[mythtv] Version numbers for services

David Engel david at istwok.net
Mon Dec 13 19:27:23 UTC 2021


On Sun, Dec 12, 2021 at 03:02:46AM +0000, Gary Buhrmaster wrote:
> On Sun, Dec 12, 2021 at 1:15 AM Peter Bennett <pb.mythtv at gmail.com> wrote:
> >
> > The old services had various version numbers. See below. In the new
> > services all have defaulted to version 1.0. Any recommendations on what
> > we should do with the new services?
> >
> > Set all versions same as the old
> > Set them all to 2.0
> > Set them all to 20.0
> > Set them to previous version + 0.1
> > Other?
> 
> Well, since you asked, here is an email that I sent to
> a dev who asked for a unicasted email regarding the
> entire services API issues, which touched upon
> API versioning (I do not know if that dev forwarded
> my comments to others).....
> 
[...]

> * The services API is far too low level.  In
>   most cases they are basically just DB updates.
> 
>   Ideally, additions/updates/replacements will
>   consider the needed functionality and not
>   be just a fancy mysql interface as they are
>   currently.

As I recall, the capture card API is one such case.  I know I cringed
when I first saw it.  It needs a sever overhaul.  I'm sure there are
other areas that are the same.

> * Services API(s) do not implement business logic
> 
>   To first approximation, none of the Services APIs implement
>   the business logic that MythTV Setup performs, allowing
>   the storage of values in database tables that are not
>   validated/valid, and can lead to issues.
> 
>   The business logic should be added such that
>   applications cannot mangle the MythTV database
>   by accident, or intentionally.

I put business logic in the recording rule API.  The intent was to
make it impossible to create invalid rules.

> * A number of the variable names are different when
>   retrieved vs the name that needs to be
>   set/updated.
> 
>   I have had to create a translation from the
>   returned names to the names to be updated
>   for my apps.  Sometimes the difference is
>   capitalization, sometimes the name is
>   related, but different (ChanNum vs
>   ChannelNumber for the Channel service).
>   Pick a name (any name) and stick with it.

Yuck.  I suppose compatibility could be preserved by changing the the
write APIs to accept the read names too when they differ.  That could
cause the internal structs to have both names, though, which would
also be very ugly.  Does (Could?) the new, API infrastructure have
support for some type of aliass where the same column could be written
using two or more names?

> * Service API(s) do not support creating dtv_multiplex entries
> 
>   While retrieving a Channel definition allows one to
>   retrieve the dtv_multiplex values, there is no way to
>   create new multiplexs or to update a dtv_muliplex.
>   If the Services API is intended to be used to replace
>   MythTV Setup there needs to be a way to create
>   dtv_multiplex entries (and likely the dtv_privatetypes)
>   if the application has it available to store.

I'm not sure how to handle dtv_multiplex, and the diseqc tree as well.
My initial reaction has always been that the multiplex parameters
(modulation, frequenty, etc.) should be part of the channel parameters
made visible throught the API.  MythTV would then magically manage
dtv_multiplex behind the scene.  My fear is that if it's not done
really, really, really well, with copious amounts of business rules,
it might lead to a very messed up dtv_multiplex.

> * Service API(s) do not support setting of a NULL value
> 
>   Json based encoding allows the specification of a
>   form such as {"key": null}.  However, the Services APIs
>   that update database columns such that the column
>   allows a DB NULL do not store NULL in that column
>   if passed the null value.
> 
>   As long as MythTV code differentiates between
>   NULLs and other values, the Services API needs
>   to allow those values to be properly evaluated and
>   stored.
> 
>   Bonus points, in addition to the bare 'null'
>   value, there is also 'true' and 'false', and while
>   most of MythTV stores such as an integer
>   1 or 0, supporting 'true' and 'false' barewords
>   might be a good idea for appropriate columns.

I don't have a good suggestion.  Just maybe use "special" values like
"<null>" when they make sense and "special" boolean columns like
SetOtherClumnNULL=true" when they don't.

> * Service API(s) do not store an empty string
> 
>   Json based encoding allows the specification of a
>   form such as {"key": ""}, however there are cases
>   where such are considered as missing, rather than
>   explicit, and are not actually set (or are set as the
>   default value, whatever that is).

In my latest API changes, which was some time ago, I started checking
a special map variable which indicated which values were actually
supplied.  I think that could work.  If the map variable says ColumnX
was supplied and it's values is an empty string, then that's what the
user wants.

> * The Services APIs should be versioned, and
>   backward compatibility required for returning only
>   the values (and setting values) that were in the
>   requested version.  Such backward compatibility
>   should be maintained for at least one previous
>   version to allow apps to migrate across supported
>   versions and master.
> 
>   Without versioning, certain apps may not be
>   able to even know that the (new) values they
>   get are important.
> 
>   Whether the API itself should take a version
>   number in the request ...?v19, or the endpoint
>   URL should be versioned (.../v19/ is up to the
>   implementer.  The important issue for apps
>   using the services API will continue to work
>   across at least the supported version and
>   master.

I don't have a strong preference on which versionaing method to use
other than we should try to be consistent.

>   Admittedly it is not always clear how the
>   BE service will need to adapt to a previous
>   versions request, but in cases where the BE
>   can do a "reasonable" thing, it should.
>   If the services API cannot support a previous
>   version (substantial changes, too old, ???)
>   it should return an appropriate error code
>   (409?).
> 
>   Thought experiment, an unversioned query
>   to determine what version(s) the service
>   supports so an app can respond according
>   to any common version that they both support
>   (or reject the processing)???
> 
> * Any Services APIs that take arguments for changes
>   should leave existing values alone if not replaced
> 
>   If, at some point in the future, the service API
>   has an added value, applications may not be
>   aware of that, and since the services API is
>   not versioned, it may not return the existing
>   value, so you may end up with resetting values.
> 
>   A few of the recent updates to the services API
>   have done this (good!), but there are a number
>   of services for which one must return all the
>   existing values, otherwise they will be updated
>   with the defaults.

Agreed.  I think I did the following on my latest change:

    Read old values from DB.

    Apply changes included in the API request.

    Sanity check the result and return an error if not sane.

    Save the changes if sane.

> * When one does a AddDBChannel one has to specify
>   a new (available) ChannelID.
> 
>   That requires one to first look for an available
>   number.  The ChannelID should be treated as
>   an opaque value, and auto incremented when
>   created by AddDBChannel (which should
>   probably return the ChannelID created?).

That's an unfortunate choice for doing add.  Add should return the new
ID that is automatically created.  For backward compatibility, use the
provided ID if the channel doesn't already exist, otherwise, return an
error.

David
-- 
David Engel
david at istwok.net


More information about the mythtv-dev mailing list