Re: [edk2-devel] [PATCH 02/11] OvmfPkg: add ShellLibs.dsc.inc

["Laszlo Ersek" <lersek@redhat.com>]

On 1/26/24 14:59, Gerd Hoffmann wrote:
>   Hi,
> 
>> My preferred style is observed for example in:
>>
>> - NetworkPkg/NetworkLibs.dsc.inc
>> - RedfishPkg/RedfishLibs.dsc.inc
>>
>> Alas, it is not observed in:
>>
>> - MdePkg/MdeLibs.dsc.inc
>> - OvmfPkg/Include/Dsc/OvmfTpmLibs.dsc.inc
>>
>> At least, from the last two, OvmfTpmLibs.dsc.inc has an excuse: it
>> provides library class resolutions for different module types.
> 
> Which is why I did it that way for TPM.  A single include file is enough
> instead of having multiple include files for multiple module types.  The
> OvmfPkg/Include/Dsc/OvmfTpmLibs.dsc.inc is placed in a location where
> this works without problems, and I've placed the new shell include at
> the same location.
> 
> I also have a WIP patch series (need to undust and rebase it ...) doing
> the same for the crypto stuff, and that has the same problem the TPM
> include has:  We have different configurations for PEI (stripped down
> config, with only the hash functions for measurements) and DXE (full
> TLS support).  And it is a single file with multiple sections too ...
> 
> tl;dr: I prefer the with-sections style.

OK!

> 
>> (2) This change makes the ShellCEntryLib resolution dependent on
>> BUILD_SHELL, which is a functional change, and it is not justified,
>> AFAICT.
> 
>> (2.1) "Bare bones" is for example
>> "MdeModulePkg/Application/SmiHandlerProfileInfo/SmiHandlerProfileInfo.inf".
>> [ ... ]
> 
>> (2.2) shell app is for example
>> "OvmfPkg/EnrollDefaultKeys/EnrollDefaultKeys.inf".
>> [ ... ]
> 
> Thanks for all this background info, much appreciated.
> 
>> Our BUILD_SHELL macro controls whether we build the shell itself, but
>> that is independent of whether we build applications that require the
>> shell to launch them. Making the ShellCEntryLib class resolution
>> dependent on BUILD_SHELL would only be valid if we also built all the
>> shell applications dependent on the BUILD_SHELL macro -- which is not
>> the case.
> 
> Given that EnrollDefaultKeys depends on the shell to launch
> I'm wondering whenever we should just change that and make the
> EnrollDefaultKeys build depend on BUILD_SHELL

This certainly sounds justified. It's hard to imagine a use case where
someone wanted an EnrollDefaultKeys.efi binary, but not a fresh shell.efi.

> (and also move it into Shell*.inc) ?

Seems to make sense, technically speaking; it's again the naming that
annoys me a bit. :( It's a utility that's supposed to be run from under
the shell, but not related to the shell itself. Hmm. :/

I guess we can live with it.

Laszlo



-=-=-=-=-=-=-=-=-=-=-=-
Groups.io Links: You receive all messages sent to this group.
View/Reply Online (#114685): https://edk2.groups.io/g/devel/message/114685
Mute This Topic: https://groups.io/mt/103935344/7686176
Group Owner: devel+owner@edk2.groups.io
Unsubscribe: https://edk2.groups.io/g/devel/leave/12367111/7686176/1913456212/xyzzy [rebecca@openfw.io]
-=-=-=-=-=-=-=-=-=-=-=-