1 ================================
2 Source Level Debugging with LLVM
3 ================================
11 This document is the central repository for all information pertaining to debug
12 information in LLVM. It describes the :ref:`actual format that the LLVM debug
13 information takes <format>`, which is useful for those interested in creating
14 front-ends or dealing directly with the information. Further, this document
15 provides specific examples of what debug information for C/C++ looks like.
17 Philosophy behind LLVM debugging information
18 --------------------------------------------
20 The idea of the LLVM debugging information is to capture how the important
21 pieces of the source-language's Abstract Syntax Tree map onto LLVM code.
22 Several design aspects have shaped the solution that appears here. The
25 * Debugging information should have very little impact on the rest of the
26 compiler. No transformations, analyses, or code generators should need to
27 be modified because of debugging information.
29 * LLVM optimizations should interact in :ref:`well-defined and easily described
30 ways <intro_debugopt>` with the debugging information.
32 * Because LLVM is designed to support arbitrary programming languages,
33 LLVM-to-LLVM tools should not need to know anything about the semantics of
34 the source-level-language.
36 * Source-level languages are often **widely** different from one another.
37 LLVM should not put any restrictions of the flavor of the source-language,
38 and the debugging information should work with any language.
40 * With code generator support, it should be possible to use an LLVM compiler
41 to compile a program to native machine code and standard debugging
42 formats. This allows compatibility with traditional machine-code level
43 debuggers, like GDB or DBX.
45 The approach used by the LLVM implementation is to use a small set of
46 :ref:`intrinsic functions <format_common_intrinsics>` to define a mapping
47 between LLVM program objects and the source-level objects. The description of
48 the source-level program is maintained in LLVM metadata in an
49 :ref:`implementation-defined format <ccxx_frontend>` (the C/C++ front-end
50 currently uses working draft 7 of the `DWARF 3 standard
51 <http://www.eagercon.com/dwarf/dwarf3std.htm>`_).
53 When a program is being debugged, a debugger interacts with the user and turns
54 the stored debug information into source-language specific information. As
55 such, a debugger must be aware of the source-language, and is thus tied to a
56 specific language or family of languages.
58 Debug information consumers
59 ---------------------------
61 The role of debug information is to provide meta information normally stripped
62 away during the compilation process. This meta information provides an LLVM
63 user a relationship between generated code and the original program source
66 Currently, debug information is consumed by DwarfDebug to produce dwarf
67 information used by the gdb debugger. Other targets could use the same
68 information to produce stabs or other debug forms.
70 It would also be reasonable to use debug information to feed profiling tools
71 for analysis of generated code, or, tools for reconstructing the original
72 source from generated code.
74 TODO - expound a bit more.
78 Debugging optimized code
79 ------------------------
81 An extremely high priority of LLVM debugging information is to make it interact
82 well with optimizations and analysis. In particular, the LLVM debug
83 information provides the following guarantees:
85 * LLVM debug information **always provides information to accurately read
86 the source-level state of the program**, regardless of which LLVM
87 optimizations have been run, and without any modification to the
88 optimizations themselves. However, some optimizations may impact the
89 ability to modify the current state of the program with a debugger, such
90 as setting program variables, or calling functions that have been
93 * As desired, LLVM optimizations can be upgraded to be aware of the LLVM
94 debugging information, allowing them to update the debugging information
95 as they perform aggressive optimizations. This means that, with effort,
96 the LLVM optimizers could optimize debug code just as well as non-debug
99 * LLVM debug information does not prevent optimizations from
100 happening (for example inlining, basic block reordering/merging/cleanup,
101 tail duplication, etc).
103 * LLVM debug information is automatically optimized along with the rest of
104 the program, using existing facilities. For example, duplicate
105 information is automatically merged by the linker, and unused information
106 is automatically removed.
108 Basically, the debug information allows you to compile a program with
109 "``-O0 -g``" and get full debug information, allowing you to arbitrarily modify
110 the program as it executes from a debugger. Compiling a program with
111 "``-O3 -g``" gives you full debug information that is always available and
112 accurate for reading (e.g., you get accurate stack traces despite tail call
113 elimination and inlining), but you might lose the ability to modify the program
114 and call functions where were optimized out of the program, or inlined away
117 :ref:`LLVM test suite <test-suite-quickstart>` provides a framework to test
118 optimizer's handling of debugging information. It can be run like this:
122 % cd llvm/projects/test-suite/MultiSource/Benchmarks # or some other level
125 This will test impact of debugging information on optimization passes. If
126 debugging information influences optimization passes then it will be reported
127 as a failure. See :doc:`TestingGuide` for more information on LLVM test
128 infrastructure and how to run various tests.
