Re: [PATCH] printf: add support for printing symbolic error codes

[Rasmus Villemoes <linux@rasmusvillemoes.dk>]

On 31/08/2019 00.21, Joe Perches wrote:
> On Sat, 2019-08-31 at 00:03 +0200, Rasmus Villemoes wrote:
>> On 30/08/2019 23.53, Joe Perches wrote:
>>>> diff --git a/lib/vsprintf.c b/lib/vsprintf.c
>>> []
>>>> @@ -2178,8 +2204,6 @@ char *pointer(const char *fmt, char *buf, char *end, void *ptr,
>>>>  		return flags_string(buf, end, ptr, spec, fmt);
>>>>  	case 'O':
>>>>  		return kobject_string(buf, end, ptr, spec, fmt);
>>>> -	case 'x':
>>>> -		return pointer_string(buf, end, ptr, spec);
>>>>  	}
>>>>  
>>>>  	/* default is to _not_ leak addresses, hash before printing */
>>>
>>> why remove this?
>>>
>>
>> The handling of %px is moved above the test for ptr being an ERR_PTR, so
>> that %px, ptr continues to be (roughly) equivalent to %08lx, (long)ptr.
> 
> Ah.
> Pity the flow of the switch/case is disrupted.

Agree, but I don't think it's that bad.

> That now deserves a separate comment.

You mean a comment like /* %px means the user explicitly wanted the
pointer formatted as a hex value. */. Or do you want (the|an additional)
comment somewhere inside the switch()?

> But why not just extend check_pointer_msg?

Partly because that would rely on all %p<foo> actually eventually
passing ptr through to that (notably plain %p does not), partly because
the way check_pointer_msg works means that it has to return a string for
its caller to print - which is ok when the errcode is found, but breaks
if it needs to format a decimal. It can't even snprintf() to a stack
buffer and return that, because, well, you can't do that, and it would
be a silly recursive snprintf anyway.

OK, so perhaps you meant check_pointer() where it might be doable, but
again, with plain %p and possibly others we don't get to that.

Rasmus