Page MenuHomeSoftware Heritage
Differential D215 Diff 719 swh/indexer/metadata_dictionary.py

Changeset View

Standalone View

View Options

swh/indexer/metadata_dictionary.py

  • This file was added.
# Copyright (C) 2017 The Software Heritage developers
ardumontUnsubmitted
Done
Inline Actions

2017

ardumont: 2017
moraneggAuthorUnsubmitted
Not Done
Inline Actions

ack

moranegg: ack
# See the AUTHORS file at the top-level directory of this distribution
# License: GNU General Public License version 3, or any later version
# See top-level LICENSE file for more information
import json
def convert(raw_content):
"""
convert raw_content recursively:
- from bytes to string
- from string to dict
Args:
- raw_content (bytes / string / dict)
Returns:
- Dict of content (if string was json, otherwise returns string)
"""
if isinstance(raw_content, bytes):
return convert(raw_content.decode())
if isinstance(raw_content, str):
try:
content = json.loads(raw_content)
if content:
return content
else:
return raw_content
except json.decoder.JSONDecodeError:
return raw_content
if isinstance(raw_content, dict):
return raw_content
class BaseMapping():
"""Base class for mappings to inherit from
To implement a new mapping:
- inherit this class
- add a local property self.mapping
- override translate function
"""
def translate(self, content_dict):
"""
Tranlsates content by parsing content to a json object
and translating with the npm mapping (for now hard_coded mapping)
Args:
- context_text (text) : should be json
Returns:
- translated_metadata (dict): jsonb form needed for the indexer
"""
translated_metadata = {}
default = 'other'
translated_metadata['other'] = {}
try:
for k, v in content_dict.items():
try:
ardumontUnsubmitted
Done
Inline Actions

You could use a dict of key: {context, translation function}.

Something like:

mapping_fn = {
    "hard_mapping_fn": translate_npm,
    "pom.xml": lambda content: translate_pom(parse_xml(content)),
  ...
}

# then parse method is simplified
class MetadataDict():
	​
	​    def __init__(self):
	​        pass
	​
	​    def parse(self, context, content):
	​        return mapping_fn(context)(content)
...

Since i don't see any object state used in the Metadata dict functions (translate_npm, etc...), those could be simple functions (and not object methods).

Hey, the parse method itself could be a function here :)

Personally, i tend to use class when there needs to be shared state (accessible through the self. instance variable).
Otherwise, i define functions.

ardumont: You could use a dict of key: {context, translation function}. Something like: ``` mapping_fn…
moraneggAuthorUnsubmitted
Not Done
Inline Actions

very useful. thanks for this, I will rework the design of the MetaDict to decide is the class is needed.

I'm trying to integrate the mapping_fn but I have some issues with calling it..

I will continue this tomorrow..

moranegg: very useful. thanks for this, I will rework the design of the MetaDict to decide is the class…
ardumontUnsubmitted
Done
Inline Actions

I will continue this tomorrow..

Sure.

I'm trying to integrate the mapping_fn but I have some issues with calling it..

Calling it should be as simple as:

mapping_fn['hard_mapping_fn'](content)
  • mapping_fn['hard_mapping_fn'] being resolved to the 'translate_npm' function.
  • 'content' being defined as a something that makes sense for the translate_npm function.

oh...

I see there is a typo in the gist i mentioned to you. Try using bracket instead of parenthesis.

mapping_fn[context](content)
ardumont: > I will continue this tomorrow.. Sure. > I'm trying to integrate the mapping_fn but I have…
moraneggAuthorUnsubmitted
Not Done
Inline Actions

not using a class and moved the function compute_metadata into the metadata_dictionary to keep the logic:

  • indexer to retrieve data from storage and store result
  • compute_xxxxx() uses tool and return the result from the tool
moranegg: not using a class and moved the function compute_metadata into the metadata_dictionary to keep…
term = self.mapping.get(k, default)
if term not in translated_metadata:
translated_metadata[term] = v
continue
if isinstance(translated_metadata[term], str):
ardumontUnsubmitted
Done
Inline Actions

parse_xml?

ardumont: `parse_xml`?
in_value = translated_metadata[term]
translated_metadata[term] = [in_value, v]
continue
if isinstance(translated_metadata[term], list):
translated_metadata[term].append(v)
continue
if isinstance(translated_metadata[term], dict):
translated_metadata[term][k] = v
continue
except KeyError:
self.log.exception(
"Problem during item mapping")
continue
except:
return None
return translated_metadata
class NpmMapping(BaseMapping):
"""
dedicated class for NPM (package.json) mapping and translation
"""
mapping = {
'repository': 'codeRepository',
'os': 'operatingSystem',
'cpu': 'processorRequirements',
'engines': 'processorRequirements',
'dependencies': 'softwareRequirements',
'bundleDependencies': 'softwareRequirements',
'peerDependencies': 'softwareRequirements',
'author': 'author',
'contributor': 'contributor',
'keywords': 'keywords',
'license': 'license',
'version': 'version',
'description': 'description',
'name': 'name',
'devDependencies': 'softwareSuggestions',
'optionalDependencies': 'softwareSuggestions',
'bugs': 'issueTracker',
'homepage': 'url'
}
def translate(self, raw_content):
content_dict = convert(raw_content)
return super().translate(content_dict)
ardumontUnsubmitted
Not Done
Inline Actions

