Skip to content

debuginfo: slices are DW_TAG_array_type's #153238

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
shua wants to merge 2 commits into
base: main
Choose a base branch
from
Open

debuginfo: slices are DW_TAG_array_type's #153238

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
31ce9e7
debuginfo: slices are DW_TAG_array_type's
shua Feb 26, 2026
18b4e24
adding debuginfo tests for str repr llvm di
shua May 5, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
49 changes: 30 additions & 19 deletions compiler/rustc_codegen_llvm/src/debuginfo/metadata.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -152,7 +152,20 @@ fn build_pointer_or_reference_di_node<'ll, 'tcx>(
cx.size_and_align_of(Ty::new_mut_ptr(cx.tcx, pointee_type))
);

let pointee_type_di_node = type_di_node(cx, pointee_type);
let pointee_type_di_node = match pointee_type.kind() {
// `&[T]` will look like `{ data_ptr: *const T, length: usize }`
ty::Slice(element_type) => type_di_node(cx, *element_type),
// `&str` will look like `{ data_ptr: *const u8, length: usize }`
ty::Str => type_di_node(cx, cx.tcx.types.u8),

// `&dyn K` will look like `{ pointer: _, vtable: _}`
// any Adt `Foo` containing an unsized type (eg `&[_]` or `&dyn _`)
// will look like `{ data_ptr: *const Foo, length: usize }`
// and thin pointers `&Foo` will just look like `*const Foo`.
//
// in all those cases, we just use the pointee_type
_ => type_di_node(cx, pointee_type),
};
Comment on lines +155 to +168
Copy link
Copy Markdown
Contributor

@Walnut356 Walnut356 May 17, 2026 •
edited by rustbot
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Oh, I think I get it. Basically, if you have a struct:

struct T {
	a: S,
	b: [S]
}

A pointer to that struct has the layout:

{
	data_ptr: *T {
		a: S,
		b: [S]
	},
	length: usize,
}

The length applies to the b field, not the whole struct. That applies recursively, so if you have some wrappers:

struct X(T);
struct W(X);

...

let ref_w = &W(X(T{...}));

the data layout would be

{
	data_ptr: *W {
		__0: X {
			__0 T {
				a: S,
				b: [S]
			}
		}
	},
	length: usize
}

All that said, I'd still need to test this a bit to see if it's the case, but I think this match is making too many assumptions and isn't accurately representing Rust's !Sized type layout, because { data_ptr: *const T, length: usize } can mean T itself is the unsized part (i.e. [T]) or that a field of T is unsized.

It might be the case that the visualization won't work out of the box, but so long as a struct's !Sized field can be differentiated from an "identically sized" Sized field, the visualizers can fix it before it reaches the user.

i.e. if we can tell the difference between &X(T) and

struct T2 {
	a: S,
	b: S,
}

struct Y([T2]);

from the field layout information alone, we're golden.

I have a full adversarial example here, where the align and padding mean there's no other way to tell them apart (with output from GDB failing to differentiate them for reference).

View changes since the review

Copy link
Copy Markdown
Contributor

@Walnut356 Walnut356 May 18, 2026 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Looked into this a bit more, I think this is due to GDB's printing behavior. The types are being output accurately:

(gdb) ptype a
type = struct *const arbitrary::A {
  data_ptr: *mut arbitrary::A,
  length: usize,
}
(gdb) ptype arbitrary::A
type = struct arbitrary::A {
  f1: u64,
  f2: u16,
  f3: [arbitrary::S],
}

I think the problem is GDB is hard-coded to treat structs with the shape of {data_ptr, length} as slices and prints them "shallowly" as slices (see: val_print_slice and value_print_inner).

I'm not really familiar with GDB's value or type representations though, so I could be wrong. @tromey does the above sound about right?

I don't think we'll get a disable-gdb-pretty-printers test for this to pass until GDB gets updated. On the other hand, visualizers should be able to override this behavior and print properly. It'd need a check for the {data_ptr, length} shape in rust_types.classify_rust_type, and the provider would need to search the pointee's last field recursively to find the one with a [T]/str type.

Copy link
Copy Markdown
Author

@shua shua May 18, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'm not sure I understand your concerns in the first comment, but I think we have the same understanding from your second. Specifically, if gdb (or lldb, etc) sees a "fat pointer" type of {data_ptr, length} it should recursively walk through the debuginfo definitions of data_ptr's type:

  1. if it is a DW_TAG_array_type with a member field DW_AT_count of -1, then it is a dynamic array, and the fat pointer's length is for this value,
  2. if it is a DW_TAG_structure_type, check the type of the last field in this struct, starting from step 1. above
  3. if it is neither of those two cases, this is some error in debuginfo output; do whatever you think is best.

