Support asdf:// URI scheme #854
Conversation
| """ | ||
| assert generic_io.resolve_uri( | ||
| '{}://somewhere.org/some-schema'.format(protocol), '#/definitions/foo' | ||
| ) == '{}://somewhere.org/some-schema#/definitions/foo'.format(protocol) |
There was a problem hiding this comment.
Can we use f-strings here and bump the minimum version of python to 3.6 for our 2.8 release?
I don't see a good reason to keep supporting python 3.5 when it is no longer supported by astropy and numpy.
There was a problem hiding this comment.
You're asking the good questions. I'll start an email discussion.
| # We're importing our own copy of urllib.parse because | ||
| # we need to patch it to support asdf:// URIs, but it'd | ||
| # be irresponsible to do this for all users of a | ||
| # standard library. | ||
| urllib_parse_spec = importlib.util.find_spec('urllib.parse') | ||
| patched_urllib_parse = importlib.util.module_from_spec(urllib_parse_spec) | ||
| urllib_parse_spec.loader.exec_module(patched_urllib_parse) | ||
| del urllib_parse_spec |
There was a problem hiding this comment.
What the advantage of doing it this way rather than
import urllib.parse as patched_urllib_parse
It's still local to this module's namespace, right?
There was a problem hiding this comment.
Unfortunately while that aliases the module reference to a new name, it still points to the same module object. Import caches the module after loading it the first time so that other imports don't have to read and parse the module's code again.
# module1.py
import urllib.parse as module1_alias
# module2.py
import urllib.parse as module2_alias
# Fire up a console:
In [1]: import module1
In [2]: import module2
In [3]: module1.module1_alias is module2.module2_alias
Out[3]: TrueThere was a problem hiding this comment.
Ah, I see. I didn't realize there was caching going on. This bypasses that. #themoreyouknow
| # urllib.parse needs to know that it should treat asdf:// | ||
| # URIs like http:// URIs for the purposes of joining | ||
| # a relative path to a base URI. | ||
| patched_urllib_parse.uses_relative.append('asdf') |
There was a problem hiding this comment.
Digging deep into the standard library. What a naughty boy. ;-) I had to go look at the library source code to see this explained.
This PR allow asdf to properly resolve relative references against
asdf://URIs. Theurllib.parsemodule maintains a list of URI schemes that should behave likehttpwith regard to resolving relative paths, and neededasdfadded to its list.There's a second change in here that fixes a long-standing issue with resolving references in metaschemas. It's common for such schemas to define an
idproperty, but the asdf library didn't know not to treat that property object as a URI. The fix is to check thatidis a string before assuming it's a URI.