C. Michael Pilato wrote:
> Max Bowsher wrote:
>> Martin Furter wrote:
>>> Hello
>>>
>>> Today svndumptool 0.4.0 has been released with many new features and
>>> bugfixes. The biggest feature since 0.3 is probably that it works with
>>> cvs2svn created dumps now (though better is to fix them first using the
>>> new command cvs2svnfix).
>> Hmm?
>>
>> Wouldn't better still be to fix cvs2svn?
>>
>> Oh. So, I just found the README.txt....
>> [[[
>>> Fixes a cvs2svn created dumpfile. Some (all?) versions of cvs2svn do not
>>> create 100% valid dumpfiles according to subversions specification of the
>>> dumpfile format. It omits the node kind for copied nodes. This command
>>> repairs those nodes.
>> ]]]
>>
>> Well, I'm inclined to alter the specification to state the Node-kind is
>> irrelevant, and should be omitted on write, and ignored on read, for a
>> copied node other than in the case of "Node-action: replace", in which
>> case the Node-kind specifies the kind of the *new* node, which may be
>> different from the copyfrom node.
>
> You'd rather change the spec to have a 4-line-long complex special-case
> condition description than to just fix cvs2svn, which appears to be
> intentionally omitting this header?
>
> (from dumpfile_delegate.py:DumpfileDelegate.copy_path():)
>
> # We don't need to include "Node-kind:" for copies; the loader
> # ignores it anyway and just uses the source kind instead.
No, I'd prefer to change to spec to avoid the disparity of having
Node-kind be defined to be the copyfrom kind for non-replace actions,
but the replacing kind for replace actions.
Max.
Received on Wed Apr 11 22:58:23 2007