I found similar confusion around how to handle this in https://sourceware.org/bugzilla/show_bug.cgi?id=30330

Copy link
Copy Markdown
Author

@shua shua May 20, 2026 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

ah, I think I understand your comments better. I thought they were pointing out something wrong in my change, but it seems that the mingw version of gdb that we test with has some incorrect handling of these fat pointers which are now showing up in the tests.

Specifically, the test creates an Rc<str> and prints it with gdb, we expect something like

Rc<str, _> {
    ptr: NonNull<RcInner<str>> {
        pointer: RcInner<str> {
            strong: Cell<usize> {value: UnsafeCell<usize> {value: 1}}, 
            weak: Cell<usize> {value: UnsafeCell<usize> {value: 1}}, 
            value: 0x6a48f0
        }
    }, 
    phantom: PhantomData<RcInner<str>>, 
    alloc: _
}

but currently mingw is printing

Rc<str, _> {
    ptr: NonNull<RcInner<str>> {
        pointer: *const RcInner<str> [
            RcInner<str> {
                strong: Cell<usize> {value: UnsafeCell<usize> {value: 1}}, 
                weak: Cell<usize> {value: UnsafeCell<usize> {value: 1}}, 
                value: 0x6a48f0}, 
            RcInner<str> {  // garbage
                strong: Cell<usize> {value: UnsafeCell<usize> {value: 13451671976132374359}}, 
                weak: Cell<usize> {value: UnsafeCell<usize> {value: 12370169555311111083}}, 
                value: 0x6a4900}
            RcInner<str> {...}, // more garbage...
            RcInner<str> {...},
            RcInner<str> {...}
        ]
    },
    phantom: PhantomData<RcInner<str>>, 
    alloc: _
}

This indicates that the debugger is incorrectly treating the NonNull field of *const RcInner<str> as if it is a pointer to some array of RcInner<str>, when it should be treating it as simply a pointer to a single RcInner<str>.


return_if_di_node_created_in_meantime!(cx, unique_type_id);

Expand Down Expand Up @@ -389,26 +402,11 @@ fn build_dyn_type_di_node<'ll, 'tcx>(
}

/// Create debuginfo for `[T]` and `str`. These are unsized.
///
/// NOTE: We currently emit just emit the debuginfo for the element type here
/// (i.e. `T` for slices and `u8` for `str`), so that we end up with
/// `*const T` for the `data_ptr` field of the corresponding wide-pointer
/// debuginfo of `&[T]`.
///
/// It would be preferable and more accurate if we emitted a DIArray of T
/// without an upper bound instead. That is, LLVM already supports emitting
/// debuginfo of arrays of unknown size. But GDB currently seems to end up
/// in an infinite loop when confronted with such a type.
///
/// As a side effect of the current encoding every instance of a type like
/// `struct Foo { unsized_field: [u8] }` will look like
/// `struct Foo { unsized_field: u8 }` in debuginfo. If the length of the
/// slice is zero, then accessing `unsized_field` in the debugger would
/// result in an out-of-bounds access.
fn build_slice_type_di_node<'ll, 'tcx>(
cx: &CodegenCx<'ll, 'tcx>,
slice_type: Ty<'tcx>,
unique_type_id: UniqueTypeId<'tcx>,
span: Span,
) -> DINodeCreationResult<'ll> {
let element_type = match slice_type.kind() {
ty::Slice(element_type) => *element_type,
Expand All @@ -423,7 +421,20 @@ fn build_slice_type_di_node<'ll, 'tcx>(

let element_type_di_node = type_di_node(cx, element_type);
return_if_di_node_created_in_meantime!(cx, unique_type_id);
DINodeCreationResult { di_node: element_type_di_node, already_stored_in_typemap: false }
let (size, align) = cx.spanned_size_and_align_of(slice_type, span);
let subrange = unsafe { llvm::LLVMDIBuilderGetOrCreateSubrange(DIB(cx), 0, -1) };
let subscripts = &[subrange];
let di_node = unsafe {
llvm::LLVMDIBuilderCreateArrayType(
DIB(cx),
size.bits(),
align.bits() as u32,
element_type_di_node,
subscripts.as_ptr(),
subscripts.len() as c_uint,
)
};
DINodeCreationResult { di_node, already_stored_in_typemap: false }
}

