Re: [RFC 1/2] ath9k: add DFS pattern detector

[Felix Fietkau <nbd@openwrt.org>]

On 2012-01-27 4:09 PM, Zefir Kurtisi wrote:
> On 01/26/2012 04:53 PM, Felix Fietkau wrote:
>> On 2012-01-26 4:34 PM, Zefir Kurtisi wrote:
>> [...]
>>> +/**
>>> + * struct pattern_detector - overloading base dfs_pattern_detector
>>> + *
>>> + * @exit(): destructor
>>> + * @add_pulse(): add radar pulse to detector
>>> + * @num_radar_types: number of different radar types
>>> + * @last_pulse_ts: time stamp of last valid pulse
>>> + * @radar_detector_specs: array of radar detection specs
>>> + * @channel_detectors: list connecting channel_detector elements
>>> + */
>>> +struct pattern_detector {
>>> +	void (*exit)(struct pattern_detector *_this);
>>> +	enum dfs_detector_result (*add_pulse)
>>> +		(struct pattern_detector *_this, struct pulse_event *pe);
>>> +
>>> +	u8 num_radar_types;
>>> +	u64 last_pulse_ts;
>>> +	struct radar_detector_specs *radar_spec;
>>> +	struct list_head channel_detectors;
>>> +};
>> To overload it this way is quite fragile. It's better to embed struct
>> dfs_pattern_detector here. In places where you need to go from the
>> struct dfs_pattern_detector to this struct, you can then use the
>> container_of macro, to get at least some form of type safety.
>> 
>> - Felix
> 
> Hi Felix,
> 
> thanks for taking a look.
> 
> Actually, for this initial post we do not need any polymorphism at
> all, I could basically make this derived class the interface in the
> PATCH to come.
> 
> Though (since I am used to this coding style) I'd like to understand
> your concern. By 'fragile' you are referring to the risk of base and
> derived class diverging, right? If it is that, I see that one is in
> trouble if you let that happen. Here with the base class being
> defined as external interface holding just two function pointers I
> thought the risk is negligible.
Yes, I'm talking about the risk of those two diverging. It may be a
small risk right now, but people tend to forget about things like that
over time, and then it becomes bigger. Also, this is very easy to avoid
in a way that does not change the compiled binary code in any way.

- Felix