promote the shifts flag to a first class theory flag

validphys2/src/validphys/commondataparser.py

@@ -128,8 +128,7 @@ class ValidApfelComb:
             mapping with the single fktables which need to be normalized and the factor
             note that when they are global factors they are promoted to conversion_factor
         - shifts:
-            mapping with the single fktables and their respective shifts
-            necessary to create "gaps" that some cfactors or datasets might expect
+            this flag is left here for compatibility purposes but has been moved to TheoryMeta
     """
 
     repetition_flag: Optional[list[str]] = None
@@ -143,6 +142,19 @@ class TheoryMeta:
 
     The theory metadata must always contain a key ``FK_tables`` which defines
     the fktables to be loaded.
+    The ``FK_tables`` is organized as a double list such that:
+
+    The inner list is concatenated
+    In practice these are different fktables that might refer to the same observable but
+    that are divided in subgrids for practical reasons.
+    The outer list instead are the operands for whatever operation needs to be computed
+    in order to match the experimental data.
+
+    In addition there are other flags that can affect how the fktables are read or used:
+    - operation: defines the operation to apply to the outer list
+    - shifts: mapping with the single fktables and their respective shifts
+              useful to create "gaps" so that the fktables and the respective experimental data
+              are ordered in the same way (for instance, when some points are missing from a grid)
 
     Example
     -------
@@ -161,19 +173,27 @@ class TheoryMeta:
     ... '''
     ... theory = yaml.safe_load(theory_raw)
     ... parse_input(theory, TheoryMeta)
-    TheoryMeta(FK_tables=[['fk1'], ['fk2', 'fk3']], operation='RATIO', conversion_factor=1.0, comment=None, apfelcomb=ValidApfelComb(repetition_flag=['fk3'], normalization=None, shifts=None))
+    TheoryMeta(FK_tables=[['fk1'], ['fk2', 'fk3']], operation='RATIO', shifts = None, conversion_factor=1.0, comment=None, apfelcomb=ValidApfelComb(repetition_flag=['fk3'], normalization=None))
 
     """
 
     FK_tables: list[list]
     operation: ValidOperation = "NULL"
     conversion_factor: float = 1.0
     comment: Optional[str] = None
+    shifts: Optional[dict] = None
     apfelcomb: Optional[ValidApfelComb] = None
     # The following options are transitional so that the yamldb can be used from the theory
     appl: Optional[bool] = False
     target_dataset: Optional[str] = None
 
+    def __post_init__(self):
+        """If a ``shifts`` flag is found in the apfelcomb object, move it outside"""
+        if self.apfelcomb is not None:
+            if self.apfelcomb.shifts is not None and self.shifts is None:
+                self.shifts = self.apfelcomb.shifts
+                self.apfelcomb.shifts = None
+
     def fktables_to_paths(self, grids_folder):
         """Given a source for pineappl grids, constructs the lists of fktables
         to be loaded"""

validphys2/src/validphys/pineparser.py

@@ -67,12 +67,7 @@ def pineko_apfelcomb_compatibility_flags(gridpaths, metadata):
         repetition_flag:
             grid_name
 
-    - shifts:
-        only for ATLASZPT8TEVMDIST
-        the points in this dataset are not contiguous so the index is shifted
 
-        shifts:
-            grid_name: shift_int
 
     Returns
     -------
@@ -108,12 +103,6 @@ def pineko_apfelcomb_compatibility_flags(gridpaths, metadata):
             if any(op in apfelcomb.repetition_flag for op in operands):
                 raise ValueError(f"The yaml info for {metadata['target_dataset']} is broken")
 
-    # Check whether the dataset has shifts
-    # NOTE: this only happens for ATLASZPT8TEVMDIST, if that gets fixed we might as well remove it
-    if apfelcomb.shifts is not None:
-        shift_info = apfelcomb.shifts
-        ret["shifts"] = [shift_info.get(op, 0) for op in operands]
-
     return ret
 
 
@@ -228,6 +217,14 @@ def pineappl_reader(fkspec):
 
     apfelcomb = pineko_apfelcomb_compatibility_flags(fkspec.fkpath, fkspec.metadata)
 
+    # Process the shifts (if any), shifts is a dictionary with {fktable_name: shift_value}
+    # since this parser doesn't know about operations, we need to convert it to a list
+    # then we just iterate over the fktables and apply the shift in the right order
+    shifts = None
+    if (shift_info := fkspec.metadata.shifts) is not None:
+        fknames = [i.name.replace(f".{EXT}", "") for i in fkspec.fkpath]
+        shifts = [shift_info.get(fname, 0) for fname in fknames]
+
     # fktables in pineapplgrid are for obs = fk * f while previous fktables were obs = fk * xf
     # prepare the grid all tables will be divided by
     if hadronic:
@@ -256,8 +253,9 @@ def pineappl_reader(fkspec):
                 raw_fktable = raw_fktable[0:1]
                 n = 1
                 protected = True
-            if apfelcomb.get("shifts") is not None:
-                ndata += apfelcomb["shifts"][i]
+
+        if shifts is not None:
+            ndata += shifts[i]
 
         # Add empty points to ensure that all fktables share the same x-grid upon convolution
         missing_x_points = np.setdiff1d(xgrid, p.x_grid(), assume_unique=True)