Re: [PATCH] zram: check comp algorithm availability earlier

[Sergey Senozhatsky <sergey.senozhatsky.work@gmail.com>]

On (05/27/15 14:58), Minchan Kim wrote:
> > > I'm not against this patch because it's better than old.
> > > But let's think more about the pr_err part.
> > > 
> > > If user try to set wrong algo name, he can see EINVAL.
> > > Isn't it enough?
> > > 
> > > I think every sane admin can think he passed wrong argument
> > > if he sees -EINVAL.
> > > So, I don't think we need to emit pr_err in here.
> > > 
> > 
> > well, it's here simply to make failure investigation easier.
> > one surely will know that supplied string was not recognized
> > as a compression algorithm name, but what was it.. "$3 instead
> > of $2... or, wait, did $i contain something wrong?". zram knew
> > what was wrong.
> > 
> > /* and you asked to put this warn here in your previous email. */
> 
> Yes, Sorry about that. At that time, you put the warning in find_backend
> and I didn't like it. Instead, I want to move it to in there.
> But more thinking about it, I don't feel we need it.
> 

no problem.

> > > The reason I am paranoid about that is that I really don't want
> > > to argue with syslog info which is part of ABI or not in future.
> > > If possible, I don't want to depend on pr_xxx.
> > > 
> > 
> > just for the record...  I don't understand this part.
> 
> I meant if we remove the pr_err in future by some reason,
> someone might shout
> 
> "No, it's ABI so if you guys removes it, it will break user interface's
> semantic". Maybe he seems to depends on parse on dmesg.
>

ah, I see. well, hopefully no one does this. anyway, will remove it.


hm, I was thinking about some sort of a troubleshooting section in zram
documentation, where we can `decrypt' some of the errors that we return
back. but still did not convince myself that it will be of any use, just
doubt that people read documentation that often.

speaking of docs,
mentioning zramctl in zram docs is actully not entirely bad idea,
I think. zramctl has been in util-linux for (almost) a year now
and sooner or later will be widely available.

	-ss