git: 2c26d77d989a - main - Remove /boot/efi from mtree, missed in 0b7472b3d8d2.

[Nathan Whitehorn]

On 3/2/21 10:19 PM, Warner Losh wrote:
>
>
> On Tue, Mar 2, 2021, 7:01 PM Ed Maste <emaste at freebsd.org 
> <mailto:emaste at freebsd.org>> wrote:
>
>     On Tue, 2 Mar 2021 at 19:30, Warner Losh <imp at bsdimp.com
>     <mailto:imp at bsdimp.com>> wrote:
>     >
>     > There has been some talk of moving the mount point to /efi, but
>     I think that went nowhere...
>
>     I thought about /efi based on arguments from the Linux world, more
>     info at
>     https://wiki.archlinux.org/index.php/EFI_system_partition#Typical_mount_points
>     <https://wiki.archlinux.org/index.php/EFI_system_partition#Typical_mount_points>.
>     I did open https://reviews.freebsd.org/D28814
>     <https://reviews.freebsd.org/D28814> to move the existing
>     uses to /efi. That said, I'm much more interested in us picking
>     something and using it in the installer/other tooling than whether
>     it's /boot/efi or /efi.
>
>
> Me too. I am actually open to either. If I had a do over, I'd pick 
> /efi honestly.
>
> But the directory presence or absence shouldn't be used to know if we 
> have an ESP to deal with or not.
>
> Warner
>

It would have been nice to know that when I posted the original patch 
for review that relied on this mechanism two weeks ago and you and Ed 
both signed off. Or when we discussed this on Friday and no one knew why 
it was in mtree -- I still honestly don't know from this thread -- and 
then I proposed removing it and everyone said that was fine. There are 
lots of other directories that we make in the installer that aren't in 
mtree. What's special about this one?

The problem here is that the installer needs to know if an ESP has been 
made. If /boot/efi is supposed to be the mountpoint for an ESP, its 
presence seemed like a reasonable way to signal that. Here are some options:
- We could try to see if a partition is actually mounted there. That's 
mildly annoying -- it involves parsing some shell output -- and results 
in making a pointless directory on all non-EFI architectures.
- We could try to have the installer signal to itself via some temp 
file. This seems fragile and complicated.
- We could duplicate the logic for whether to make an ESP in the first 
place in several places, which seems like a terrible idea.

But this needs to be settled *now*. It's already very, very late to get 
this into 13.0 and we can't have another cycle of changing our minds 
about reviewed code.
-Nathan