On Apr 4, 2019, at 9:06 AM, Laszlo Ersek <lersek@redhat.com> wrote:

On 04/04/19 17:10, Andrew Fish wrote:

On Apr 4, 2019, at 3:45 AM, Laszlo Ersek <lersek@redhat.com> wrote:

On 04/04/19 06:09, Andrew Fish wrote:

On Apr 3, 2019, at 8:42 PM, Ni, Ray <ray.ni@intel.com> wrote:

Mike, Laszlo,
It's a good idea to store the shell binaries into the assets of each stable tag.

If we go in this way, it means "build" requires network connection to download the
shell binary from the assets of a certain release.
Do you think it's acceptable?

Ray,

The other option would be to have a configuration step, like installing Python or the C compilers, that copies the binary. You need a network connection to clone the git repo and to stay in sync with it. I guess you could model that as a git submodule, or actually have a script that grabs the binary you want from a remote system, and fall back to the local copy if you don't have a network connection. 

Or we can separate the binary download and build into two phases so build phase
can be independent on network connection.

Is there any known practice/solution for such requirement (stable sub-component binaries
needed by a production image generation)?

I think to some extent this kind of thing is driven by the customers build rules. Basically what the customer think of as their manifest of parts for software version X.

I suggested PREBUILD because I took it as a given, from Mike's problem
statement, that "build" had to ensure, internally, the local
availability of the shell binary.

If that's a not requirement, then IMO it's much better to leave it to
organizations to fetch the prerequisites of their platform builds. I'd
say that's out of scope for upstream edk2 -- if they need the shell
binary to be available off-line, at their build time, they can download
it earlier and cache it locally.

Laszlo,

I guess for edk2 projects the maintainers own the manifest. So the edk2 projects that need the Shell should define how that works. I don't think we need to define a generic solution for 3rd parties as I'd guess Red Hat and Microsoft probably already have tools and strategies to deal with cobbling together software from different packages.  

So I guess we should ask the maintainers of the ekd2 packages does the version of the Shell matter? If no then just pre-install a shell binary as part of the setup. If the version matters then we should look into doing something a little more fancy, and use the pre-installed shell binary as the fallback. 

Is there anyway to tell the Shell version from the Shell PE/COFF? One option could be a build warning if the shell is old and just have the user manually update the shell if needed. 

As a co-maintainer under OvmfPkg and ArmVirtPkg, I prefer to build the
shell from source at all times, namely from the source code that is part
of the entire edk2 tree at a given commit / checkout.

I don't see any possibility or desire for the virtual firmware packages
(RPMs) that I have a say in to consume/ship a pre-built UEFI shell binary.

--*--

The reason I recommend for us (the TianoCore community) to offer the
shell as a prebuilt binary too, somewhere on the web, is because it
would help UEFI users (in the most general sense).

Thanks,
Laszlo