The following bugs have been fixed in AMC v3.6.4:

• If the getcwd system call failed AMC would abort because a path name could not be normalized. In some cases (NFS or strict permissions on parent directories), the getcwd function can fail. Because this is non-fatal, the un-normalized name is used instead.
• A portability problem with struct utsname existed where the code assumed that there was space in the structure that could be written to. This assumption is not guaranteed by POSIX and may break on some architectures that just have pointers in the structure to constant areas.
• If there was an error processing an include file the depth counter was not properly re-balanced. This could cause problems later on if project files were ever parsed on the fly.
• Duplicate CGL procedures were not flagged as errors. Rather, the newer definition overshadowed the old one. This was intentional to allow circular recursion before the dynamic_call built-in was added. Now that it is there, there is no reason to be silent about this kind of error.
• An off-by-one index into the internationalization table that caused some messages to be generated incorrectly.
• During the processing of a module, the header keyword list was active, thus modules could not consist of syntax that had keywords the same as the module header. The keyword list is now re-set.
• There was a bug in the substr built-in that would have allowed negative numbers as indicies to cause problems.

• A new variable is defined during the processing of a project file that allows projects written using AMC to be "directory independant" - their location in the file system can be automatically determined.
• Added a new CGL built-in, mkdir to ensure that a directory exists.
• A new directive allows idempotent inclusion (only include a file once) in the project files.
• Added a new CGL built-in, fn_normalize to compute an absolute file name.
• CGL code can now insert and process custom records in the interface files. It used to be that only chunks of output text were allowed, now user-defined record type are allowed.
• CGL maps have been improved (with a small syntax change). It is now possible for more than one value to share the same piece of code.
• Added a generic, efficient sorting mechanism to CGL.
• A new CGL built-in called strcmp provides a way to decide if a string is less than, greater than, or equal to another string.