Re: [PATCH 1/2] Add namespace tags that can be used for matching without pinning a ns

[Jarkko Sakkinen]

On Fri, Feb 05, 2021 at 08:25:35AM +0000, David Howells wrote:
> Jarkko Sakkinen <jarkko@xxxxxxxxxx> wrote:
> 
> > > + * init_ns_common - Initialise the common part of a namespace
> > 
> > Nit: init_ns_common()
> 
> Interesting.  The majority of code doesn't put the brackets in.
> 
> > I've used lately (e.g. arch/x86/kernel/cpu/sgx/ioctl.c) along the lines:
> > 
> > * Return:
> > * - 0:          Initialization was successful.
> > * - -ENOMEM:    Out of memory.
> 
> Actually, looking at kernel-doc.rst, this isn't necessarily the recommended
> approach as it will much everything into one line, complete with dashes, and
> can't handle splitting over lines.  You probably meant:
> 
>       * Return:
>       * * 0		- OK to runtime suspend the device
>       * * -EBUSY	- Device should not be runtime suspended

A line beginning with dash, lines up just as well, as one beginning with
an asterisk. I've also tested this with "make htmldocs".

This is Mauro's response to my recent patch:

https://lore.kernel.org/lkml/20210125105353.5c695d42@xxxxxxxx/

So, what I can make up from this is that they are equally good
alternatives.

What I'm not still fully registering is the dash after the return value.

I mean double comma is used after parameter. Why this weird dash syntax
is used after return value I have no idea, and the kernel-doc.rst does
not provide any explanation.

> 
> > * Return:
> > * - 0:          Initialization was successful.
> > * - -ENOMEM:    Out of memory.
> > 
> > Looking at the implementation, I guess this is a complete representation of
> > what it can return?
> 
> It isn't.  It can return at least -ENOSPC as well, but it's awkward detailing
> the errors from functions it calls since they can change and then the
> description here is wrong.  I'm not sure there's a perfect answer to that.
> 
> David

What if you just add this as the last entry:

* * -errno:     Otherwise.

/Jarkko