/// Get the debuginfo node for the given type.
Expand Down Expand Up @@ -454,7 +465,7 @@ pub(crate) fn spanned_type_di_node<'ll, 'tcx>(
}
ty::Tuple(elements) if elements.is_empty() => build_basic_type_di_node(cx, t),
ty::Array(..) => build_fixed_size_array_di_node(cx, unique_type_id, t, span),
ty::Slice(_) | ty::Str => build_slice_type_di_node(cx, t, unique_type_id),
ty::Slice(_) | ty::Str => build_slice_type_di_node(cx, t, unique_type_id, span),
ty::Dynamic(..) => build_dyn_type_di_node(cx, t, unique_type_id),
ty::Foreign(..) => build_foreign_type_di_node(cx, t, unique_type_id),
ty::RawPtr(pointee_type, _) | ty::Ref(_, pointee_type, _) => {
Expand Down
1 change: 1 addition & 0 deletions src/etc/gdb_lookup.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -103,6 +103,7 @@ def __call__(self, valobj):
printer.add(RustType.StdString, StdStringProvider)
printer.add(RustType.StdOsString, StdOsStringProvider)
printer.add(RustType.StdStr, StdStrProvider)
printer.add(RustType.StdBoxStr, StdBoxStrProvider)
printer.add(RustType.StdSlice, StdSliceProvider)
printer.add(RustType.StdVec, StdVecProvider)
printer.add(RustType.StdVecDeque, StdVecDequeProvider)
Expand Down
20 changes: 20 additions & 0 deletions src/etc/gdb_providers.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -142,6 +142,20 @@ def display_hint():
return "array"


class StdBoxStrProvider(printer_base):
def __init__(self, valobj):
self._valobj = valobj
self._length = int(valobj["length"])
self._data_ptr = valobj["data_ptr"]

def to_string(self):
return self._data_ptr.lazy_string(encoding="utf-8", length=self._length)

@staticmethod
def display_hint():
return "string"


class StdVecProvider(printer_base):
def __init__(self, valobj):
self._valobj = valobj
Expand Down Expand Up @@ -203,6 +217,12 @@ def __init__(self, valobj, is_atomic=False):
self._is_atomic = is_atomic
self._ptr = unwrap_unique_or_non_null(valobj["ptr"])
self._value = self._ptr["data" if is_atomic else "value"]
# FIXME(shua): the debuginfo template type should be 'str' not 'u8'
if self._ptr.type.target().name == "alloc::rc::RcInner<str>":
length = self._valobj["ptr"]["pointer"]["length"]
u8_ptr_ty = gdb.Type.pointer(gdb.lookup_type("u8"))
ptr = self._value.address.reinterpret_cast(u8_ptr_ty)
self._value = ptr.lazy_string(encoding="utf-8", length=length)
self._strong = unwrap_scalar_wrappers(self._ptr["strong"])
self._weak = unwrap_scalar_wrappers(self._ptr["weak"]) - 1

Expand Down
3 changes: 3 additions & 0 deletions src/etc/rust_types.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -37,12 +37,14 @@ class RustType(Enum):
StdNonZeroNumber = 29
StdPath = 30
StdPathBuf = 31
StdBoxStr = 32


STD_STRING_REGEX = re.compile(r"^(alloc::([a-z_]+::)+)String$")
STD_STR_REGEX = re.compile(r"^&(mut )?str$")
STD_SLICE_REGEX = re.compile(r"^&(mut )?\[.+\]$")
STD_OS_STRING_REGEX = re.compile(r"^(std::ffi::([a-z_]+::)+)OsString$")
STD_BOX_STR_REGEX = re.compile(r"^(alloc::([a-z_]+::)+)Box<str,.+>$")
STD_VEC_REGEX = re.compile(r"^(alloc::([a-z_]+::)+)Vec<.+>$")
STD_VEC_DEQUE_REGEX = re.compile(r"^(alloc::([a-z_]+::)+)VecDeque<.+>$")
STD_BTREE_SET_REGEX = re.compile(r"^(alloc::([a-z_]+::)+)BTreeSet<.+>$")
Expand All @@ -67,6 +69,7 @@ class RustType(Enum):
RustType.StdString: STD_STRING_REGEX,
RustType.StdOsString: STD_OS_STRING_REGEX,
RustType.StdStr: STD_STR_REGEX,
RustType.StdBoxStr: STD_BOX_STR_REGEX,
RustType.StdSlice: STD_SLICE_REGEX,
RustType.StdVec: STD_VEC_REGEX,
RustType.StdVecDeque: STD_VEC_DEQUE_REGEX,
Expand Down
66 changes: 66 additions & 0 deletions tests/codegen-llvm/debuginfo-unsize-field.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,66 @@
//@ compile-flags:-g -Copt-level=0 -C panic=abort

