It is not clear what Yassine’s bug is. Maybe it is about round-tripping from urlparse() → urlunparse(). If so, it could be solved by fixing either of the following two problems:

1. urlunparse() forgets the initial pair of slashes when netloc="". That might be addressed by Issue 22852, and documented as a limitation in the mean time.

2. urlunparse() accepts invalid components, such as netloc="", path="//evil.com", which transforms the path into a hostname. Yassine preferred to percent-encode the path and pass it through, though I think an exception would be more sensible. Or just documenting that there is little or no validation.

When considering the second problem of validation, you have to be aware that urlunparse() is documented to handle schemes like “mailto:” not listed in “uses_netloc”. According to RFC 6068, mailto://evil.com is valid syntax, and is decoded to netloc="", path="//evil.com". In this case, netloc="evil.com" would probably be invalid instead.