If there is nothing inside, you don't need to define it :)

ardumont: If there is nothing inside, you don't need to define it :)
ardumontUnsubmitted
Done
Inline Actions

why do you want to name it differently, the class already holds the npm inside.
I found the plain 'mapping' name sufficient.

ardumont: why do you want to name it differently, the class already holds the npm inside. I found the…
class MavenMapping(BaseMapping):
"""
dedicated class for Maven (pom.xml) mapping and translation
"""
ardumontUnsubmitted
Not Done
Inline Actions

If that code is the same on all derived classes, this could prove it belongs to the BaseMapping's translate definition.
The only override then being the mapping variable.

ardumont: If that code is the same on all derived classes, this could prove it belongs to the…
moraneggAuthorUnsubmitted
Not Done
Inline Actions

now it's the same because I'm lazy and I haven't started working on other files, but in cases where the file doesn't contain json it should be decoded and/or parsed differently so this is why i didn't keep it in the translate method.

Also I wanted to keep only one task in the translate method which the semantical translation from a dict with original terms to a dict with CodeMeta terms.

moranegg: now it's the same because I'm lazy and I haven't started working on other files, but in cases…
ardumontUnsubmitted
Not Done
Inline Actions

now it's the same because I'm lazy...

sure, aren't we all in some form or another? :)

ardumont: > now it's the same because I'm lazy... sure, aren't we all in some form or another? :)
mapping = {
'license': 'license',
'version': 'version',
'description': 'description',
'name': 'name',
'prerequisites': 'softwareRequirements',
'repositories': 'codeRepository',
'groupId': 'identifier',
'ciManagement': 'contIntegration',
'issuesManagement': 'issueTracker',
}
def translate(self, raw_content):
content = convert(raw_content)
# parse content from xml to dict
return super().translate(content)
class DoapMapping(BaseMapping):
mapping = {
ardumontUnsubmitted
Done
Inline Actions

Since you do need a distinct mapping dict for all your translation, you could use classes...
(Sorry for proposing yet again another implementation but there are more in your code now :)

To avoid having out of context dictionaries defined all over the place (npm_mapping, doap_mapping, etc...).
And since those are only used in a specific context (the <techno> mapping), we could define a class per mapping which enclose said mapping, something like:

class NpmMapping:
    mapping = {<npm-mapping-dict-here>}

    def translate(self, content):
        return translate(content, self.mapping)

class PomMapping:
    mapping = {<pom-mapping-here>}

    def translate(self, content):
        return translate(parse_xml(content), self.mapping)

class DoapMapping:
    mapping = {<doap-mapping-dict-here>} 

    def translate(self, content):
        return translate(parse_xml(content), self.mapping)

### Note that we could factor again code here :) ###

# mapping_tool_fn becomes:
mapping_tool_fn = {
    "hard_mapping_npm": NpmMapping(),
    "pom_xml": PomMapping(),
    "doap_xml": DoapMapping(),
}

# finally compute-metadata:
def compute_metadata(context, raw_content):
    content = convert(raw_content)
    if content is None:
        return None
    translated_metadata = mapping_tool_fn[context].translate(content)
    return translated_metadata


etc...

But feel free to continue on your initial way for now, it's just suggestion.

ardumont: Since you do need a distinct mapping dict for all your translation, you could use classes...
moraneggAuthorUnsubmitted
Not Done
Inline Actions

I really like this approach and I could overload the "translate()" function if needed with a parent class metadataMapping()
and the child classes could have your design for now with the hard coded mapping.

Eventually I will use the CodeMeta crosswalk table to generate the mappings so this might change a bit.

Thanks for the help with the design of the component !

moranegg: I really like this approach and I could overload the "translate()" function if needed with a…
ardumontUnsubmitted
Not Done
Inline Actions

I really like this approach

cool

and I could overload the "translate()" ...

indeed :)

ardumont: > I really like this approach cool > and I could overload the "translate()" ... indeed :)
}
def translate(self, raw_content):
content = convert(raw_content)
# parse content from xml to dict
return super().translate(content)
def parse_xml(content):
"""
Parses content from xml to a python dict
Args:
- content (text): the string form of the raw_content ( in xml)
Returns:
- parsed_xml (dict): a python dict of the content after parsing
"""
# check if xml
# use xml parser to dict
return content
mapping_tool_fn = {
"hard_mapping_npm": NpmMapping(),
"pom_xml": MavenMapping(),
"doap_xml": DoapMapping()
}
def compute_metadata(context, raw_content):
"""
first landing method: a dispatcher that sends content
to the right function to carry out the real parsing of syntax
and translation of terms
Args:
- context (text) : defines to which function/tool
the content is sent
- content (text): the string form of the raw_content
Returns:
- translated_metadata (dict): jsonb form needed for the indexer
to store in storage
"""
if raw_content is None or raw_content is b"":
return None
# TODO: keep mapping not in code (maybe fetch crosswalk from storage?)
# if fetched from storage should be done once for batch of sha1s
dictionary = mapping_tool_fn[context]
translated_metadata = dictionary.translate(raw_content)
# print(translated_metadata)
return translated_metadata
def main():
raw_content = """{"name": "test_name", "unknown_term": "ut"}"""
raw_content1 = b"""{"name": "test_name",
"unknown_term": "ut",
"prerequisites" :"packageXYZ"}"""
result = compute_metadata("hard_mapping_npm", raw_content)
result1 = compute_metadata("pom_xml", raw_content1)
print(result)
print(result1)
if __name__ == "__main__":
main()
About · Developer information · Legal