132 Debugging information format
133 ============================
135 LLVM debugging information has been carefully designed to make it possible for
136 the optimizer to optimize the program and debugging information without
137 necessarily having to know anything about debugging information. In
138 particular, the use of metadata avoids duplicated debugging information from
139 the beginning, and the global dead code elimination pass automatically deletes
140 debugging information for a function if it decides to delete the function.
142 To do this, most of the debugging information (descriptors for types,
143 variables, functions, source files, etc) is inserted by the language front-end
144 in the form of LLVM metadata.
146 Debug information is designed to be agnostic about the target debugger and
147 debugging information representation (e.g. DWARF/Stabs/etc). It uses a generic
148 pass to decode the information that represents variables, types, functions,
149 namespaces, etc: this allows for arbitrary source-language semantics and
150 type-systems to be used, as long as there is a module written for the target
151 debugger to interpret the information.
153 To provide basic functionality, the LLVM debugger does have to make some
154 assumptions about the source-level language being debugged, though it keeps
155 these to a minimum. The only common features that the LLVM debugger assumes
156 exist are :ref:`source files <format_files>`, and :ref:`program objects
157 <format_global_variables>`. These abstract objects are used by a debugger to
158 form stack traces, show information about local variables, etc.
160 This section of the documentation first describes the representation aspects
161 common to any source-language. :ref:`ccxx_frontend` describes the data layout
162 conventions used by the C and C++ front-ends.
164 Debug information descriptors
165 -----------------------------
167 In consideration of the complexity and volume of debug information, LLVM
168 provides a specification for well formed debug descriptors.
170 Consumers of LLVM debug information expect the descriptors for program objects
171 to start in a canonical format, but the descriptors can include additional
172 information appended at the end that is source-language specific. All debugging
173 information objects start with a tag to indicate what type of object it is.
174 The source-language is allowed to define its own objects, by using unreserved
175 tag numbers. We recommend using with tags in the range 0x1000 through 0x2000
176 (there is a defined ``enum DW_TAG_user_base = 0x1000``.)
178 The fields of debug descriptors used internally by LLVM are restricted to only
179 the simple data types ``i32``, ``i1``, ``float``, ``double``, ``mdstring`` and
189 <a name="LLVMDebugVersion">The first field of a descriptor is always an
190 ``i32`` containing a tag value identifying the content of the descriptor.
191 The remaining fields are specific to the descriptor. The values of tags are
192 loosely bound to the tag values of DWARF information entries. However, that
193 does not restrict the use of the information supplied to DWARF targets.
195 The details of the various descriptors follow.
197 Compile unit descriptors
198 ^^^^^^^^^^^^^^^^^^^^^^^^
203 i32, ;; Tag = 17 (DW_TAG_compile_unit)
204 metadata, ;; Source directory (including trailing slash) & file pair
205 i32, ;; DWARF language identifier (ex. DW_LANG_C89)
206 metadata ;; Producer (ex. "4.0.1 LLVM (LLVM research group)")
207 i1, ;; True if this is optimized.
209 i32 ;; Runtime version
210 metadata ;; List of enums types
211 metadata ;; List of retained types
212 metadata ;; List of subprograms
213 metadata ;; List of global variables
214 metadata ;; List of imported entities
215 metadata ;; Split debug filename
216 i32 ;; Debug info emission kind (1 = Full Debug Info, 2 = Line Tables Only)
219 These descriptors contain a source language ID for the file (we use the DWARF
220 3.0 ID numbers, such as ``DW_LANG_C89``, ``DW_LANG_C_plus_plus``,
221 ``DW_LANG_Cobol74``, etc), a reference to a metadata node containing a pair of
222 strings for the source file name and the working directory, as well as an
223 identifier string for the compiler that produced it.
225 Compile unit descriptors provide the root context for objects declared in a
226 specific compilation unit. File descriptors are defined using this context.
