Bug 1803, Upcoming breaking changes for 4.10.0-beta1

[Steve Lime]

Umberto: How do you add classes and layers at runtime? In the context of
a layer or a map
correct? I don't believe that what Tamas is proposing would change that
capability at all. 

If you look at the list:

classObj.layer
webObj.map
legendObj.map
layerObj.map

are back references and you don't want folks monkeying around with
them.

classObj.label
legendObj.label
mapObj.scalebar
mapObj.legend 
mapObj.querymap
mapObj.web
mapObj.symbolset
mapObj.labelcache
mapObj.reference

Already exist in their parent objects so there is no need to create new
ones in place of the
existing ones. Users should get the reference to the existing structure
and modify it in place.
I'm not sure that given the potential problems that it even makes sense
to have constructors
for these. I assume this is the contentious spot...

(If there is a glaring need to have, for example, an unattached 
labelObj laying around then 
couldn't a populateFrom capability be developed for a labelObj, or
perhaps setLabel() for a
classObj?)

classObj.metadata
fontSetObj.fonts
mapObj.configoptions
webObj.metadata

Are hash tables and Sean created a nice interface for them to use.
Modifying directly shouldn't
be allowed. I think this is what was ok'd yesterday.

Note sure about  imageObj.format...

I guess I don't see the big deal with this change. My 2 cents. 

Steve

>>> Umberto Nicoletti <umberto.nicoletti at GMAIL.COM> 8/17/2006 8:25 AM
>>>
On 8/17/06, Daniel Morissette <dmorissette at mapgears.com> wrote:
> Tamas Szekeres wrote:
> >
> >> I think you can safely hide the labelPathObj, that should not be
> >> exposed. I think the others
> >> should be immutable too (assuming I understand the problem
correctly),
> >> the question is how
> >> many scripts will the change break and should backward
compatability
> >> breakage be limited
> >> to major releases (e.g. 5.0).
> >>
> >
> > Those scripts are *erroneous* !  It would be desirable to force
the
> > users to put them into a good shape as soon as possible.
> >
> >
>
> It seems that I should take position as release manager but
> unfortunately I do not know or use SWIG MapScript much so I have to
rely
> on those who know to make an opinion on whether the current stuff is
> dangerous enough to warrant breaking a few scripts with 4.10.
>
> Based on the understanding I have of the problem from reading this
> thread and previous discussions, it seems to me that if those
scripts
> are doing something that is just waiting to bomb then changing SWIG
> MapScript in v4.10 to make those object references immutable is not
a
> backwards compatibility issue, it is a bugfix and a service we are
> making to those users by forcing them to fix their scripts... and in
the
> end their apps will just be more robust.
>

Daniel,
IMO  we are talking about fixing pieces of code we are not even sure
they are broken. For instance I have some Java mapscript tests (and an
application) running just fine that add layers and classes at run time
and that's why I am so reluctant in approving this proposal. Only once
we get a test case that reproduces this errors reliably then we can go
and change the code.

Unfortunately I couldn't join the IRC meeting because I am on vacation
otherwise I'd have spoken  earlier (I am writing this on the road).

Regards,
Umberto

> The real question that I would throw at the SWIG MapScript experts
is:
> "Is the practice of overwriting those object references really
dangerous
> or not?"
>
> If the answer is yes then I think that solves the question and we
need
> to apply the fix and force users to fix their scripts.
>
> If there are object references that can safely be overwritten (such
as
> colorObj) then my opinion is that we should not touch them in 4.10.
>
> Daniel
> --
> Daniel Morissette
> http://www.mapgears.com/ 
>