mirror of
https://github.com/ISEAL-Community/iseal-core.git
synced 2024-11-25 00:00:19 +01:00
util/generate-hugo-content.py: add support for fsc.csv
Currently the only difference is the name of the module column and that there are no mandatory fields so we need to be careful there.
This commit is contained in:
parent
f1807806d1
commit
d37bfc453f
@ -44,8 +44,19 @@ def parseSchema(schema_df):
|
||||
# For example Certifying Body, FSC audit, Certificate, etc
|
||||
cluster = row["idss element cluster"].capitalize()
|
||||
|
||||
# For example Assurance, Certification, Core, Impact, etc
|
||||
module = row["idss schema module"].capitalize()
|
||||
# Extract the module (whether from IDSS or extension), for example
|
||||
# Assurance, Certification, Core, Impact, etc
|
||||
if "idss schema module" in df.columns:
|
||||
module = row["idss schema module"]
|
||||
|
||||
# Use the default home layout
|
||||
layout = "home"
|
||||
elif "fsc extension module" in df.columns:
|
||||
module = row["fsc extension module"]
|
||||
|
||||
# Since we know this is the FSC schema we can set the layout to
|
||||
# use the custom fsc layout instead of the default home layout.
|
||||
layout = "fsc"
|
||||
|
||||
# Generate a "safe" version of the element name for use in URLs and
|
||||
# files by combining the cluster and the element name. This could
|
||||
@ -79,8 +90,12 @@ def parseSchema(schema_df):
|
||||
if row["element controlled values or terms"]:
|
||||
controlled_vocab = True
|
||||
|
||||
controlled_vocabulary_src=f"data/controlled-vocabularies/{element_name_safe}.txt"
|
||||
controlled_vocabulary_dst=f"site/content/terms/{element_name_safe}/vocabulary.txt"
|
||||
controlled_vocabulary_src = (
|
||||
f"data/controlled-vocabularies/{element_name_safe}.txt"
|
||||
)
|
||||
controlled_vocabulary_dst = (
|
||||
f"site/content/terms/{element_name_safe}/vocabulary.txt"
|
||||
)
|
||||
|
||||
copyfile(controlled_vocabulary_src, controlled_vocabulary_dst)
|
||||
|
||||
@ -89,7 +104,7 @@ def parseSchema(schema_df):
|
||||
else:
|
||||
controlled_vocab = False
|
||||
|
||||
if row["mandatory?"] == "MANDATORY":
|
||||
if "mandatory?" in df.columns and row["mandatory?"] == "MANDATORY":
|
||||
required = True
|
||||
else:
|
||||
required = False
|
||||
@ -129,6 +144,8 @@ def parseSchema(schema_df):
|
||||
if cluster:
|
||||
indexLines.append(f"cluster: '{cluster}'\n")
|
||||
indexLines.append(f"policy: '{policy}'\n")
|
||||
if layout:
|
||||
indexLines.append(f"layout: '{layout}'\n")
|
||||
## TODO: use some real date...?
|
||||
# indexLines.append(f"date: '2019-05-04T00:00:00+00:00'\n")
|
||||
indexLines.append("---")
|
||||
|
Loading…
Reference in New Issue
Block a user