227 These descriptors are collected by a named metadata ``!llvm.dbg.cu``. They
228 keep track of subprograms, global variables, type information, and imported
229 entities (declarations and namespaces).
239 i32, ;; Tag = 41 (DW_TAG_file_type)
240 metadata, ;; Source directory (including trailing slash) & file pair
243 These descriptors contain information for a file. Global variables and top
244 level functions would be defined using this context. File descriptors also
245 provide context for source line correspondence.
247 Each input file is encoded as a separate file descriptor in LLVM debugging
250 .. _format_global_variables:
252 Global variable descriptors
253 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
258 i32, ;; Tag = 52 (DW_TAG_variable)
259 i32, ;; Unused field.
260 metadata, ;; Reference to context descriptor
262 metadata, ;; Display name (fully qualified C++ name)
263 metadata, ;; MIPS linkage name (for C++)
264 metadata, ;; Reference to file where defined
265 i32, ;; Line number where defined
266 metadata, ;; Reference to type descriptor
267 i1, ;; True if the global is local to compile unit (static)
268 i1, ;; True if the global is defined in the compile unit (not extern)
269 {}*, ;; Reference to the global variable
270 metadata, ;; The static member declaration, if any
273 These descriptors provide debug information about global variables. They
274 provide details such as name, type and where the variable is defined. All
275 global variables are collected inside the named metadata ``!llvm.dbg.cu``.
277 .. _format_subprograms:
279 Subprogram descriptors
280 ^^^^^^^^^^^^^^^^^^^^^^
285 i32, ;; Tag = 46 (DW_TAG_subprogram)
286 metadata, ;; Source directory (including trailing slash) & file pair
287 metadata, ;; Reference to context descriptor
289 metadata, ;; Display name (fully qualified C++ name)
290 metadata, ;; MIPS linkage name (for C++)
291 i32, ;; Line number where defined
292 metadata, ;; Reference to type descriptor
293 i1, ;; True if the global is local to compile unit (static)
294 i1, ;; True if the global is defined in the compile unit (not extern)
295 i32, ;; Virtuality, e.g. dwarf::DW_VIRTUALITY__virtual
296 i32, ;; Index into a virtual function
297 metadata, ;; indicates which base type contains the vtable pointer for the
299 i32, ;; Flags - Artificial, Private, Protected, Explicit, Prototyped.
301 {}*, ;; Reference to the LLVM function
302 metadata, ;; Lists function template parameters
303 metadata, ;; Function declaration descriptor
304 metadata, ;; List of function variables
305 i32 ;; Line number where the scope of the subprogram begins
308 These descriptors provide debug information about functions, methods and
309 subprograms. They provide details such as name, return types and the source
310 location where the subprogram is defined.
318 i32, ;; Tag = 11 (DW_TAG_lexical_block)
319 metadata, ;; Source directory (including trailing slash) & file pair
320 metadata, ;; Reference to context descriptor
322 i32, ;; Column number
323 i32 ;; Unique ID to identify blocks from a template function
326 This descriptor provides debug information about nested blocks within a
327 subprogram. The line number and column numbers are used to dinstinguish two
328 lexical blocks at same depth.
333 i32, ;; Tag = 11 (DW_TAG_lexical_block)
334 metadata, ;; Source directory (including trailing slash) & file pair
335 metadata ;; Reference to the scope we're annotating with a file change
336 i32, ;; DWARF path discriminator value
339 This descriptor provides a wrapper around a lexical scope to handle file
340 changes in the middle of a lexical block.
342 .. _format_basic_type:
344 Basic type descriptors
345 ^^^^^^^^^^^^^^^^^^^^^^
350 i32, ;; Tag = 36 (DW_TAG_base_type)
351 metadata, ;; Source directory (including trailing slash) & file pair (may be null)
352 metadata, ;; Reference to context
353 metadata, ;; Name (may be "" for anonymous types)
354 i32, ;; Line number where defined (may be 0)
356 i64, ;; Alignment in bits
357 i64, ;; Offset in bits
359 i32 ;; DWARF type encoding
362 These descriptors define primitive types used in the code. Example ``int``,
363 ``bool`` and ``float``. The context provides the scope of the type, which is
364 usually the top level. Since basic types are not usually user defined the
365 context and line number can be left as NULL and 0. The size, alignment and
366 offset are expressed in bits and can be 64 bit values. The alignment is used
367 to round the offset when embedded in a :ref:`composite type
368 <format_composite_type>` (example to keep float doubles on 64 bit boundaries).
