On 28/12/12 08:42, Louis-Dominique Dubeau wrote:

> A dependency analyzer would also have to check for calls to document()
> and unparsed-text() (and maybe other calls I'm forgetting right now),
> because the result of the transformation can be dependent on input from
> these functions.

As I understand it, that sounds like a variation on the Halting problem 
(non-Comp Sci people see ). 
Once you start computing file paths in the code being studied, 
dependency analysis becomes very, very, tricky.

However, there are some rules of thumb which mean that we can take 
shortcuts. Things like "if file X.a depends on file Y.b, then a file Z.b 
in the same directory as Y.b is also a dependency if X != Y"