// Check that debug information for structs with embedded str and [u8] slices is distinct from
// structs with embedded u8

#![crate_type = "lib"]

// NOTE: regex for the CHECK directives,
// depending on the target, u8/usize are basic types or typedefs
// linux: !1 = !DIBasicType(name: "u8",
// win: !1 = !DIDerivedType(tag: .*, name: "u8",
// and references types are
// linux: name: "&debuginfo_unsize_field::Foo"
// win: name: "ref$<debuginfo_unsize_field::Foo>"

// CHECK: ![[U8:[0-9]+]] = !DI{{Basic|Derived}}Type({{.*}}name: "u8",

pub struct Foo {
a: u32,
b: str,
}
// CHECK: !DICompositeType(tag: DW_TAG_structure_type, name: "{{&|ref\$<}}{{[^"]+}}::Foo{{>?}}", {{.*}}elements: ![[FOO_REF_ELEMS:[0-9]+]]
// CHECK: ![[FOO_REF_ELEMS]] = !{![[FOO_REF_PTR:[0-9]+]], ![[FOO_REF_LEN:[0-9]+]]}
// CHECK: ![[FOO_REF_PTR]] = !DIDerivedType(tag: DW_TAG_member, name: "data_ptr", {{.*}}baseType: ![[FOO_PTR:[0-9]+]]
// CHECK: ![[FOO_PTR]] = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: ![[FOO:[0-9]+]]
// CHECK: ![[FOO]] = !DICompositeType(tag: DW_TAG_structure_type, name: "Foo", {{.*}}elements: ![[FOO_ELEMS:[0-9]+]]
// CHECK: ![[FOO_ELEMS]] = !{![[FOO_A:[0-9]+]], ![[FOO_B:[0-9]+]]}
// CHECK: ![[FOO_A]] = !DIDerivedType(tag: DW_TAG_member, name: "a"
// CHECK: ![[FOO_B]] = !DIDerivedType(tag: DW_TAG_member, name: "b", {{.*}}baseType: ![[U8_SLICE:[0-9]+]]
//
// CHECK: ![[U8_SLICE]] = !DICompositeType(tag: DW_TAG_array_type, baseType: ![[U8]], {{.*}}elements: ![[U8_SLICE_ELEMS:[0-9]+]]
// CHECK: ![[U8_SLICE_ELEMS]] = !{![[U8_SLICE_RANGE:[0-9]+]]}
// this is special to embedded slices, there is no upper bound on the number of elements,
// that info is stored in the length metadata for a reference to the parent struct
// CHECK: ![[U8_SLICE_RANGE]] = !DISubrange(count: -1, lowerBound: 0)
//
// CHECK: ![[FOO_REF_LEN]] = !DIDerivedType(tag: DW_TAG_member, name: "length", {{.*}}baseType: ![[USIZE:[0-9]+]]
// CHECK: ![[USIZE]] = !DI{{Basic|Derived}}Type({{.*}}name: "usize"
pub struct Bar {
a: u32,
b: [u8],
}
// CHECK: !DICompositeType(tag: DW_TAG_structure_type, name: "{{&|ref\$<}}{{[^"]+}}::Bar{{>?}}", {{.*}}elements: ![[BAR_REF_ELEMS:[0-9]+]]
// CHECK: ![[BAR_REF_ELEMS]] = !{![[BAR_REF_PTR:[0-9]+]], ![[BAR_REF_LEN:[0-9]+]]}
// CHECK: ![[BAR_REF_PTR]] = !DIDerivedType(tag: DW_TAG_member, name: "data_ptr", {{.*}}baseType: ![[BAR_PTR:[0-9]+]]
// CHECK: ![[BAR_PTR]] = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: ![[BAR:[0-9]+]]
// CHECK: ![[BAR]] = !DICompositeType(tag: DW_TAG_structure_type, name: "Bar", {{.*}}elements: ![[BAR_ELEMS:[0-9]+]]
// CHECK: ![[BAR_ELEMS]] = !{![[BAR_A:[0-9]+]], ![[BAR_B:[0-9]+]]}
// CHECK: ![[BAR_A]] = !DIDerivedType(tag: DW_TAG_member, name: "a"
// CHECK: ![[BAR_B]] = !DIDerivedType(tag: DW_TAG_member, name: "b", {{.*}}baseType: ![[U8_SLICE]]
// CHECK: ![[BAR_REF_LEN]] = !DIDerivedType(tag: DW_TAG_member, name: "length", {{.*}}baseType: ![[USIZE:[0-9]+]]
pub struct Baz {
a: u32,
b: u8,
}
// CHECK: !DIDerivedType(tag: DW_TAG_pointer_type, name: "{{&|ref\$<}}{{[^"]+}}::Baz{{>?}}", {{.*}}baseType: ![[BAZ:[0-9]+]]
// CHECK: ![[BAZ]] = !DICompositeType(tag: DW_TAG_structure_type, name: "Baz", {{.*}}elements: ![[BAZ_ELEMS:[0-9]+]]
// CHECK: ![[BAZ_ELEMS]] = !{![[BAZ_A:[0-9]+]], ![[BAZ_B:[0-9]+]]}
// CHECK: ![[BAZ_A]] = !DIDerivedType(tag: DW_TAG_member, name: "a"
// CHECK: ![[BAZ_B]] = !DIDerivedType(tag: DW_TAG_member, name: "b", {{.*}}baseType: ![[U8]]

