[Zeek-Dev] support for event handlers using a subset of parameters

[Jan Grashöfer]

On 06/02/2019 21:00, Jon Siwek wrote:
> On Wed, Feb 6, 2019 at 1:30 PM Vlad Grigorescu <vlad at es.net> wrote:
> 
>> I _think_ I like Seth's idea of records, but I'm still thinking it through. It would formalize a growing trend towards moving more parameters into records anyway. If we're worried about backwards compatibility, then maybe we have a built in version number in each record. Whenever fields are added/removed, or there are more subtle contextual changes, the version number could increase.
> 
> Explicit versioning is a neat idea to maybe try expanding on.  I'm not
> quite sure how it would look for the user to, when they write their
> code, make it explicit that they expect the semantics of the record to
> match version XYZ.

If I get that right, this would mean that every record type comes with 
an individual version number, independent from the package or zeek 
version. While I can see that this allows to achieve backwards 
compatibility I am quite sure it will end in a complete mess. I expect 
the resulting code to be harder to write and to read, assuming that most 
of the records come with versioning constraints.

Actually, I like Jon's solution most. From a coder's perspective, the 
only thing I have to keep in mind is that parameters are matched by 
name. If I want to rename I can fallback to matching by order. That 
said, I would guess renaming a parameter is not a critical use case. I 
cannot imagine a situation where I rely on that feature. For me, 
renaming parameters makes code harder to read if I am familiar with the 
original definition. One could even think about introducing 
syntactically explicit renaming of parameters (which would probably 
break backwards compatibility again).

On 06/02/2019 02:40, Robin Sommer wrote:
> The following would be even worst in terms of confusion:
> 
>     global my_event: event(a: string, b: string);
>     event my_event(b: string)
> 
> Now I need to know if the language goes by order of parameters or by
> parameter name.

I can see that this might be confusing at first but I would argue that 
this is a one-time confusion. Versioned records in contrast would 
require me to think about their history every time I try to use a type.

Jan