SPH derived fields require using `ds.add_field`

[chrishavlin]

When you attempt to add an SPH derived field as a new global field, it is not identified as an SPH field:

import yt 

def _Ha_cloudy(field, data):
    return 2*data['PartType0','Density']


yt.add_field(
    name=("PartType0", "my_density"),
    function=_Ha_cloudy,
    sampling_type="particle",
)

ds = yt.load_sample("snapshot_033")
print(ds.field_info[('PartType0', 'my_density')].is_sph_field) # False

And any operations that should work for SPH fields fail (e.g., projection plots) despite ds._sph_ptypes containing PartType0.

If you define the new field just for the dataset, it is correctly identified as an SPH field:

import yt 

ds = yt.load_sample("snapshot_033")

def _Ha_cloudy(field, data):
    return 2*data['PartType0','Density']

ds.add_field(
    name=("PartType0", "my_density"),
    function=_Ha_cloudy,
    sampling_type="particle",
)
print(ds.field_info[('PartType0', 'my_density')].is_sph_field) # True

and everything works as usual.

I find this behavior confusing... and it's not obvious from the docs on whether this is the intended behavior.

Any thoughts on whether this a bug or just an undocumented behavior (that should be documented)?

From a quick look at the code, I'm guessing the issue arises because the DerivedField.ds is set to None for global fields so is_sph_field will always return False for global fields:

yt/yt/fields/derived_field.py

Lines 211 to 222 in e962538

	@property
	def is_sph_field(self):
	if self.sampling_type == "cell":
	return False
	is_sph_field = False
	if self.is_alias:
	name = self.alias_name
	else:
	name = self.name
	if hasattr(self.ds, "_sph_ptypes"):
	is_sph_field |= name[0] in (self.ds._sph_ptypes + ("gas",))
	return is_sph_field

[jzuhone]

@chrishavlin for SPH frontends "gas" is an alias for the SPH filed type (in this case "PartType0"), so what is normally done is that if we use yt.add_field we simply define new "gas" fields.

I would argue that since (using the current example) "PartType0" is a type proper to the frontend (just like "enzo", "flash", "gamer", etc. are field types proper to their frontends), we should in general not be adding derived fields for this field type (and similar ones), unless it's being done in that frontend's fields.py.

[chrishavlin]

pretty sure I tried it with "gas" as well and had the same issue... will check again.

[chrishavlin]

ya, still not recognized as sph.

import yt 

def _Ha_cloudy(field, data):
    return 2*data['PartType0','Density']

yt.add_field(
    name=("gas", "my_density"),
    function=_Ha_cloudy,
    sampling_type="particle",
)

ds = yt.load_sample("snapshot_033")
print(ds.field_info[('gas', 'my_density')].is_sph_field) # False

[jzuhone]

@chrishavlin if you try adding it as "gas", does the projection plot still fail (probably because it depends on is_sph_field)?

[chrishavlin]

Yup, fails with the usual non-sph particle error when trying to use a projection plot:

SlicePlot, ProjectionPlot, and OffAxisProjectionPlot can only plot fields that
are defined on a mesh or for SPH particles, but received the following N-body
particle fields:

    [('gas', 'my_density')]

Did you mean to use ParticlePlot or plot a deposited particle field instead?

The way is_sph_field is evaluated with the derived field relies on an internal reference to a ds attribute, which does not get set when creating the global derived field. Works fine when using ds.add_field though.

[chrishavlin]

but I'm getting the feeling that this is probably more of a bug than an intended behavior, ya? the derived field probably needs some better state management so that the ds reference gets passed down to the derived field when it's added to the dataset's available fields?