A while ago I read about an idea to make it easier to avoid common programming mistakes in PHP regarding the handling of strings. There are dozens of attacks that one must pay attention to when using strings: you have to escape your string one way when you embed it in an SQL statement, escape it in a different way when outputting it as part of a web-page (XSL attacks), and escape it in a third way when you output it as part of a HTTP-header. It’s not surprising that eventually somewhere something will be not escaped in the right way.
Wells suggests a SafeString class to encapsulate all Strings in a class with different access methods that automatically escape your string the right way. So if you were to output the string back to the user, you’d call a toHTML() method that properly escapes any HTML-tags and special characters embedded in the string. A method to access the raw string would be called “UnsafeRawString” to remind the programmer that the string contains “tainted” user-input. While it is still possible to do something wrong, these parts stick out in the code (for example, one might use String->toHTML() when using it in an SQL statement – obviously wrong, but much easier to find). See “Making Wrong Code look Wrong” for the underlying philosophy.
I really like the idea, but I see a couple of practical problems with this idea:
- All strings, including Server variables and Super-Globals, should be automatically converted to the new String class. Otherwise the programmer has to constantly figure out if he/she is dealing with an encapsulated string or not.
- You’d need a database abstraction layer that will return these kind of strings as results of queries.
- All the existing PHP string operations (from strcmp to soundex) must be usable. This can be tricky, but interestingly PHP5 offers a way with __call to overload the object with arbitrarily named functions (see overload() function in PHP4). With some eval-magic this could be doable. Technically you wouldn’t want anybody to ever to work with the UnsafeRawString…