Re: [PATCH v2] pahole: do not return an error when printing only a specific class

["Alexis Lothoré" <alexis.lothore@bootlin.com>]

Hi Alan,

On Thu Jul 31, 2025 at 11:10 AM CEST, Alan Maguire wrote:
> On 30/07/2025 16:26, Alexis Lothoré (eBPF Foundation) wrote:

[...]

>> Signed-off-by: Alexis Lothoré (eBPF Foundation) <alexis.lothore@bootlin.com>
>
> a few small things below, but
>
> Reviewed-by: Alan Maguire <alan.maguire@oracle.com>

[...]

>>  		case JOB_STEAL:
>> -			if (cus__steal_now(dcus->cus, job->cu, dcus->conf) == LSK__STOP_LOADING)
>> +			ret = cus__steal_now(dcus->cus, job->cu, dcus->conf);
>> +			if (ret == LSK__ABORT || ret == LSK__STOP_LOADING) {
>> +				dcus->lsk_status = ret;
>>  				goto out_abort;
>
> nit: I was thinking we could rename out_abort (rather than out_error) to
> out_early since we can exit due to either ABORT or STOP_LOADING now.

Yes, true, I did it for dwarf_cus__process_cus, but I forgot to do it here.
I'll update it here as well.

[...]

>> @@ -3386,15 +3386,16 @@ static enum load_steal_kind pahole_stealer(struct cu *cu, struct conf_load *conf
>>  	/*
>>  	 * If we found all the entries in --class_name, stop
>>  	 */
>> -	if (list_empty(&class_names)) {
>> -dump_and_stop:
>> +	if (list_empty(&class_names))
>>  		ret = LSK__STOP_LOADING;
>> -	}
>>  dump_it:
>>  	if (first_obj_only)
>>  		ret = LSK__STOP_LOADING;
>>  filter_it:
>>  	return ret;
>> +dump_and_abort:
>> +	ret = LSK__ABORT;
>> +	return ret;
>
> nit: we could just return LSK__ABORT at all the places we do "goto
> dump_and_abort" and avoid adding a new goto label here I think.

Ok, I'll add those few returns and get rid of this goto

Alexis

-- 
Alexis Lothoré, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com