369 The offset is the bit offset if embedded in a :ref:`composite type
370 <format_composite_type>`.
372 The type encoding provides the details of the type. The values are typically
373 one of the following:
381 DW_ATE_signed_char = 6
383 DW_ATE_unsigned_char = 8
385 .. _format_derived_type:
387 Derived type descriptors
388 ^^^^^^^^^^^^^^^^^^^^^^^^
393 i32, ;; Tag (see below)
394 metadata, ;; Source directory (including trailing slash) & file pair (may be null)
395 metadata, ;; Reference to context
396 metadata, ;; Name (may be "" for anonymous types)
397 i32, ;; Line number where defined (may be 0)
399 i64, ;; Alignment in bits
400 i64, ;; Offset in bits
401 i32, ;; Flags to encode attributes, e.g. private
402 metadata, ;; Reference to type derived from
403 metadata, ;; (optional) Name of the Objective C property associated with
404 ;; Objective-C an ivar, or the type of which this
405 ;; pointer-to-member is pointing to members of.
406 metadata, ;; (optional) Name of the Objective C property getter selector.
407 metadata, ;; (optional) Name of the Objective C property setter selector.
408 i32 ;; (optional) Objective C property attributes.
411 These descriptors are used to define types derived from other types. The value
412 of the tag varies depending on the meaning. The following are possible tag
417 DW_TAG_formal_parameter = 5
419 DW_TAG_pointer_type = 15
420 DW_TAG_reference_type = 16
422 DW_TAG_ptr_to_member_type = 31
423 DW_TAG_const_type = 38
424 DW_TAG_volatile_type = 53
425 DW_TAG_restrict_type = 55
427 ``DW_TAG_member`` is used to define a member of a :ref:`composite type
428 <format_composite_type>` or :ref:`subprogram <format_subprograms>`. The type
429 of the member is the :ref:`derived type <format_derived_type>`.
430 ``DW_TAG_formal_parameter`` is used to define a member which is a formal
431 argument of a subprogram.
433 ``DW_TAG_typedef`` is used to provide a name for the derived type.
435 ``DW_TAG_pointer_type``, ``DW_TAG_reference_type``, ``DW_TAG_const_type``,
436 ``DW_TAG_volatile_type`` and ``DW_TAG_restrict_type`` are used to qualify the
437 :ref:`derived type <format_derived_type>`.
439 :ref:`Derived type <format_derived_type>` location can be determined from the
440 context and line number. The size, alignment and offset are expressed in bits
441 and can be 64 bit values. The alignment is used to round the offset when
442 embedded in a :ref:`composite type <format_composite_type>` (example to keep
443 float doubles on 64 bit boundaries.) The offset is the bit offset if embedded
444 in a :ref:`composite type <format_composite_type>`.
446 Note that the ``void *`` type is expressed as a type derived from NULL.
448 .. _format_composite_type:
450 Composite type descriptors
451 ^^^^^^^^^^^^^^^^^^^^^^^^^^
456 i32, ;; Tag (see below)
457 metadata, ;; Source directory (including trailing slash) & file pair (may be null)
458 metadata, ;; Reference to context
459 metadata, ;; Name (may be "" for anonymous types)
460 i32, ;; Line number where defined (may be 0)
462 i64, ;; Alignment in bits
463 i64, ;; Offset in bits
465 metadata, ;; Reference to type derived from
466 metadata, ;; Reference to array of member descriptors
467 i32, ;; Runtime languages
468 metadata, ;; Base type containing the vtable pointer for this type
469 metadata, ;; Template parameters
470 metadata ;; A unique identifier for type uniquing purpose (may be null)
473 These descriptors are used to define types that are composed of 0 or more
474 elements. The value of the tag varies depending on the meaning. The following
475 are possible tag values:
479 DW_TAG_array_type = 1
480 DW_TAG_enumeration_type = 4
481 DW_TAG_structure_type = 19
482 DW_TAG_union_type = 23
483 DW_TAG_subroutine_type = 21
484 DW_TAG_inheritance = 28
486 The vector flag indicates that an array type is a native packed vector.
488 The members of array types (tag = ``DW_TAG_array_type``) are
489 :ref:`subrange descriptors <format_subrange>`, each
490 representing the range of subscripts at that level of indexing.
