[Tech] Fcp woes II

[juergen urner]

Matthew Toseland schrieb:
> On Sunday 09 March 2008 15:25, juergen urner wrote:
>   
>> There is no distinction between ClientPutDiskDir and ClientPutComplexDir
>> in PersistentPut*. This makes it hard to tell both apart. I need this to 
>> do automatic
>> type conversions.
>>     
>
> We can make a distinction if we need to, however there really isn't much 
> difference, ClientPutDiskDir is just a simple way to access the same 
> functionality, any ClientPutDiskDir can be expressed as a ClientPutComplexDir 
> if need be.
>   

Yes, I noticed that. And I noticed that 'Filename' is not present in 
PeristentPutDir
when I pass ClientPutComplexDir. just a bit of ugly parsing to find out 
'Filename'
I'd like to avoid. So, at least passing 'Filename' as indicator would be 
helpful.


>> Just thinking aloud ..I wish ClientPut* (PersistentPut*) would go. 
>> Afaics a new message
>> would make live much easier on both sides. Any thoughts?
>>
>> Put
>>     Identifer=any
>>     NItems=N
>>     Persistence=whatever
>>     (...)
>> EndMessage
>>
>> ...emidiately followed by N items to put
>>
>> DataItem
>>     DataLength=N
>>     Name=any
>>     (...)
>> EndMessage
>> FileItem
>>     Filename=filename
>>     Name=any
>>      (...)
>> EndMessage
>> (...)
>>     
>
> Why is this better?
>   

Put* is somewhat overcomplicated. I am already scared of future 
extensions to it.

As I see it, all that is needed is is to break the container into its 
peaces to
make a nice and clean api for both sides on only one message and one page
in the wiki.


>> Btw, on input Fcp does not seem to care if Files.N.* in ClientPutComplexDir
>> start at 0. This broke persistents.
>>     
>
> You mean it expects it to start at 0? Or what?
>   

Yes something breaks. Too lazy to run more tests, but NodeHello does
not arrive anymore.



Btw, I ran a test throwing 'Plum' as message terminator at the node.

 >>> If '=' not in chunk:
 >>>    endOfMessageEncountered()

Fcp doesn' t care at all?