We've been hit by a bug in a real world application after upgrading 1.4.1 to 1.6, where previously valid URLs are no longer valid, which looks to be due to using java.net.URI for validating the path of a URL.
Our application went to validate URLs similar to the following:
This is no longer valid in 1.6.1, but the following cases are:
It seems paths in UrlValidator are being parsed/validated as host-names, per java.net.URI's validation.
It looks like this may have been introduced by the following change:
Specifically due to now using java.net.URI to validate a path. The usage is as follows in org.apache.commons.validator.routines.UrlValidator:
It looks like URI is trying to parse the path as a hostname when the schema and hostname are not specified.
Example to reproduce:
Same example with other parts, no longer throwing exception:
Even though java.net.URI states string components can be null, it seems the URL built internally, which is validated, is slightly different. So when specifying a hostname with URI, internally it constructs:
Using no hostname, in the same way as UrlValidator, the following is constructed and validated internally:
Therefore it looks like there's either a bug in java.net.URI, or its usage is not correctly documented.
A potential fix is to change org.apache.commons.validator.routines.UrlValidator to pass an empty string in the hostname. Internally, in java.net.URI, this produces:
Thus the hostname is empty, which is considered empty, and the correct path is validated.
Would this fix be appropriate, or considered too fragile?
Alternatively the fix could be to extract similar logic to java.net.URI, to validate the path, which appears to be just checking the characters are valid and between a certain range. This logic can be seen in java.net.URI.parseHierarchical, which calls upon checkChars.