[cvsnt] Re: Keyword 'properties' breaks rcs keyword ordering
tmh at nodomain.org
Mon Jun 13 22:44:37 BST 2005
Bertho Stultiens wrote:
>>Tony Hoyle write:
>> notes:@fixes a bug in the widgets code; to test.@
>> to_test:@development team 3@;
> Actually, this is quite acceptable for CVS. There has always been some
> problems with the definitions in the format and that is why most parsers
> take a liberal approach. They parse each newphrase part as:
> <newkeyword> <junktext> ;
> Where junktext may contain embedded string encapsulations (using @). The
> parser simply reads until it encounters the first unescaped semicolon.
> If I read the cvsnt code correctly, then it uses a similar method to
> circumvent the ambiguities of the original rcsfile definition. This method
It wasn't that flexible... it only expected '@' at the start and end of
the text, with embedded '@@' characters. The above text would break an
old cvs/cvsnt wherever it was put.
> has been established (silently) for many years. Note that even cvsnt defines
> the 'properties' extension using the same established format. My only
> problem is its position in the grammatical sense.
I've now moved it (and due to the way cvsnt writes RCS files this works
retroactively).. at the moment it's not really an issue as only the
testing builds really have a way to set properties (using cvs -s during
commmit) and the stable builds just have the code to handle them so they
get preserved (there's a 'cvs admin' call but it's really just for testing).
More information about the cvsnt