#[no_mangle]
pub fn test<'a>(a: &'a Foo, b: &'a Bar, c: &'a Baz) -> &'a u8 {
// just use this somehow so the debuginfo isn't removed
&a.b.as_bytes()[0]
}
15 changes: 15 additions & 0 deletions tests/debuginfo/strings-and-strs.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,6 +1,7 @@
//@ min-gdb-version: 14.0
// LLDB 1800+ tests were not tested in CI, broke, and now are disabled
//@ ignore-lldb
//@ ignore-x86_64-pc-windows-gnu #153238 this target in CI doesn't print Rc<str> correctly

//@ compile-flags:-g
//@ disable-gdb-pretty-printers
Expand All @@ -23,6 +24,12 @@
//@ gdb-command:print str_in_rc
//@ gdb-check:$5 = alloc::rc::Rc<&str, alloc::alloc::Global> {ptr: core::ptr::non_null::NonNull<alloc::rc::RcInner<&str>> {pointer: 0x[...]}, phantom: core::marker::PhantomData<alloc::rc::RcInner<&str>>, alloc: alloc::alloc::Global}

//@ gdb-command:print box_str
//@ gdb-check:$6 = alloc::boxed::Box<str, alloc::alloc::Global> [87, 111, 114, 108, 100]

//@ gdb-command:print rc_str
//@ gdb-check:$7 = alloc::rc::Rc<str, alloc::alloc::Global> {ptr: core::ptr::non_null::NonNull<alloc::rc::RcInner<str>> {pointer: alloc::rc::RcInner<str> {strong: core::cell::Cell<usize> {value: core::cell::UnsafeCell<usize> {value: 1}}, weak: core::cell::Cell<usize> {value: core::cell::UnsafeCell<usize> {value: 1}}, value: 0x[...]}}, phantom: core::marker::PhantomData<alloc::rc::RcInner<str>>, alloc: alloc::alloc::Global}

// === LLDB TESTS ==================================================================================
//@ lldb-command:run
//@ lldb-command:v plain_string
Expand All @@ -40,6 +47,12 @@
//@ lldb-command:v str_in_rc
//@ lldb-check:(alloc::rc::Rc<&str, alloc::alloc::Global>) str_in_rc = strong=1, weak=0 { value = "Hello" { [0] = 'H' [1] = 'e' [2] = 'l' [3] = 'l' [4] = 'o' } }

//@ lldb-command:v box_str
//@ lldb-check:(alloc::boxed::Box<unsigned char[], alloc::alloc::Global>) box_str = { __0 = { pointer = { pointer = { data_ptr = 0x[...] "World" length = 5 } } _marker = } __1 = }

//@ lldb-command:v rc_str
//@ lldb-check:(alloc::rc::Rc<unsigned char[], alloc::alloc::Global>) rc_str = strong=1, weak=0 { value = "World" }

#![allow(unused_variables)]

pub struct Foo<'a> {
Expand All @@ -53,6 +66,8 @@ fn main() {
let str_in_tuple = ("Hello", "World");

let str_in_rc = std::rc::Rc::new("Hello");
let box_str: Box<str> = "World".into();
let rc_str: std::rc::Rc<str> = "World".into();
zzz(); // #break
}

Expand Down
Loading