[OGR-GML] fids written by the GML driver are not read back correctly

[sebastien.grignard@…]

All fids written by the GML driver (F...) are ignored by the reader.

[sebastien.grignard@…]

This patch tries to detect if the GML fid was written by the OGR driver to assign it correctly

[warmerdam]

Sebastian,

I'm still not comfortable with the FID handling.  It seems like we could
get very screwy FID values in some situations, and non-unique fids could be
pretty bad karma. 

I'll leave this bug report open for the time being and think about it a bit. 

[warmerdam]

Chaitanya,

Could you please revise the patch to work properly against trunk, and to do some more careful evaluation of the FID values. In particular I would like:

• The FID values to be more completely scanned to ensure they are strictly a string of numeric digits, potentially prefixed by "F", and with no other characters, spaces, etc.
• Each time an FID value is extracted, iNextGMLId should be set to the maximum of the existing iNextGMLId value, and the FID value plus 1 - the objective here is to keep track of the next safe FID we could assign if we need to switch to serial fid assignment, without risk of conflicting with any existing FIDs.
• Keep a flag indicating whether we have encountered any illegal FIDs, in which case we will assign the FIDs serially even if the current FID seems legitimate.

If/when you are satisfied with these changes, please then extent autotest/ogr/ogr_gml_read.py to test these new capabilities and special cases, including introducing new (small) test files as needed.