Delete rules for vector<vector<T>> and for nested containers

[lucic71]

Closes #15

Translates vector<vector<T>> as Vec<Vec<T>> instead of Vec<Value<Vec<T>>>. Nested arrays also lose the extra Value. This allows deleting the rules for vector<vector<T>>.

This adds a new PtrKind:

pub trait SliceParent<T> {
    fn root(&self) -> Rc<dyn Any>;
    fn borrow_slice<'a>(&self, root: &'a dyn Any) -> Ref<'a, [T]>;
    fn borrow_slice_mut<'a>(&self, root: &'a dyn Any) -> RefMut<'a, [T]>;
    fn address(&self) -> usize;
}

PtrKind {
  Nested(Rc<dyn SliceParent<T>>)
  ...
}

root accesses the parent, borrow_slice/borrow_slice_mut are used for reading/writing into the nested row and address is sued for identity.

[nunoplopes]

this dyn doubles the size of PtrKind. Needs benchmarking.

[lucic71]

The size is the same. It's wrapped in an Rc, same as ReinterpretedView

[lucic71]

I will check if micro already contains benchmarks for arrays, otherwise I will add some

[nunoplopes]

Nop. size of Rc dyn is 2x size of Rc without dyn.

[lucic71]

Sorry, I meant that it does not increase the size of PtrKind because we already have a Weak<dyn> which is Field

[lucic71]

BTW I benchmarked brunsli and woff2 and I saw a slowdown between 1.2% and 1.4%. I will also benchmark micro

[nunoplopes]

Add a few tests that subtract pointers to ensure addresses are being computed correctly across multiple nested levels.
e.g.:

int v[3][4];
assert(&v[0][3] - &v[0][1] == 2);
assert(v[1] - (&v[0][0] + 3)  == 1);

[lucic71]

The second assert is UB. It's rejected by clang in constexpr mode: https://godbolt.org/z/T43K6M3v1. The pointers are not from the same array object

[nunoplopes]

Is this change expected?

--- a/woff2/out/refcount/src/bin/woff2_compress.rs
+++ b/woff2/out/refcount/src/bin/woff2_compress.rs
@@ -640,7 +640,7 @@ pub struct woff2_Font {
     pub num_tables: u16,
     #[offset(8)]
     #[byte_size(48)]
-    pub tables: BTreeMap<u32, Value<woff2_Font_Table>>,
+    pub tables: BTreeMap<u32, Value<(Value<u32>, Value<woff2_Font_Table>)>>,
 }
 #[derive(DeepClone, Record, ByteRepr, VaArg, FnPtrArg, Default)]
 #[byte_size(80)]
@@ -651,7 +651,7 @@ pub struct woff2_FontCollection {
     pub header_version: u32,
     #[offset(8)]
     #[byte_size(48)]
-    pub tables: BTreeMap<u32, Value<Ptr<woff2_Font_Table>>>,
+    pub tables: BTreeMap<u32, Value<(Value<u32>, Value<Ptr<woff2_Font_Table>>)>>,
     #[offset(56)]
     #[byte_size(24)]
     pub fonts: Value<Vec<woff2_Font>>,
@@ -704,7 +704,8 @@ pub fn ReadTrueTypeFont_33(
     {
         return false;
     }
-    let intervals: Value<BTreeMap<u32, Value<u32>>> = Rc::new(RefCell::new(BTreeMap::new()));
+    let intervals: Value<BTreeMap<u32, Value<(Value<u32>, Value<u32>)>>> =
+        Rc::new(RefCell::new(BTreeMap::new()));
     let mut i: u16 = 0_u16;
     'loop_: while ({ (i as i32) } < { (font.with(|__s| __s.num_tables) as i32) }) {
         let table: Value<woff2_Font_Table> = Rc::new(RefCell::new(<woff2_Font_Table>::default()));

[lucic71]

Is this change expected?

--- a/woff2/out/refcount/src/bin/woff2_compress.rs
+++ b/woff2/out/refcount/src/bin/woff2_compress.rs
@@ -640,7 +640,7 @@ pub struct woff2_Font {
     pub num_tables: u16,
     #[offset(8)]
     #[byte_size(48)]
-    pub tables: BTreeMap<u32, Value<woff2_Font_Table>>,
+    pub tables: BTreeMap<u32, Value<(Value<u32>, Value<woff2_Font_Table>)>>,
 }
 #[derive(DeepClone, Record, ByteRepr, VaArg, FnPtrArg, Default)]
 #[byte_size(80)]
@@ -651,7 +651,7 @@ pub struct woff2_FontCollection {
     pub header_version: u32,
     #[offset(8)]
     #[byte_size(48)]
-    pub tables: BTreeMap<u32, Value<Ptr<woff2_Font_Table>>>,
+    pub tables: BTreeMap<u32, Value<(Value<u32>, Value<Ptr<woff2_Font_Table>>)>>,
     #[offset(56)]
     #[byte_size(24)]
     pub fonts: Value<Vec<woff2_Font>>,
@@ -704,7 +704,8 @@ pub fn ReadTrueTypeFont_33(
     {
         return false;
     }
-    let intervals: Value<BTreeMap<u32, Value<u32>>> = Rc::new(RefCell::new(BTreeMap::new()));
+    let intervals: Value<BTreeMap<u32, Value<(Value<u32>, Value<u32>)>>> =
+        Rc::new(RefCell::new(BTreeMap::new()));
     let mut i: u16 = 0_u16;
     'loop_: while ({ (i as i32) } < { (font.with(|__s| __s.num_tables) as i32) }) {
         let table: Value<woff2_Font_Table> = Rc::new(RefCell::new(<woff2_Font_Table>::default()));

Yes, see #523

[lucic71]

I benchmarked the tests I added in Cpp2Rust/cpp2rust-testsuite#8, numbers only for refcount:

test	master (ms)	master vs cpp	nested-containers vs master
matmul_vector	139	164x	1.78x
matmul_array	104	59x	5.62x
row_ptr_vector	192	6x	2x
row_ptr_array	181	6.7x	1.87x