fix validate credentials

This commit is contained in:
Cayo Puigdefabregas 2024-02-15 18:59:57 +01:00
parent c218245707
commit 5aa22419ac
3 changed files with 96 additions and 25 deletions

View File

@ -1,7 +1,8 @@
import csv import csv
import json import json
import base64
import copy import copy
import base64
import jsonschema
import pandas as pd import pandas as pd
from pyhanko.sign import signers from pyhanko.sign import signers
@ -176,20 +177,10 @@ class ImportForm(forms.Form):
self._schema = schema.first() self._schema = schema.first()
try: try:
self.json_schema = json.loads(self._schema.data) self.json_schema = self._schema.get_credential_subject_schema()
props = [x for x in self.json_schema["allOf"] if 'properties' in x.keys()]
prop = props[0]['properties']
self.properties = prop['credentialSubject']['properties']
except Exception: except Exception:
raise ValidationError("Schema is not valid!") raise ValidationError("Schema is not valid!")
if not self.properties:
raise ValidationError("Schema is not valid!")
# TODO we need filter "$ref" of schema for can validate a csv
self.json_schema_filtered = copy.copy(self.json_schema)
allOf = [x for x in self.json_schema["allOf"] if '$ref' not in x.keys()]
self.json_schema_filtered["allOf"] = allOf
return data return data
def clean_file_import(self): def clean_file_import(self):
@ -197,17 +188,19 @@ class ImportForm(forms.Form):
self.file_name = data.name self.file_name = data.name
df = pd.read_excel(data) df = pd.read_excel(data)
# convert dates to iso 8601
for col in df.select_dtypes(include='datetime').columns:
df[col] = df[col].dt.strftime("%Y-%m-%d")
data_pd = df.fillna('').to_dict() data_pd = df.fillna('').to_dict()
if not data_pd: if not data_pd:
self.exception("This file is empty!") self.exception("This file is empty!")
head_row = {x: '' for x in self.properties.keys()}
for n in range(df.last_valid_index()+1): for n in range(df.last_valid_index()+1):
row = {} row = {}
for k in data_pd.keys(): for k in data_pd.keys():
if k in self.properties.keys(): row[k] = data_pd[k][n] or ''
row[k] = data_pd[k][n] or ''
user = self.validate_jsonld(n, row) user = self.validate_jsonld(n, row)
self.rows[user] = row self.rows[user] = row
@ -229,13 +222,15 @@ class ImportForm(forms.Form):
def validate_jsonld(self, line, row): def validate_jsonld(self, line, row):
try: try:
# check = credtools.validate_json(row, self.json_schema_filtered) jsonschema.validate(instance=row, schema=self.json_schema)
check = credtools.validate_json(row, self.json_schema) except jsonschema.exceptions.ValidationError as err:
if check is not True: msg = "line {}: {}".format(line+1, err)
raise ValidationError("Not valid row")
except Exception as e:
msg = "line {}: {}".format(line+1, e)
return self.exception(msg) return self.exception(msg)
# try:
# check = credtools.validate_json(row, self.json_schema)
# if check is not True:
# raise ValidationError("Not valid row")
# except Exception as e:
user, new = User.objects.get_or_create(email=row.get('email')) user, new = User.objects.get_or_create(email=row.get('email'))
if new: if new:

View File

@ -541,6 +541,28 @@ class Schemas(models.Model):
def description(self, value): def description(self, value):
self._description = value self._description = value
def get_credential_subject_schema(self):
sc = self.get_data()
properties = sc["allOf"][1]["properties"]["credentialSubject"]["properties"]
required = sc["allOf"][1]["properties"]["credentialSubject"]["required"]
if "id" in required:
required.remove("id")
schema = {
"$schema": "https://json-schema.org/draft/2020-12/schema",
"type": "object",
"properties": properties,
"required": required,
"additionalProperties": False
}
return schema
def get_data(self):
return json.loads(self.data)
class VerificableCredential(models.Model): class VerificableCredential(models.Model):
""" """
Definition of Verificable Credentials Definition of Verificable Credentials

View File

@ -9,6 +9,7 @@ import requests
from pyld import jsonld from pyld import jsonld
import jsonref import jsonref
from jsonpath_ng import jsonpath, parse from jsonpath_ng import jsonpath, parse
from datetime import datetime
# def remove_null_values(dictionary): # def remove_null_values(dictionary):
@ -205,16 +206,53 @@ def schema_to_xls_comment(schema, xls_file_path):
# Get the xlsxwriter workbook and worksheet objects # Get the xlsxwriter workbook and worksheet objects
workbook = writer.book workbook = writer.book
matches_title = parse('$.title').find(schema)
title = matches_title[0].value if matches_title else 'no title'
matches_desc = parse('$.description').find(schema)
desc = matches_desc[0].value if matches_desc else 'no description'
matches_id = parse("$['$id']").find(schema)
idschema = matches_id[0].value if matches_id else 'no schema'
matches_subject_desc = parse('$..credentialSubject.description').find(schema)
subject_desc = matches_subject_desc[0].value if matches_subject_desc else 'no subject description'
workbook.set_properties({
'title': title,
'subject': desc,
'author': 'IdHub Orchestral',
'category': subject_desc,
'keywords': 'schema, template, plantilla',
'created': datetime.now().date(), #datetime.date(2018, 1, 1),
'comments': 'Created with Python for IdHub'})
workbook.set_custom_property('Schema', idschema)
worksheet = writer.sheets['Full1'] worksheet = writer.sheets['Full1']
# Define a format for the required header cells # Define a format for the required header cells
req_format = workbook.add_format({'border': 1}) req_f = workbook.add_format({'border': 1})
# cell_format = workbook.add_format({'bold': True, 'font_color': 'red'}) req_da = workbook.add_format({'border': 1, 'num_format': 'yyyy-mm-dd'})
req_in = workbook.add_format({'border': 1, 'num_format': '0'})
req_st = workbook.add_format({'border': 1, 'num_format': '@'})
opt_da = workbook.add_format({'num_format': 'yyyy-mm-dd'})
opt_in = workbook.add_format({'num_format': '0'})
opt_st = workbook.add_format({'num_format': '@'})
fmts = {
'string' : {True: req_st, False: opt_st},
'date' : {True: req_da, False: opt_da},
'integer' : {True: req_in, False: opt_in}
}
# Write comments to the cells # Write comments to the cells
for i, header in enumerate(headers): for i, header in enumerate(headers):
if header in req: fmt = {}
worksheet.set_column(i,i, None, req_format) #if header in req:
# fmt = req_format
# worksheet.set_column(i,i, None, req_format)
# Get the description for the current field # Get the description for the current field
if 'description' in matches[0][header]: if 'description' in matches[0][header]:
description = matches[0][header]['description'] description = matches[0][header]['description']
@ -222,6 +260,22 @@ def schema_to_xls_comment(schema, xls_file_path):
# Write the description as a comment to the corresponding cell # Write the description as a comment to the corresponding cell
worksheet.write_comment(0, i, description) worksheet.write_comment(0, i, description)
# Get the type for the current field
if 'type' in matches[0][header]:
type_field = matches[0][header]['type']
format_field = None
if 'format' in matches[0][header]:
format_field = matches[0][header]['format']
if type_field is not None:
if format_field is not None and format_field == 'date':
type_field = 'date'
fmt = fmts[type_field][header in req] # Add type format
print(f'header {header} with fmt {fmt}\n')
worksheet.set_column(i,i, None, fmt)
# Close the Pandas Excel writer and output the Excel file # Close the Pandas Excel writer and output the Excel file
worksheet.autofit() worksheet.autofit()
writer.close() writer.close()