Re: [PATCH 01/11] perf tools: Introduce struct perf_maps_opts

[David Ahern <dsahern@gmail.com>]

On 02/13/2012 12:05 PM, Arnaldo Carvalho de Melo wrote:
> Em Mon, Feb 13, 2012 at 11:50:29AM -0700, David Ahern escreveu:
>> The cleanup might make my multiple tid/pid patch easier: e.g.,
>>
>> struct perf_target{
>> 	...
>> 	char errmsg[128];
>> };
>>
>> Then if the tid/pid string parsing fails in perf_evlist__create_maps and
>> friends the errmsg can be put into the buffer for the callers to get a
>> more useful message to the user as to what happened.
>>
>> Today's perf if you give it an invalid pid, scandir fails and the
>> command spits out the usage statement. Which is completely confusing --
>> ie., not clear that the command failed b/c the pid does not exist.
> 
> Humm, ok, but then I think we should have an enum + a strerror(3)
> equivalent, i.e.:
> 
> 	enum perf_target_error perf_evlist__create_maps(...);
> 
> 	int perf_target__strerror(struct perf_target *target, int errnum,
> 				  char *buf, size_t buflen);

ok, so you are proposing an internal generation of enum error codes and
correlating them to strings rather than adding a buffer into
perf_target. If that's the case perhaps we need a libperf-wide design:

enum perf_error
perf__strerror(enum perf_error)

which effectively taps an array similar to _sys_errlist_internal based
on enum index.

> 
> Please see 'man strerror_r", and make it work like the POSIX compliant
> variant.

No globals are in use, so I would expect the _r to be redundant. I have
glibc source; scanning __strerror_r implementation ....

> 
> Ok, so it may be better to first process Kim's patches and then you
> rework yours?

The current patch is ready to go; I just don't like the error handling
and lack of a useful message. That said, it is no worse than what
happens today.

David

> 
> - Arnaldo