[Idr] RFC-compliant use of Extended Length for new BGP attributes

I  am  interested  in input from the list on loose behavior we've seen
from various BGP specs.  The specific example in question is a case we
are seeing in the wild where a NEW_AS_PATH attribute has 8 data bytes,
but  sets the Extended Length bit and uses a 2-byte length field.  RFC
1771 explicitly states "Extended Length may be used only if the length
of  the  attribute value  is  greater  than 255 octets."  However, the
superceding  RFC  4271 does not have this explicit language, therefore
would  imply that  either  of the following cases would technically be
compliant:

A.  Ext-Len is set (1), 2-byte length field is used, regardless of
length of attribute

B.    Ext-Len   is   clear  (0),  1-byte  length  field  is  uses,
and attribute is <= 255 bytes

Furthermore, I've      seen      loose      language,      such     as
draft-kumaki-pce-bgp-disco-attribute-00.txt  that  seems  to  imply to
always set Ext-Len bit regardless.

Specifically,   is  anyone  aware  of  any  new  BGP  extensions  that
would REQUIRE  Ext-Len  to  be set, even if attribute is <= 255 bytes?
Our  thought  is  that if we receive a case A above with length <= 255
bytes,  we  are  free to propogate to downstream neighbors as a case B
(generically, for any attribute in BGP)."

-- 
-m

====================================================================
Key fingerprint = 8E48 6CD0 6D2B 538E 264B  D1D0 21E2 E7EE A40F 2A0B
gpg --keyserver wwwkeys.pgp.net --recv-keys 0xA40F2A0B
====================================================================