Improved navigation and clarity in details view.

This commit is contained in:
lwark
2024-09-13 15:23:11 -05:00
parent 744394f236
commit c7d83401e7
15 changed files with 270 additions and 177 deletions

View File

@@ -168,7 +168,8 @@ class BasicSubmission(BaseClass):
'extraction_info', 'comment', 'barcode',
'platemap', 'export_map', 'equipment', 'tips'],
# NOTE: Fields not placed in ui form
form_ignore=['reagents', 'ctx', 'id', 'cost', 'extraction_info', 'signed_by', 'comment', 'namer', 'submission_object', "tips"] + recover,
form_ignore=['reagents', 'ctx', 'id', 'cost', 'extraction_info', 'signed_by', 'comment', 'namer',
'submission_object', "tips", 'contact_phone'] + recover,
# NOTE: Fields not placed in ui form to be moved to pydantic
form_recover=recover
)
@@ -229,7 +230,8 @@ class BasicSubmission(BaseClass):
return SubmissionType.query(cls.__mapper_args__['polymorphic_identity'])
@classmethod
def construct_info_map(cls, submission_type:SubmissionType|None=None, mode: Literal["read", "write"]="read") -> dict:
def construct_info_map(cls, submission_type: SubmissionType | None = None,
mode: Literal["read", "write"] = "read") -> dict:
"""
Method to call submission type's construct info map.
@@ -242,7 +244,7 @@ class BasicSubmission(BaseClass):
return cls.get_submission_type(submission_type).construct_info_map(mode=mode)
@classmethod
def construct_sample_map(cls, submission_type:SubmissionType|None=None) -> dict:
def construct_sample_map(cls, submission_type: SubmissionType | None = None) -> dict:
"""
Method to call submission type's construct_sample_map
@@ -609,7 +611,7 @@ class BasicSubmission(BaseClass):
new_dict = {}
for key, value in dicto.items():
# logger.debug(f"Checking {key}")
missing = value is None or value in ['', 'None']
missing = value in ['', 'None', None]
match key:
case "reagents":
new_dict[key] = [PydReagent(**reagent) for reagent in value]
@@ -628,7 +630,7 @@ class BasicSubmission(BaseClass):
case "id":
pass
case _:
# logger.debug(f"Setting dict {key} to {value}")
logger.debug(f"Setting dict {key} to {value}")
new_dict[key.lower().replace(" ", "_")] = dict(value=value, missing=missing)
# logger.debug(f"{key} complete after {time()-start}")
new_dict['filepath'] = Path(tempfile.TemporaryFile().name)
@@ -648,7 +650,7 @@ class BasicSubmission(BaseClass):
return super().save()
@classmethod
def get_regex(cls, submission_type:SubmissionType|str|None=None):
def get_regex(cls, submission_type: SubmissionType | str | None = None):
# logger.debug(f"Attempting to get regex for {cls.__mapper_args__['polymorphic_identity']}")
logger.debug(f"Attempting to get regex for {submission_type}")
try:
@@ -707,7 +709,8 @@ class BasicSubmission(BaseClass):
# item.__mapper_args__['polymorphic_identity'] == polymorphic_identity][0]
model = cls.__mapper__.polymorphic_map[polymorphic_identity].class_
except Exception as e:
logger.error(f"Could not get polymorph {polymorphic_identity} of {cls} due to {e}, falling back to BasicSubmission")
logger.error(
f"Could not get polymorph {polymorphic_identity} of {cls} due to {e}, falling back to BasicSubmission")
case _:
pass
if attrs is None or len(attrs) == 0:
@@ -1199,6 +1202,8 @@ class BasicSubmission(BaseClass):
dlg = SubmissionComment(parent=obj, submission=self)
if dlg.exec():
comment = dlg.parse_form()
if comment in ["", None]:
return
self.set_attribute(key='comment', value=comment)
# logger.debug(self.comment)
self.save(original=False)
@@ -1368,49 +1373,6 @@ class BacterialCulture(BasicSubmission):
return input_dict
# class ViralCulture(BasicSubmission):
#
# id = Column(INTEGER, ForeignKey('_basicsubmission.id'), primary_key=True)
# __mapper_args__ = dict(polymorphic_identity="Viral Culture",
# polymorphic_load="inline",
# inherit_condition=(id == BasicSubmission.id))
#
# # @classmethod
# # def get_regex(cls) -> str:
# # """
# # Retrieves string for regex construction.
# #
# # Returns:
# # str: string for regex construction
# # """
# # return "(?P<Viral_Culture>RSL(?:-|_)?VE(?:-|_)?20\d{2}-?\d{2}-?\d{2}(?:(_|-)?\d?([^_0123456789\sA-QS-Z]|$)?R?\d?)?)"
#
# @classmethod
# def custom_sample_autofill_row(cls, sample, worksheet: Worksheet) -> int:
# """
# Extends parent
# """
# # logger.debug(f"Checking {sample.well}")
# # logger.debug(f"here's the worksheet: {worksheet}")
# row = super().custom_sample_autofill_row(sample, worksheet)
# df = pd.DataFrame(list(worksheet.values))
# # logger.debug(f"Here's the dataframe: {df}")
# idx = df[df[0] == sample.well]
# if idx.empty:
# new = f"{sample.well[0]}{sample.well[1:].zfill(2)}"
# # logger.debug(f"Checking: {new}")
# idx = df[df[0] == new]
# # logger.debug(f"Here is the row: {idx}")
# row = idx.index.to_list()[0]
# return row + 1
#
# @classmethod
# def custom_info_parser(cls, input_dict: dict, xl: Workbook | None = None, custom_fields: dict = {}) -> dict:
# input_dict = super().custom_info_parser(input_dict=input_dict, xl=xl, custom_fields=custom_fields)
# logger.debug(f"\n\nInfo dictionary:\n\n{pformat(input_dict)}\n\n")
# return input_dict
class Wastewater(BasicSubmission):
"""
derivative submission type from BasicSubmission
@@ -2239,6 +2201,9 @@ class BasicSample(BaseClass):
"""
gui friendly dictionary, extends parent method.
Args:
full_data (bool): Whether to use full object or truncated. Defaults to False
Returns:
dict: well location and name (sample id, organism) NOTE: keys must sync with WWSample to_sub_dict above
"""