg, UefiCpuPkg, ...) lives too, and share a common
> git history.
>

Agreed.

> ArmVirtPkg and OvmfPkg move very closely together with the core, most
> significant ArmVirtPkg and OvmfPkg contributions need changes (and
> therefore introduce new dependencies) on the core. Managing such
> dependencies is a nightmare evein with git submodules; it only works if
> the git history is shared. This problem is not theoretical, it already
> has a bad effect on edk2-platforms.
>
> For a recent example, my latest OvmfPkg patch series:
>
> https://bugzilla.tianocore.org/show_bug.cgi?id=1512#c18
>
> merged as commit range 61d3b2d4279e..1158fc8e2c7b, started by improving
> the logging in MdeModulePkg/PiSmmCore (a1ddad95933e), and fixing a bug
> in UefiCpuPkg/PiSmmCpuDxeSmm (90e11edd16c7).
>
> I don't necessarily mind if *new* virtual platforms are outside of the
> edk2 tree, but if I'm completely honest about "why", it's because I
> don't use those new platforms. And that's a *selfish* reason -- if I
> want ArmVirtPkg and OvmfPkg to benefit from sharing and interleaving
> their histories with the core, then other virtual platforms deserve the
> same, even if I don't use them.
>
> (In fact, I think that even edk2-platforms should never have been split
> out of edk2 -- but that ship has sailed. I believe I argued against
> separating edk2-platforms, but my reasons weren't strong or convincing
> enough.)
>

Yes, keeping platforms in sync with the core is more painful than it
should be. If we move all platforms out of the core, what are we going
to do for validation? Sure, we'll get a nice tickbox from Azure that
all the semicolons line up nicely, but being able to build something
that can be tested on actual hardware is essential IMO.