Test 101 of the NCML handler's regression tests show the following error:
[nori:tests jimg$] besstandalone -c bes.conf -i test_101.bescmd | more
Here's what's in the test_101.bescmd (the ncml tests automatically generate the test.bescmd files, but errant tests save them in testsuite.dir/101/, ... and I copied it out and will attach it here):
And here's the NcML file (_ ../data/ncml/var_orgName_Structure.ncml_)
Test 97 shows a different, but almost certainly related error:
Here's the bescmd (also attached; test_97.bescmd)
The response _ besstandalone -c $abs_builddir/bes.conf -i ./test.bescmd_
Note that there are two variables in the response.
The response should only have one (see the baseline here baselines/var_orgName_Structure_constraint_1.ncml)
Here's a stab at what's going on. There error from test 101 is a bit easier to understand. The error message comes from libdap's CE expression parser here ce_expr_tab.cc and also in expr.yy at line 1134 (see below). I think what's happening is that the call to set_send_p() leaves the CE parser thinking that the CE has already been parsed (because it uses a check of send_p to determine if the CE was already parsed). SInce send_p was set, the code thinks the whole array has been projected - the attempt to set it to a different set of start, ..., values fails.
For test 97, I think that the call to send_p might be setting send_p for the whole Structure - that would explain why the two fields were sent.