Re: Followup of yesterday's design session "refactoring the REST"

[Roger Pau Monné <roger.pau@xxxxxxxxxx>]

On Thu, Jul 09, 2020 at 07:56:58AM +0200, Jürgen Groß wrote:
> Yesterday's design session at Xen Developer Summit "Hypervisor Team: .."
> had one topic regarding whether we should find specific maintainers of
> all the files currently assigned to "THE REST" in order to lower the
> amount of reviews for those assigned to be "THE REST" maintainers.
> 
> Modifying the MAINTAINERS file adding "REST@x.y" as REST maintainer
> and running the rune:
> 
> git ls-files | while true; do f=`line`; [ "$f" = "" ] && exit; \
> echo $f `./scripts/get_maintainer.pl -f $f | awk '{print $(NF)}'`; \
> done | awk '/REST/ { print $1}'
> 
> shows that basically the following files are covered by "THE REST":
> 
> - files directly in /
> - config/
> - most files in docs/ (not docs/man/)
> - misc/ (only one file)
> - scripts/
> - lots of files in xen/common/
> - xen/crypto/
> - lots of files in xen/drivers/
> - lots of files in xen/include/
> - xen/scripts/
> - some files in xen/tools/
> 
> I have attached the file list.

Thanks! I still have to go over the list in more detail, just some
comments below.

> So the basic idea to have a "hypervisor REST" and a "tools REST"
> wouldn't make a huge difference, if we don't assign docs/ to "tools
> REST".
> 
> So I think it would make sense to:
> 
> - look through the docs/ and xen/include/ files whether some of those
>   can be assigned to a component already having dedicated maintainers
> 
> - try to find maintainers for the other files, especially those in
>   xen/common/ and xen/drivers/ (including the related include files, of
>   course)

I think it's important that xen/common files (specially the ones
containing interfaces exposed to guests) have at least one maintainer
from each supported architecture (Arm and x86 ATM), plus whatever
common code maintainers we want to have. It's sometimes easy (at least
for me) to forget about other arches or make wrong assumptions about
them when modifying common code.

Drivers could also benefit from something similar IMO, where common
code has a shared group of maintainers, for example IOMMU should
ideally have a mix of maintainers from the current implementations
(Arm/Intel/AMD) plus again whatever common code maintainers we want
to have.

Roger.