492 The members of enumeration types (tag = ``DW_TAG_enumeration_type``) are
493 :ref:`enumerator descriptors <format_enumerator>`, each representing the
494 definition of enumeration value for the set. All enumeration type descriptors
495 are collected inside the named metadata ``!llvm.dbg.cu``.
497 The members of structure (tag = ``DW_TAG_structure_type``) or union (tag =
498 ``DW_TAG_union_type``) types are any one of the :ref:`basic
499 <format_basic_type>`, :ref:`derived <format_derived_type>` or :ref:`composite
500 <format_composite_type>` type descriptors, each representing a field member of
501 the structure or union.
503 For C++ classes (tag = ``DW_TAG_structure_type``), member descriptors provide
504 information about base classes, static members and member functions. If a
505 member is a :ref:`derived type descriptor <format_derived_type>` and has a tag
506 of ``DW_TAG_inheritance``, then the type represents a base class. If the member
507 of is a :ref:`global variable descriptor <format_global_variables>` then it
508 represents a static member. And, if the member is a :ref:`subprogram
509 descriptor <format_subprograms>` then it represents a member function. For
510 static members and member functions, ``getName()`` returns the members link or
511 the C++ mangled name. ``getDisplayName()`` the simplied version of the name.
513 The first member of subroutine (tag = ``DW_TAG_subroutine_type``) type elements
514 is the return type for the subroutine. The remaining elements are the formal
515 arguments to the subroutine.
517 :ref:`Composite type <format_composite_type>` location can be determined from
518 the context and line number. The size, alignment and offset are expressed in
519 bits and can be 64 bit values. The alignment is used to round the offset when
520 embedded in a :ref:`composite type <format_composite_type>` (as an example, to
521 keep float doubles on 64 bit boundaries). The offset is the bit offset if
522 embedded in a :ref:`composite type <format_composite_type>`.
532 i32, ;; Tag = 33 (DW_TAG_subrange_type)
537 These descriptors are used to define ranges of array subscripts for an array
538 :ref:`composite type <format_composite_type>`. The low value defines the lower
539 bounds typically zero for C/C++. The high value is the upper bounds. Values
540 are 64 bit. ``High - Low + 1`` is the size of the array. If ``Low > High``
541 the array bounds are not included in generated debugging information.
543 .. _format_enumerator:
545 Enumerator descriptors
546 ^^^^^^^^^^^^^^^^^^^^^^
551 i32, ;; Tag = 40 (DW_TAG_enumerator)
556 These descriptors are used to define members of an enumeration :ref:`composite
557 type <format_composite_type>`, it associates the name to the value.
565 i32, ;; Tag (see below)
568 metadata, ;; Reference to file where defined
569 i32, ;; 24 bit - Line number where defined
570 ;; 8 bit - Argument number. 1 indicates 1st argument.
571 metadata, ;; Reference to the type descriptor
573 metadata ;; (optional) Reference to inline location
574 metadata ;; (optional) Reference to a complex expression.
577 These descriptors are used to define variables local to a sub program. The
578 value of the tag depends on the usage of the variable:
582 DW_TAG_auto_variable = 256
583 DW_TAG_arg_variable = 257
585 An auto variable is any variable declared in the body of the function. An
586 argument variable is any variable that appears as a formal argument to the
589 The context is either the subprogram or block where the variable is defined.
590 Name the source variable name. Context and line indicate where the variable
591 was defined. Type descriptor defines the declared type of the variable.
598 i32, ;; DW_TAG_expression
602 Complex expressions describe variable storage locations in terms of
603 prefix-notated DWARF expressions. Currently the only supported
604 operators are ``DW_OP_plus``, ``DW_OP_deref``, and ``DW_OP_piece``.
606 The ``DW_OP_piece`` operator is used for (typically larger aggregate)
607 variables that are fragmented across several locations. It takes two
608 i32 arguments, an offset and a size in bytes to describe which piece
609 of the variable is at this location.
612 .. _format_common_intrinsics:
614 Debugger intrinsic functions
615 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
617 LLVM uses several intrinsic functions (name prefixed with "``llvm.dbg``") to
618 provide debug information at various points in generated code.
