2013/7/30 Joerg Mayer <jmayer@xxxxxxxxx>:
> On Mon, Jul 29, 2013 at 10:42:02PM +0200, Jakub Zawadzki wrote:
>> On Mon, Jul 29, 2013 at 01:25:41PM -0700, Guy Harris wrote:
>> >
>> > On Jul 29, 2013, at 1:12 PM, Jakub Zawadzki <darkjames-ws@xxxxxxxxxxxx> wrote:
>> >
>> > > Is anyone attached to hf_ variables? ;)
>> > >
>> > > There's no real need of them, and we can just replace them with
>> > > header_field_info structure.
>> >
>> > I.e., pass a pointer to the relevant structure, rather than an array index into an internal array that can supply that pointer?
>>
>> Yes.
>>
>> > I'm not sure there's a compelling reason (other than the effort involved in converting code) to keep the variables.
>>
>> Being devil's advocate I think there's no good reason to convert it.
>> Saving few CPU cycles by not doing:
>> - load variable into memory
>> - avoid PROTO_REGISTRAR_GET_NTH()
>>
>> and some cycles on startup + max 2-3 MB less in binary doesn't seems like good deal
>> against converting 90% of dissectors codebase :)
>
> If we don't loose functionality, I'd probably like to get rid of the extra
> variables (both ett_ and hf_) as it makes coding a bit more convenient.
> In the ett_ case: currently expansion is done similarly for all elements
> using the same ett variable, how would that be done without these vars?
>
> Can you give a sample diff how a dissector would need to be changed?
> Also: What would be the migration plan: All at once or incremental?
If we plan making such big changes which require touching practically
every dissector, it would be nice to migrate dissection related global variables
into a context structure which would allow us to run several dissections on
several files in one binary.
Cheers,
Balint
