xmlquery issue when values are set to match another variable

[bartgol]

I'm getting this error with --listall (or when query the HIST_N variable):

$ ./xmlquery HIST_N
ERROR: Entry HIST_N was listed as type int but value '$STOP_N' is not valid int
$ ./xmlquery --listall
ERROR: Entry HIST_N was listed as type int but value '$STOP_N' is not valid int

Did something change in CIME related to xmlquery? I don't think we recently changed the default for HIST_N, so I'm guessing this is a new thing? I'm using the tag cime6.0.226 (4388509).

[jedwards4b]

This should work and recursively give you the value of STOP_N. I just tried in the latest cime master and it works as expected. (cime6.0.250)

[bartgol]

Ok, then maybe it's simply time to update the cime submodule in our fork. Thanks for checking!

Edit: uhm, I tried to checkout cime6.0.250, and still got the same error. @jgfouca could there be something funky in e3sm?

[jgfouca]

It may be E3SM-only. I've assigned @jasonb5

[bartgol]

I may have a suspect. In generic_xml.py, we have

        for m in reference_re.finditer(item_data):
            var = m.groups()[0]
            print (f"found m={m}, var={var}")
            logger.debug("find: {}".format(var))
            # The overridden versions of this method do not simply return None
            # so the pylint should not be flagging this
            # pylint: disable=assignment-from-none
            ref = self.get_value(var, subgroup=subgroup)
            ...

However, the referenced var may be in a different subgroup. E.g., in my case, HIST_N is in the group run_drv_history, while STOP_N is in run_begin_stop_restart. Removing the subgroup arg from the get_value call, makes things work. However, that means that the referenced var must be unique across the XML file, which may be a restriction.

[rljacob]

I just tried ./xmlquery HIST_N on latest e3sm master with a case built back on May 14 and it works for me.

[jgfouca]

@rljacob , try this case: SMS.ne256pg2_ne256pg2.F2010-SCREAMv1

[rljacob]

Still works. See /home/jacob/E3SM/cime/scripts/Fscream on Chrysalis.

[jgfouca]

Now I'm really confused.

[rljacob]

Is it an old case you're trying to query? If you're doing that with a newer CIME, it may not work.

[jgfouca]

I just tried again, making sure i was on lastest master with subs updated, same error. This is on mappy.

[rljacob]

Oh weird. I tried actually doing "create_test" and going to the casedir and now I see the error.
Try making your case with: ./create_newcase --case Fscream --compset F2010-SCREAMv1 --res ne256pg2_ne256pg2 and going to the casedir it creates.

[jedwards4b]

It will only occur with create_test and not with create_newcase since the setting of
HIST_N=$STOP_N happens in config_tests.xml

[rljacob]

I went to a test directory created in our overnight testing and it was fine. If it wasn't, all the tests would fail since there would be no coupler history output without changing HIST_N from its default.

[jedwards4b]

I don't think that's the case - you can set the value of HIST_N - the reported problem is that you cannot query the value after setting HIST_N=$STOP_IN

[jedwards4b]

Is that overnight test directory on mappy - where they are apparently having problems? If so can you try ./xmlquery HIST_N in one of those ERS tests?

[rljacob]

Oh I see what you're saying. Its always the case that HIST_N is set as <HIST_N>$STOP_N</HIST_N> in a test. But ./xmlquery is working in at least one testdir I tried.

[jgfouca]

To be clear, all the basics are running fine. It's only xmlquery for specific items that's the problem.

[rljacob]

No I'm trying Chrysalis. I've seen ./xmlquery HIST_N give an error on a test dir I just created but it works fine in a test dir created by the overnight testing. On the same machine.

[rljacob]

It must be something about that case and the SMS test. I tried ./create_test --no-build ERP.ne256pg2_ne256pg2.F2010-SCREAMv1 and ./xmlquery HIST_N worked..

[bartgol]

It must be something about that case and the SMS test. I tried ./create_test --no-build ERP.ne256pg2_ne256pg2.F2010-SCREAMv1 and ./xmlquery HIST_N worked..

For ERS/ERP tests, I suspect HIST_N is set to an integer, and not $STOP_N?

[jedwards4b]

Not true: https://github.com/ESMCI/cime/blob/master/CIME/data/config/config_tests.xml#L270

[rljacob]

By default, STOP_N is 5 for SMS and 11 for ERS.

[bartgol]

I can confirm that also on my laptop ./xmlquery HIST_N works with an ERP test, but fails with an SMS one. My theory above regarding HIST_N and STOP_N being in different groups must be wrong (or incomplete).

[jasonb5]

@bartgol The issues is with HIST_N and STOP_N being in different groups. The reason ERS, ERP, IRT, etc continue work is due to it having STOP_N defined in env_test.xml. Any test with <HIST_N>$STOP_N</HIST_N> and <STOP_N>...</STOP_N> defined in config_tests.xml will work still. But anything relying on $STOP_N being populated from somewhere else will fail, e.g. PEA, PEM, PET, etc.

I'm thinking the fix is probably making sure values not provided in config_tests.xml are given a default when the file is generated. Changing the grouping could work too, might be simpler than the first. I don't think we could remove the subgroup from

cime/CIME/case/case.py

Lines 553 to 557 in 67ade4a

	item = env_file.get_resolved_value(
	item,
	allow_unresolved_envvars=allow_unresolved_envvars,
	subgroup=subgroup,
	)

since it'd probably re-introduce the origin issue from #4572.

[bartgol]

What if we allowed to specify indirect parameters via "scoping"? I'm thinking something like setting HIST_N to $run_begin_stop_restart::STOP_N. We'd still process $-containing strings as params to be resolved, but we'd add an extra step:

        for m in reference_re.finditer(item_data):
            var = m.groups()[0]
            logger.debug("find: {}".format(var))
            # The overridden versions of this method do not simply return None
            # so the pylint should not be flagging this
            # pylint: disable=assignment-from-none
            tokens = var.split('::')
			if len(tokens)==2:
               ref = self.get_value(tokens[0], subgroup=tokens[1])
			else:
               ref = self.get_value(var, subgroup=subgroup)

I used the :: token since I'm thinking about C++ scopes, but any other "safe" string would do (double underscore, @, ...)

[github-actions]

This issue is stale because it has been open 90 days with no activity. Remove stale label or comment or this will be closed in 5 days.