Re: [Qemu-devel] [PATCH v3 ] doc: Introduce coding style for errors

[Thomas Huth]

On 27.01.2016 20:20, Lluís Vilanova wrote:
> Lluís Vilanova writes:
> 
>> Thomas Huth writes:
>>> On 20.01.2016 15:10, Lluís Vilanova wrote:
>>>> Thomas Huth writes:
>>>>
>>>>> On 18.01.2016 21:26, Eric Blake wrote:
>>>>>> On 01/15/2016 06:54 AM, Lluís Vilanova wrote:
>>>>>>> Gives some general guidelines for reporting errors in QEMU.
>>>>>>>
>>>>>>> Signed-off-by: Lluís Vilanova <address@hidden>
>>>>>>> ---
>>>>>>> HACKING |   36 ++++++++++++++++++++++++++++++++++++
>>>>>>> 1 file changed, 36 insertions(+)
>>>>> ...
>>>>>>> +Functions in this header are used to accumulate error messages in an 
>>>>>>> 'Error'
>>>>>>> +object, which can be propagated up the call chain where it is finally 
>>>>>>> reported.
>>>>>>> +
>>>>>>> +In its simplest form, you can immediately report an error with:
>>>>>>> +
>>>>>>> +    error_setg(&error_fatal, "Error with %s", "arguments");
>>>>>>
>>>>>> This paradigm doesn't appear anywhere in the current code base
>>>>>> (hw/ppc/spapr*.c has a few cases of error_setg(&error_abort), but
>>>>>> nothing directly passes error_fatal).  It's a bit odd to document
>>>>>> something that isn't actually used.
>>>>
>>>>> +1 for _not_ documenting this here: IMHO this looks ugly. If we want
>>>>> something like this, I think we should introduce a proper
>>>>> error_report_fatal() function instead.
>>>>
>>>> That's a bit of a chicken and egg problem. My main intention was to 
>>>> provide a
>>>> best practices summary on reporting messages/errors, since QEMU's code is 
>>>> really
>>>> heterogeneous on that regard. But there seems to be no consensus on some 
>>>> details
>>>> of what the proper way should be with the current interfaces.
>>>>
>>>> Utility functions for "regular messages", warnings, fatals and aborts would
>>>> definitiely be an improvement IMHO, but I dont have time to adapt existing 
>>>> code
>>>> to these (and I was told not to add unused utility functions for this).
>>>>
>>>> Now, if I were able to add such functions, it'd be something like:
>>>>
>>>> // Generate message "as is"; not sure if this should exist.
>>>> message_report(fmt, ...)
> 
>>> Not sure what this should be good for? We've already got error_report()
>>> that generates messages "as is", haven't we?
> 
>> Well, it just seemed wrong to me using error_report() to report "regular
>> messages" :)
> 
> 
>>>> // Generate message with prepended file/line information for the caller.
>>>> // Calls exit/abort on the last two.
>>>> error_report_{warn,fatal,abort}(fmt, ...)
>>>>
>>>> // Same with an added message from strerror.
>>>> error_report_{warn,fatal,abort}_errno(fmt, ...)
>>>>
>>>> But, should I add these without providing extensive patches that refactor 
>>>> code
>>>> to use them?
> 
>>> Maybe create a patch that introduces the _fatal and _abort functions
>>> (I'd skip the _warn functions for now), and use them in one or two files
>>> (e.g. replace the error_setg(&error_abort, ...) in spapr.c). That should
>>> not be that much of work, and could be a good base for further discussion?
> 
>> I can do that. But then should 'error_fatal' and 'error_abort' be officially
>> deprecated in favour of error_report_fatal() and error_report_abort()?
> 
> Sorry, I see this is misleading. I mean deprecate directly using
> "error_setg(error_fatal)"; you can still decide to pass error_fatal as an 
> error
> object to other user functions.

Since we hardly got any of these in the code right now, I don't see an
urgent need to explicitely say that this should be deprecated. I hope
that people rather will use the new functions automatically instead
since these sounds much more intuitive, IMHO.

 Thomas