Re: [9front] [PATCH] Unmount to remove sharp devices.

[hiro <23hiro@gmail.com>]

hating those runes.

did you consider: chdev REGEXP
could be the expanded version that allows driverA|driverB|driver[CDE]
could also allow ! or -v for reversing the meaning.

alternatively, why not use 0 instead of ∅ and '!' instead of ∉ ?

On 5/25/22, Jacob Moody <moody@mail.posixcafe.org> wrote:
> On 5/23/22 11:37, Jacob Moody wrote:
>> There was some followup discussion for this on irc that I want to
>> make sure is captured here:
>>
>> The command written to /dev/drivers will change to be one of:
>> chdev DEVS
>> chdev -DEVS
>
> I am having some second thoughts about this. While this syntax does
> replicate
> chmod, I am not entirely convinced that is what we want. When
> being added to /sys/src/libauth/newns:/^nsop/ you have to special case
> around
> the ARGBEGIN block. Combine that with the fact that this decision would
> lock
> chdev(1) from ever accepting it's own flags, it felt like this was
> introducing special case cruft.
> Another consideration was what the syntax would be for removing all
> drivers,
> currently it would just be chdev '' or something similar. Which is fine but
> fat fingering that would be an annoying way to kill your current namespace.
> These caveats may be entirely acceptable, but I was bothered enough to
> attempt to find another way.
>
> Instead, I thought perhaps a rune prefix might be best. So you end up with:
>
> chdev ∅
> for removing all drivers.
>
> chdev ∉DEVS
> For removing just the specified drivers.
>
> chdev ∈DEVS
> For removing access to all but the specified drivers.
>
> If a prefix is not given, ∈ is assumed.
>
> I did implement this idea, to field it for quirks,
> but wanted to get some feedback before redoing man page changes.
>
> The downsides I see here are:
>
> 1. Deviation from chmod
> 2. This steals three runes that could have otherwise been used for drivers
> 3. ∉ is a bit more involved to type then '-'
>
> With that said, I do much prefer this interface. Another option may be
> to use rune prefixes, but also provide flag synonyms to the prefixes in
> chdev(1).
>
> There is also an unrelated quirk I stumbled across when reviewing changes.
> As is currently, devshr checks for if the noattach has been set before
> allowing a process to post a fd. I have changed this to check for access to
> devmnt/'#M',
> which is what replicates the 'can mount' knob that RFNOMNT tweaks. This
> seems fine to me,
> but should we also replicate this check over in to devsrv for consistency?
>
> Thanks,
> moody
>