Hi Pete,
sorry for the delay, but the department opened today. Previously I was
at home using the ugliest gprs. Now I am in the office booting the
machines...
Is Andy coming here tomorrow?
Fabrizio
Peter Elmer wrote:
> [Move discussion to the new xrootd mailing list from BaBar KanSOS HN]
>
> Hi Fabrizio,
>
> Perhaps you, Heinz and I could talk by phone briefly after lunch? Are you
> available? Heinz, do you have any time?
>
> Pete
>
> On Thu, Aug 19, 2004 at 10:35:05AM +0000, Fabrizio Furano wrote:
>
>>Hi all,
>>
>> I am back after the vacations outage....
>>
>> Yes, I looked at the code, and I think that the fastest way to get to
>>the result is to borrow some components from xrootd/heinz's client.
>>These could be used to substitute the few classes that TXNetfile uses
>>from root.
>>
>> I see mainly three points in the work to do, and I have answers for
>>two of them:
>>
>>- client sockets: I can build a simple socket class similar to TSocket
>>and use it from TXSocket.
>>- private members coming from TNetfile/TFile: they are not many, 5-7
>>data members to incorporate in the new interface
>>- Configuration flags and parameters: now they come from TEnv. What do
>>you suggest for them in the new client? We might use:
>> - envvars... I know Pete says they are *evil*
>> - pass to the ctor some optional data structure
>> - use the default parameters as constants. After all, they have
>> never been changed...
>>
>>
>>My suggestion, to speed up the work, is to start from the latest
>>TXNetFile inside root, and making it independent from the root classes
>>while still in the framework. When the result is reached, then we can
>>deal with a standalone makefile.
>>
>>I am not enthusiast of the idea of doing the opposite thing: starting
>>from another client's structure and putting inside all the other
>>features. I believe that this is more complicated and prone to bugs
>>everywhere.
>>
>>Fabrizio
>>
>>[log in to unmask] wrote:
>>
>>>Hi Pete,
>>>
>>>
>>>
>>>>Do you see a reason why we shouldn't just implement write() in Heinz's
>>>>client and go from there?
>>>
>>>I see nothing wrong in using it as a starting point. However, we do have
>>>some missing protocol pieces that need to be added (e.g., connection
>>>multiplexing). I would dare say that Fabrizio should comment as well.
>>>
>>>
>>>
>>>>Ah, right. So what do we have to turn on to get the olbd's to allow
>>>>(redirection for) writing? I see:
>>>>
>>>>olb.path s /store
>>>>
>>>>The doc on this is a bit terse. I guess that "s" actually means "rs" by
>>>>default so I would need something like:
>>>>
>>>>olb.path w /whatever
>>>>
>>>>to allow me to write files to /kanga/whatever
>>>
>>>Almost right (see below). Terse, eh? You're always welcome to suggest
>>>expansions.
>>>
>>>
>>>
>>>>? (And turn off the staging, which wouldn't be relevant in the case of
>>>>the "output buffers", for example. In that case "whatever" would actually
>>>>be "store", but that is a detail.)
>>>
>>>Not quite. In fact, you really do need to say "ws" because, in fact, you
>>>will be creating files. Temporally, it is difficult to figure wout whether
>>>you are trying to update an existing file or create one from whole cloth.
>>>So, you could get rid of "s" but then you would not be able to create any
>>>new files. I guess I need to check what the xrootd is configured to do by
>>>default (I know we decided this based on Objectivity years ago).
>>>
>>>Andy
>
>
>
>
> -------------------------------------------------------------------------
> Peter Elmer E-mail: [log in to unmask] Phone: +41 (22) 767-4644
> Address: CERN Division PPE, Bat. 32 2C-14, CH-1211 Geneva 23, Switzerland
> -------------------------------------------------------------------------
|