[Qgis-developer] WFS feature IDs: when required, when not?

[Bill Clay]

All,

Thanks to Andrea Peri, I have just discovered that WFS 1.0.0 apparently 
does NOT require a WFS server to report a unique feature ID with every 
feature it transmits (a typical newbie misconception?).

The OGC specs are so nested and versioned, it's hard to be certain I've 
understood them correctly.  Could someone be kind enough to enlighten me 
on the following?

1. Can you confirm or correct the following understandings:

a. Every WFS server (versions 1.0.0 and 1.1.0) must have a permanent 
unique identifier for every feature.

b. WFS GetFeature responses version 1.0.0 may or may NOT provide a 
unique "fid" attribute with each <feature> element, provided the layer 
is not editable (WFS-T).

c. WFS GetFeature responses version 1.1.0 MUST provide a unique "fid" 
attribute with each <feature> element.

2. Are you aware of any common implementation of WFS 1.0.0 that does NOT 
always report a "fid" attribute with every <feature> element?  (I 
understand TinyOFS can be configured not to do so.)

3. Do you believe that WFS services that do not always provide a "fid" 
with every feature are unusual enough that the QGIS WFS client can 
simply disable all feature caching for such servers?

The proposal at item 3 would require GetFeatures to be requested for the 
entire canvas extent every time any previously un-fetched area is 
exposed on the canvas.  Practically speaking, this means potentially 
long delays on every pan and zoom-out on maps containing WFS layers with 
many features that are hosted by such servers.

Doubtless this is old news to everyone but me.  Sorry for the static.

Bill Clay