Add implied assertion in WheelInfo #145
Conversation
|
Thanks! |
|
Looks like this gets hit in many tests though. |
|
Hum, I'll dive into it more. |
Carreau
force-pushed
the
assrt-http
branch
2 times, most recently
from
fbb7c55
to
ab71634
Compare
|
Ok, some of the conftest.py where mocking the url without https, let's see. |
| @@ -257,7 +257,7 @@ def _mock_fetch_bytes(arg, *args, **kwargs): | |||
|
|
|||
| msg = "Access-Control-Allow-Origin" | |||
| with pytest.raises(ValueError, match=msg): | |||
| await micropip.install("htps://x.com/xxx-1.0.0-py3-none-any.whl") | |||
There was a problem hiding this comment.
It's not entirely clear if the typo here was on purpose.
There was a problem hiding this comment.
Looks pretty accidental to me. If it were on purpose you would expect a comment.
|
Worse case I'll push a better error in |
Some simple indexes (anaconda), only list the full-pathname of download URL (assuming same domain, which make sens). WheelInfo.download does not work on this case at it tries to make the requests on the current page domain. Thus we implicitely assume that URL start with http (well https:// would be better, but when you prototype local index could be http.) The error can be weird if we let it propagate (bad ziplife as you try to decode often a 404 html page as ZIP). This thus just add an assert at construction time to catch the error early and that we start with actual protocol. It should in the end be pushed earler in the code (likely at parsing time), where we are likely to know the index URL and be able to resolve URLs at that time. I think there was also a missing comma in test parametrisations, and I added a !r in a few places as I had some error with empty filenames, without quotes, so hard to see. The conftest was also update to explicitely pass a fake URL instead of just a filename. I'm not entirely sure we should allow constructing URLs with things that don't have a protocol though, as this can be confusing.
| @@ -7,17 +7,21 @@ | |||
| "path", | |||
| [ | |||
| SNOWBALL_WHEEL, | |||
| f"/{SNOWBALL_WHEEL}" f"a/{SNOWBALL_WHEEL}", | |||
There was a problem hiding this comment.
I think this is a mistake, and a , is needed.
| @pytest.mark.parametrize("protocol", ["https:", "file:", "emfs:", ""]) | ||
| @pytest.mark.parametrize( | ||
| "protocol", | ||
| ["http:", "https:", "file:", "emfs:", ""], |
There was a problem hiding this comment.
I have no clue what emfs is..
There was a problem hiding this comment.
It stands for "Emscripten file system". It's needed because file: points to files in the actual file system. Probably could use some comments and better documentation.
| def test_parse_wheel_url1(protocol, path): | ||
| pytest.importorskip("packaging") | ||
| from micropip.transaction import WheelInfo | ||
|
|
||
| url = protocol + path | ||
| url = protocol + path if protocol else "file:///" + path |
There was a problem hiding this comment.
Why not put file:/// in the parametrize decorator and do:
| url = protocol + path if protocol else "file:///" + path | |
| url = protocol + path |
There was a problem hiding this comment.
It's me I'm dumb I need to put that lower; I want to check that WheelInfo.from_url(url) parses things that have no protocol as local files.
Some simple indexes (anaconda), only list the full-pathname of download URL (assuming same domain, which make sens).
WheelInfo.download does not work on this case at it tries to make the requests on the current page domain.
Thus we implicitely assume that URL start with http (well https:// would be better, but when you prototype local index could be http.)
The error can be weird if we let it propagate (bad ziplife as you try to decode often a 404 html page as ZIP).
This thus just add an assert at construction time to catch the error early.
It should in the end be pushed earler in the code (likely at parsing time), where we are likely to know the index URL and be able to resolve URLs at that time.