625 void %llvm.dbg.declare(metadata, metadata)
627 This intrinsic provides information about a local element (e.g., variable).
628 The first argument is metadata holding the alloca for the variable. The second
629 argument is metadata containing a description of the variable.
636 void %llvm.dbg.value(metadata, i64, metadata)
638 This intrinsic provides information when a user source variable is set to a new
639 value. The first argument is the new value (wrapped as metadata). The second
640 argument is the offset in the user source variable where the new value is
641 written. The third argument is metadata containing a description of the user
644 Object lifetimes and scoping
645 ============================
647 In many languages, the local variables in functions can have their lifetimes or
648 scopes limited to a subset of a function. In the C family of languages, for
649 example, variables are only live (readable and writable) within the source
650 block that they are defined in. In functional languages, values are only
651 readable after they have been defined. Though this is a very obvious concept,
652 it is non-trivial to model in LLVM, because it has no notion of scoping in this
653 sense, and does not want to be tied to a language's scoping rules.
655 In order to handle this, the LLVM debug format uses the metadata attached to
656 llvm instructions to encode line number and scoping information. Consider the
657 following C fragment, for example:
671 Compiled to LLVM, this function would be represented like this:
675 define void @foo() #0 {
677 %X = alloca i32, align 4
678 %Y = alloca i32, align 4
679 %Z = alloca i32, align 4
680 call void @llvm.dbg.declare(metadata !{i32* %X}, metadata !10), !dbg !12
681 ; [debug line = 2:7] [debug variable = X]
682 store i32 21, i32* %X, align 4, !dbg !12
683 call void @llvm.dbg.declare(metadata !{i32* %Y}, metadata !13), !dbg !14
684 ; [debug line = 3:7] [debug variable = Y]
685 store i32 22, i32* %Y, align 4, !dbg !14
686 call void @llvm.dbg.declare(metadata !{i32* %Z}, metadata !15), !dbg !17
687 ; [debug line = 5:9] [debug variable = Z]
688 store i32 23, i32* %Z, align 4, !dbg !17
689 %0 = load i32* %X, align 4, !dbg !18
691 store i32 %0, i32* %Z, align 4, !dbg !18
692 %1 = load i32* %Y, align 4, !dbg !19
694 store i32 %1, i32* %X, align 4, !dbg !19
698 ; Function Attrs: nounwind readnone
699 declare void @llvm.dbg.declare(metadata, metadata) #1
701 attributes #0 = { nounwind ssp uwtable "less-precise-fpmad"="false"
702 "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf"
703 "no-infs-fp-math"="false" "no-nans-fp-math"="false"
704 "stack-protector-buffer-size"="8" "unsafe-fp-math"="false"
705 "use-soft-float"="false" }
706 attributes #1 = { nounwind readnone }
709 !llvm.module.flags = !{!8}
712 !0 = metadata !{i32 786449, metadata !1, i32 12,
713 metadata !"clang version 3.4 (trunk 193128) (llvm/trunk 193139)",
714 i1 false, metadata !"", i32 0, metadata !2, metadata !2, metadata !3,
715 metadata !2, metadata !2, metadata !""} ; [ DW_TAG_compile_unit ] \
716 [/private/tmp/foo.c] \
718 !1 = metadata !{metadata !"t.c", metadata !"/private/tmp"}
719 !2 = metadata !{i32 0}
720 !3 = metadata !{metadata !4}
721 !4 = metadata !{i32 786478, metadata !1, metadata !5, metadata !"foo",
722 metadata !"foo", metadata !"", i32 1, metadata !6,
723 i1 false, i1 true, i32 0, i32 0, null, i32 0, i1 false,
724 void ()* @foo, null, null, metadata !2, i32 1}
725 ; [ DW_TAG_subprogram ] [line 1] [def] [foo]
726 !5 = metadata !{i32 786473, metadata !1} ; [ DW_TAG_file_type ] \
728 !6 = metadata !{i32 786453, i32 0, null, metadata !"", i32 0, i64 0, i64 0,
729 i64 0, i32 0, null, metadata !7, i32 0, null, null, null}
