John Swinbank added a comment - 27/Jul/15 10:53 AM Forgive my naivety, but on first reading I'm not sure that I follow. You mention that "older version table is read in". Do you mean that this test relies on some old table that has been saved in afwdata or equivalent, and that breaks? Or do you just mean that when you persist a new ChebyshevBoundedField it makes an old-style table that you can't then successfully read back? In the latter case, I guess this is pretty straightforward fix (I guess it just needs a few . s to be replaced by _ s in ChebyshevBoundedField.cc ). I'd suggest filing a new ticket for it, but would imagine it could be turned around quickly. If there's some deeper backwards compatibility issue here, that might be a different matter.

Russell Owen added a comment - 27/Jul/15 11:16 AM - edited An existing unit test from HSC triggers this issue (see below). The test intentionally tries to unpersist an ExposureInfo table that uses an older table format. The test fails while trying to unpersist a ChebyshevBoundedField, which is presumably part of an ApCorrMap. Jim Bosch understands the details better than I do, but I think what is happening is that when afw table reads an older format of data it auto-converts field separators "." into "_". That fails for ChebyshevBoundedField because it incorrectly uses "." instead of "_". In any case it is a certainty that ChebyshevBoundedField should use "_" instead of "." as its field separator, and the current code is incorrect (probably because that was not changed when it was brought over from HSC, which uses the older field separator). tests/testValidPolygon.py   E: Empty WCS extension, using FITS header .... ====================================================================== ERROR: testExposureCatalogBackwardsCompatibility (__main__.ValidPolygonTestCase) Test that we can read an ExposureCatalog written with an old version of the code. ---------------------------------------------------------------------- Traceback (most recent call last): File "tests/testValidPolygon.py", line 96, in testExposureCatalogBackwardsCompatibility cat2 = afwTable.ExposureCatalog.readFits(filename2) File "/Users/rowen/LSST/lsstsw/build/afw/python/lsst/afw/table/tableLib.py", line 11090, in readFits return _tableLib.ExposureCatalog_readFits(*args) NotFoundError: File "src/table/Schema.cc", line 239, in SchemaItem<T> lsst::afw::table::detail::SchemaImpl::find(const std::string &) const [T = int] Field or subfield withname 'order.x' not found with type 'I'. {0} File "src/table/io/InputArchive.cc", line 105, in boost::shared_ptr<Persistable> lsst::afw::table::io::InputArchive::Impl::get(int, const lsst::afw::table::io::InputArchive &) loading object with id=2, name='ChebyshevBoundedField' {1} File "src/table/io/InputArchive.cc", line 105, in boost::shared_ptr<Persistable> lsst::afw::table::io::InputArchive::Impl::get(int, const lsst::afw::table::io::InputArchive &) loading object with id=1, name='ApCorrMap' {2} lsst::pex::exceptions::NotFoundError: 'Field or subfield withname 'order.x' not found with type 'I'. {0}; loading object with id=2, name='ChebyshevBoundedField' {1}; loading object with id=1, name='ApCorrMap' {2}'     ---------------------------------------------------------------------- Ran 5 tests in 0.049s   FAILED (errors=1)

Russell Owen added a comment - 27/Jul/15 4:21 PM Fixed as a trivial change to ChebyshevBoundedField.cc in afw on tickets/ DM-2981 . A unit test on that branch failed before the fix (due to this problem) and worked afterwards. Paul Price reviewed the change as part of reviewing DM-2981 .