Make the .message field on exceptions non-empty
#2947
Conversation
This allows subclasses of SanicException to access their message via the `message` attribute. This makes it consistent with the `status_code`, `quiet`, `headers` attributes that were previously present on the class. Currently, the message attribute is present on the class but not on the instance, so accessing SanicException().message will return an empty string "" instead of the actual message "Internal Server Error", which you can get by calling the __str__() method or str(). This is a bit surprising, since the .message attribute shows up in autocomplete and type-checking. It also happens for the other exceptions, like str(BadRequest()) == "" as well. I set the message attribute on instances of SanicException and added tests for this new behavior.
ekzhang
force-pushed
the
ekzhang/sanic-message-attribute
branch
from
4fbd274
to
464fedb
Compare
sanic/exceptions.py
Outdated
| @@ -86,6 +87,7 @@ def __init__( | |||
| self.status_code = status_code or self.status_code | |||
| self.quiet = quiet | |||
| self.headers = headers | |||
| self.message = message # type: ignore | |||
There was a problem hiding this comment.
There is a type error here — it looks like message the argument of __init__ can take a bytes object, but the class attribute can only be a str.
There was a problem hiding this comment.
I suspect the option for it to be bytes might be because of STATUS_CODES messages being in bytes (assuming someone passed one of those to init), and this may be necessary to keep for compatibility.
Can you get away with typing without the ignore if you add some logic to decode the argument before storing to self.message, if bytes was received?
There was a problem hiding this comment.
Ah, that makes sense! I'll update my PR to do that :)
sanic/exceptions.py
Outdated
| @@ -86,6 +87,7 @@ def __init__( | |||
| self.status_code = status_code or self.status_code | |||
| self.quiet = quiet | |||
| self.headers = headers | |||
| self.message = message # type: ignore | |||
There was a problem hiding this comment.
I suspect the option for it to be bytes might be because of STATUS_CODES messages being in bytes (assuming someone passed one of those to init), and this may be necessary to keep for compatibility.
Can you get away with typing without the ignore if you add some logic to decode the argument before storing to self.message, if bytes was received?
|
Hi @Tronic, I updated the constructor to decode bytes arguments to str. It is a bit more verbose but works without disabling the type-checker on that line now. I also updated the test! The current CI failures appear to be from an unrelated error to this changeset. |
There was a problem hiding this comment.
LGTM, if not barring minor cleanup (e.g. removal of "utf8" argument on decode, perhaps some restructuring). I will assume @ahopkins to do another review before merging; from my point this is approved as is or with changes. Thanks for the contribution!
Thanks for reviewing so quickly! I followed the pattern of the existing code to use |
This allows subclasses of SanicException to access their message via the
messageattribute. This makes it consistent with thestatus_code,quiet,headersattributes that were previously present on the class.Currently, the message attribute is present on the class but not on the instance, so accessing SanicException("Failed!").message will return an empty string "" instead of the actual message "Failed!", which you can get by calling the
__str__()method orstr()function.This is a bit surprising, since the .message attribute shows up in autocomplete and type-checking. It also happens for the other exceptions, like BadRequest().message == "" as well.
I set the message attribute on instances of SanicException and added tests for this new behavior.
Doc reference
https://sanic.dev/en/guide/best-practices/exceptions.html#exception-properties