make BudgetItem.freq a property

This makes BudgetItem frequency arrays properties instead of attributes. As an attribute, BudgetItem.freq can be set directly but this doesn't update the precomp cache or any of the frequency arrays of the cal_objs and noise_objs for instances of Budget. This is a problem if run is called without any keyword arguments after changing freq. Now that freq is a property, the setter clears the precomp cache and recurses through the cal_objs and noise_objs if applicable.

So this works correctly now

traces1 = budget.run(freq=freq1)
budget.freq = freq2
traces2 = budget.run()

gwinc/__init__.py

@@ -44,8 +44,8 @@ def freq_from_spec(spec=None):
         return spec
     elif spec is None:
         spec = DEFAULT_FREQ
-    fspec = spec.split(':')
     try:
+        fspec = spec.split(':')
         if len(fspec) == 2:
             fspec = fspec[0], DEFAULT_FREQ.split(':')[1], fspec[1]
         return np.logspace(
@@ -53,7 +53,7 @@ def freq_from_spec(spec=None):
             np.log10(float(fspec[2])),
             int(fspec[1]),
         )
-    except (ValueError, IndexError):
+    except (ValueError, IndexError, AttributeError):
         raise InvalidFrequencySpec(f'Improper frequency specification: {spec}')
 
 
@@ -154,7 +154,7 @@ def load_budget(name_or_path, freq=None, bname=None):
     mod, modpath = load_module(modname)
     Budget = getattr(mod, bname)
     if freq is None:
-        freq = getattr(Budget, 'freq', None)
+        freq = getattr(Budget, '_freq', None)
     freq = freq_from_spec(freq)
     ifopath = os.path.join(modpath, 'ifo.yaml')
     if not ifo and os.path.exists(ifopath):

[Jameson Rollins]

how do you think an AttributeError would be raised here? this should only ever be called on strings, and I don't think any of their attributes are relevant to the parsing here.

[Kevin Kuns]

If spec is not np.ndarray and is not None, then spec is not a string when entering the try block. spec.split(':') raises an AttributeError in this case. This was also previously the case and I fixed it this way when updating the rest of the frequency specification. This could also be solved by adding an else case after the case that spec is None. Open to other suggestions.