A problem with .NET that isn't talked about is the problems caused by using dynamic code generation. In a nut shell, dynamic code generation, which is used in XML Serialization, Regular Expressions, and XSLT transformations, can lead to memory leaks.
While the Common Language Runtime (CLR) can unload whole App Domains, it cannot unload individual assemblies. Code generation relies on creating temporary assemblies. These assemblies are often loaded into the primary app domain, meaning that cannot be unloaded until the application exits.
For libraries like XML Serialization, this isn't much of a problem. Usually the serialization code for a given type is cached, which limits the application to one temporary assembly per type. But there are XMLSerializer overloads that don't use caching. If a developer uses one of these without providing some sort of application-level cache, memory can be slowly leaked as new instances of essentially the same code is loaded into memory. For more information specifically on XML, see .NET Memory Leak: XmlSerializing your way to a Memory Leak.
The potential problems are far worse for code that habitually rewrites itself at runtime such as those written in LISP. For that sort of language, either the code has to be run entirely interpreted, that is never being compiled to IL code, or placed in separate AppDomains that can be purged from time to time. Either way, it poses a challenge for language developers targeting the Common Language Runtime.
While the Common Language Runtime (CLR) can unload whole App Domains, it cannot unload individual assemblies. Code generation relies on creating temporary assemblies. These assemblies are often loaded into the primary app domain, meaning that cannot be unloaded until the application exits.
For libraries like XML Serialization, this isn't much of a problem. Usually the serialization code for a given type is cached, which limits the application to one temporary assembly per type. But there are XMLSerializer overloads that don't use caching. If a developer uses one of these without providing some sort of application-level cache, memory can be slowly leaked as new instances of essentially the same code is loaded into memory. For more information specifically on XML, see .NET Memory Leak: XmlSerializing your way to a Memory Leak.
The potential problems are far worse for code that habitually rewrites itself at runtime such as those written in LISP. For that sort of language, either the code has to be run entirely interpreted, that is never being compiled to IL code, or placed in separate AppDomains that can be purged from time to time. Either way, it poses a challenge for language developers targeting the Common Language Runtime.
