analyze: misc fixes for simple_buffer

[spernsteiner]

This branch has fixes for various minor issues uncovered while testing on simple_buffer.c (#1103).

• Fix borrowck handling of casts to *mut my_struct when my_struct has lifetimes
• Fix various cases of compile errors introduced by DynOwned rewrites
• Fix precedence of method calls when generating rewritten source code
• Treat realloc(NULL, n) like malloc(n)
• Support nullable pointers in the ZeroizeTy initializers used for malloc/memset rewrites

[ahomescu]

What's the magic 3 value here?

[spernsteiner]

Operator precedence, roughly. Things that bind less tightly than . need to be parenthesized when used as a method receiver. E.g. (2 + 2).method()

[ahomescu]

Might be useful to add a comment for it, and/or change it to a const with a good name.

[spernsteiner]

Changing all the precedence values to named consts might be a good idea. I think that should be a separate PR, though

[ahomescu]

Could this just be src_ptr.unwrap_or_default()? Or use unwrap_or_else so you're not constructing a Box every time?

[spernsteiner]

Zero-sized boxes don't allocate - that includes boxed zero-length arrays, as in this case. As for unwrap_or_default, I generally prefer .unwrap_or_else(Vec::new) etc. to make it clear what the result type is intended to be.

[ahomescu]

Why is it Box<(i32)>? Is that inner tuple needed? (or is it something else)

[spernsteiner]

That's not a tuple (1-tuples require a trailing comma, like (i32,)). When reusing part of a type annotation as written in the original source code (in this case, the i32 from *mut i32), we wrap the reused type in parentheses so we don't have to worry about precedence or other interactions with the surrounding rewrite. In theory, these parentheses should be easy to detect and remove with an "unused parens" lint in the cases where they're not needed.