[mythtv] XMLTV timestamps, TimeOffset and mythfilldatabase

[William Uther]

Hi all,
   Firstly, this is a non-urgent email.  If you're working hard on 0.18, 
put this aside until later :).

   I'm writing a new XMLTV grabber for Australia ( 
http://www.cse.unsw.edu.au/~willu/xmltv/ ).  In the process I came 
across some strangeness in the way Myth handles the time information in 
XMLTV data.  I propose to clean it up, but don't want to tread on any 
toes.

Here is the way things currently work:

In the XMLTV spec, a timestamp without a timezone spec is defined to be 
UTC.  In mythtv-setup you can set a TimeOffset to "None", "Auto", or 
another time offset.  This gets used in mythfilldatabase.

It is in mythfilldatabase that things get a little convoluted:

   - If "Auto" time offset is selected, then this is calculated _once_ 
based on current time when mythfilldatabse starts running.  (in 
parseFile())
   - Then, for each program:
     - The start and end time have the timezone offset added to their 
text (parseProgram() calls addTimeOffset()).  To achieve this:
       - If the TimeOffset setting is not "None" then:
         - addTimeOffset() parses the text representation into a 
QDateTime var
         - If there is no timezone spec on the timestamp then 
addTimeOffset() assumes that the timezone is equal to the TimeOffset 
setting.  This will cause the two to cancel, leaving the time treated 
as local time, and rendering the TimeOffset setting useless.
         - addTimeOffset() then adds the timezone spec and the 
timeoffset to the original timestamp and re-writes the textual 
timestamp in this 'local time'.
     - Finally, the start and end time texts are reparsed by a second 
time parser (parseProgram() calls fromXMLTVDate())
       - fromXMLTVDate() ignores any timezone spec.

This seems strange for a number of reasons:

   - If there is no timezone spec then the result is to read the 
timestamp as localtime, not UTC as in the XMLTV spec.
   - If there is no timezone spec then the TimeOffset value is ignored 
entirely.
   - Otherwise, if the TimeOffset is set to "None" then any timezone 
spec in the timestamp is ignored, but left in the text field.  In other 
cases the timezone spec is honoured, but stripped from the text.
   - If the TimeOffset is set to "Auto" then the same offset is used for 
all programs, even if the timezone changes in the middle of a grab 
(think about the week overlapping the start or end of daylight time).
   - The timestamp is parsed, converted back into text, and parsed 
again... using two different parsers.

I think we need these features:

   - A mechanism to ignore timezone offsets and just treat things as 
localtime: "None" TimeOffset setting
   - Timestamps follow the xmltv spec (i.e. missing timezone spec = UTC)
   - "Auto" TimeOffset setting should base the time offset on the time 
the program is showing, not the time mythfilldatabase is run
   - The timestamps should be parsed once by one piece of code
   - I'm not sure if the internal textual representation should have a 
timezone spec.  If it doesn't, then the internal textual representation 
is different to the xmltv representation (but that is not a big deal).  
If it does, then it does... I'm not sure where these are used anyway.

I propose to (In my spare time, some time):

   - Combine the addTimeOffset() and fromXMLTVDate() functions so we 
have one parser
   - Change the meaning of a missing timezone spec to mean UTC instead 
of local time (in line with the XMLTV spec).  Setting the TimeOffset 
value to "None" will allow people to achieve the old effect - the 
timestamp interpreted as local time.
   - Make "Auto" TimeOffset do the conversion to local time for each 
program.

Any thoughts, comments?

Will          :-}

--
Dr William Uther                           National ICT Australia
Phone: +61 2 9385 6357               Computer Science and Engineering
Email: willu at cse.unsw.edu.au          University of New South Wales
Web: http://www.cse.unsw.edu.au/~willu/     Sydney, Australia