thrift-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From james anderson <>
Subject Re: where is an accurate bnf?
Date Sat, 03 Jul 2010 21:54:36 GMT

On 2010-07-03, at 22:55 , Michael Walsh wrote:

> Hi James,
> The official code repo is on apache svn. The current release is    
thrift.bnf has the same internal modification date as the git clone.

> I don't know why we have any git resources at all as they just seem  
> to keep confusing people.

the points of confusion are

1. binary encodings appear to stretch the proviso, that "the order of  
elements could be in some cases rearranged" to eliminate the name  
constituent from the field-begin entirely. perhaps because, without  
reordering the clauses, it would be impossible to recognize a binary  

2. the message-type, as transposed to be the first constituent of a  
message-begin appears to actually itself have two constituents: the  
protocol identifier and the protocol version number. is the protocol  
identifier domain documented?

3. the field-type names appear, but between the slee, etal paper, the  
wiki, and the code (at least from reading the cpp, java, and ruby  
implementation) the domains for t_string and t_binary are never  
conclusively specified. t_string implies at base unsigned 8-bit,  
while descriptions of t_binary imply signed.

4. is there a specification for standard encodings for the various  
symbolic indicators for binary protocols?

> M
> Sent from my iPhone
> On 3 Jul 2010, at 19:11, james anderson <>  
> wrote:
>> good evening;
>> the git-hub mirrored bnf appears inaccurate at least in the  
>> message-begin production.
>> is there a more up-to-date version?

View raw message