Re: [Qemu-devel] [PATCH 09/19] sockets: Chardev open error reporting, sockets part

[Anthony Liguori <anthony@codemonkey.ws>]

On 02/14/2012 11:24 AM, Markus Armbruster wrote:
> Markus Armbruster<armbru@redhat.com>  writes:
>
>> Anthony Liguori<aliguori@us.ibm.com>  writes:
> [Anthony asking for error_set() instead of error_report()...]
>>> Basically, same thing here and the remaining functions.  Let's not
>>> introduce additional uses of error_report().
>>>
>>> That said, I imagine you don't want to introduce a bunch of error
>>> types for these different things and that's probably not productive
>>> anyway.
> [...]
>>> So let's compromise and introduce a generic QERR_INTERNAL_ERROR that
>>> takes a single human readable string as an argument.  We can have a
>>> wrapper for it that also records location information in the error
>>> object.
>>
>> This series goes from stderr to error_report().  That's a relatively
>> simple step, which makes it relatively easy to review.  I'm afraid
>> moving all the way to error.h in one step wouldn't be as easy.  Kevin
>> suggests to do it in a follow-up series, and I agree.

The trouble I have is not about doing things incrementally, but rather touching 
a lot of code incrementally.  Most of the code you touch could be done 
incrementally with error_set().

For instance, you could touch inet_listen_opts() and just add an Error ** as the 
last argument.  You can change all callers to simply do:

Error *err = NULL;

...
inet_listen_opts(..., &err);
if (err) {
    error_report_err(err);
    return -1;
}

And it's not really all that different from the series as it stands today.  I 
agree that aggressively refactoring error propagation is probably not necessary 
as a first step, but if we're going to touch a lot of code, we should do it in a 
way that we don't have to immediately touch it again next.

>>
>> Can you point to an existing conversion from error_report() to error.h,
>> to give us an idea how it's supposed to be done?
>
> Ping?

Sorry, I mentally responded bug neglected to actually respond.

All of the QMP work that Luiz is doing effectively does this so there are ample 
examples right now.  The change command is probably a good place to start.

>