GMTDataArrayAccessor: Fallback to default grid registration and gtype if the grid source file doesn't exist

[seisman]

Description of proposed changes

If the netCDF source file doesn't exist, running grdinfo on the netCDF source file gives the following error:

grdinfo [ERROR]: Cannot find file /tmp/xxx/xxx.nc
grdinfo [ERROR]: Must specify one or more input files

The error is unfriendly and confusing. Actually, it makes no sense if the grid source file doesn't exist. In this case, it should fallback to registration=0 and gtype=0 unless we implement the idea in #2194 (comment).

Partially address #1984 and #524 (comment)

Please note that this PR doesn't solve the problems in #1984 and #524, because the registration and gtype information are incorrect, but at least it doesn't report confusing errors.

[seisman]

Ping @maxrjones and @weiji14 for reviewing.

[seisman]

I've added a unit test in 37869d4, and the test helps find a new bug.

At the end of the test, I manually set the registration and gtype information and expect the two assert statements to pass, but they don't. So there must be another bug.

 # manually set the registration and gtype
dataset.height.gmt.registration = 1
dataset.height.gmt.gtype = 1
# the registration and gtype should be correct now.
assert dataset.height.gmt.registration == 1
assert dataset.height.gmt.gtype == 1

[seisman]

I've added a unit test in 37869d4, and the test helps find a new bug.

At the end of the test, I manually set the registration and gtype information and expect the two assert statements to pass, but they don't. So there must be another bug.

 # manually set the registration and gtype
dataset.height.gmt.registration = 1
dataset.height.gmt.gtype = 1
# the registration and gtype should be correct now.
assert dataset.height.gmt.registration == 1
assert dataset.height.gmt.gtype == 1

The documentation of xarray accessor explains why they don't work as expected:

Accessors are created once per DataArray and Dataset instance. New instances, like those created from arithmetic operations or when accessing a DataArray from a Dataset (ex. ds[var_name]), will have new accessors created.

So every time dataset.height.gmt is used, the __init__ function is called. It's the limitation of the xarray accessors and I think we can do nothing here.

[weiji14]

The documentation of xarray accessor explains why they don't work as expected:

Accessors are created once per DataArray and Dataset instance. New instances, like those created from arithmetic operations or when accessing a DataArray from a Dataset (ex. ds[var_name]), will have new accessors created.

So every time dataset.height.gmt is used, the __init__ function is called. It's the limitation of the xarray accessors and I think we can do nothing here.

The accessor dataset.height.gmt is created once, but the variable or property should be able to be modified using a setter method. Still think this is an issue, but need time to look into it. Don't we have some tests that actually modify the accessor values?

[seisman]

I've added a unit test in 37869d4, and the test helps find a new bug.
At the end of the test, I manually set the registration and gtype information and expect the two assert statements to pass, but they don't. So there must be another bug.

 # manually set the registration and gtype
dataset.height.gmt.registration = 1
dataset.height.gmt.gtype = 1
# the registration and gtype should be correct now.
assert dataset.height.gmt.registration == 1
assert dataset.height.gmt.gtype == 1

The documentation of xarray accessor explains why they don't work as expected:

Accessors are created once per DataArray and Dataset instance. New instances, like those created from arithmetic operations or when accessing a DataArray from a Dataset (ex. ds[var_name]), will have new accessors created.

So every time dataset.height.gmt is used, the __init__ function is called. It's the limitation of the xarray accessors and I think we can do nothing here.

The documentation of xarray accessor explains why they don't work as expected:

Accessors are created once per DataArray and Dataset instance. New instances, like those created from arithmetic operations or when accessing a DataArray from a Dataset (ex. ds[var_name]), will have new accessors created.

So every time dataset.height.gmt is used, the __init__ function is called. It's the limitation of the xarray accessors and I think we can do nothing here.

The accessor dataset.height.gmt is created once, but the variable or property should be able to be modified using a setter method. Still think this is an issue, but need time to look into it. Don't we have some tests that actually modify the accessor values?

OK. I've removed this test and added a simpler test instead. The old complicated test was moved to #1984 so that we can work on this later.

[seisman]

I wonder if we should change these lines to the following codes, which I think are more readable:

 self._source = self._obj.encoding.get("source")
if self._source is not None and Path(self._source).exists():
try:
# Get grid registration and grid type from the last two columns of
# the shortened summary information of `grdinfo`.
self._registration, self._gtype = map(
int, grdinfo(self._source, per_column="n").split()[-2:]
)
except ValueError:
self._registration = 0
self._gtype = 0
else:
self._registration = 0
self._gtype = 0

[weiji14]

So we will silently fallback to registration=0 and gtype=0 if grdinfo cannot determine the grid registration and type? Ok I guess, but please add the code comments back 🙂

[seisman]

Ping @GenericMappingTools/pygmt-maintainers for review.

[weiji14]

Sorry, 2 more typos, otherwise ok.

[michaelgrund]

Only minor consistency changes, otherwise looks good!