730 ; [ DW_TAG_subroutine_type ] \
731 [line 0, size 0, align 0, offset 0] [from ]
732 !7 = metadata !{null}
733 !8 = metadata !{i32 2, metadata !"Dwarf Version", i32 2}
734 !9 = metadata !{metadata !"clang version 3.4 (trunk 193128) (llvm/trunk 193139)"}
735 !10 = metadata !{i32 786688, metadata !4, metadata !"X", metadata !5, i32 2,
736 metadata !11, i32 0, i32 0} ; [ DW_TAG_auto_variable ] [X] \
738 !11 = metadata !{i32 786468, null, null, metadata !"int", i32 0, i64 32,
739 i64 32, i64 0, i32 0, i32 5} ; [ DW_TAG_base_type ] [int] \
740 [line 0, size 32, align 32, offset 0, enc DW_ATE_signed]
741 !12 = metadata !{i32 2, i32 0, metadata !4, null}
742 !13 = metadata !{i32 786688, metadata !4, metadata !"Y", metadata !5, i32 3,
743 metadata !11, i32 0, i32 0} ; [ DW_TAG_auto_variable ] [Y] \
745 !14 = metadata !{i32 3, i32 0, metadata !4, null}
746 !15 = metadata !{i32 786688, metadata !16, metadata !"Z", metadata !5, i32 5,
747 metadata !11, i32 0, i32 0} ; [ DW_TAG_auto_variable ] [Z] \
749 !16 = metadata !{i32 786443, metadata !1, metadata !4, i32 4, i32 0, i32 0} \
750 ; [ DW_TAG_lexical_block ] [/private/tmp/t.c]
751 !17 = metadata !{i32 5, i32 0, metadata !16, null}
752 !18 = metadata !{i32 6, i32 0, metadata !16, null}
753 !19 = metadata !{i32 8, i32 0, metadata !4, null} ; [ DW_TAG_imported_declaration ]
754 !20 = metadata !{i32 9, i32 0, metadata !4, null}
756 This example illustrates a few important details about LLVM debugging
757 information. In particular, it shows how the ``llvm.dbg.declare`` intrinsic and
758 location information, which are attached to an instruction, are applied
759 together to allow a debugger to analyze the relationship between statements,
760 variable definitions, and the code used to implement the function.
764 call void @llvm.dbg.declare(metadata !{i32* %X}, metadata !10), !dbg !12
765 ; [debug line = 2:7] [debug variable = X]
767 The first intrinsic ``%llvm.dbg.declare`` encodes debugging information for the
768 variable ``X``. The metadata ``!dbg !12`` attached to the intrinsic provides
769 scope information for the variable ``X``.
773 !12 = metadata !{i32 2, i32 0, metadata !4, null}
774 !4 = metadata !{i32 786478, metadata !1, metadata !5, metadata !"foo",
775 metadata !"foo", metadata !"", i32 1, metadata !6,
776 i1 false, i1 true, i32 0, i32 0, null, i32 0, i1 false,
777 void ()* @foo, null, null, metadata !2, i32 1}
778 ; [ DW_TAG_subprogram ] [line 1] [def] [foo]
780 Here ``!12`` is metadata providing location information. It has four fields:
781 line number, column number, scope, and original scope. The original scope
782 represents inline location if this instruction is inlined inside a caller, and
783 is null otherwise. In this example, scope is encoded by ``!4``, a
784 :ref:`subprogram descriptor <format_subprograms>`. This way the location
785 information attached to the intrinsics indicates that the variable ``X`` is
786 declared at line number 2 at a function level scope in function ``foo``.
788 Now lets take another example.
792 call void @llvm.dbg.declare(metadata !{i32* %Z}, metadata !15), !dbg !17
793 ; [debug line = 5:9] [debug variable = Z]
795 The third intrinsic ``%llvm.dbg.declare`` encodes debugging information for
796 variable ``Z``. The metadata ``!dbg !17`` attached to the intrinsic provides
797 scope information for the variable ``Z``.
801 !16 = metadata !{i32 786443, metadata !1, metadata !4, i32 4, i32 0, i32 0} \
802 ; [ DW_TAG_lexical_block ] [/private/tmp/t.c]
803 !17 = metadata !{i32 5, i32 0, metadata !16, null}
805 Here ``!15`` indicates that ``Z`` is declared at line number 5 and
806 column number 0 inside of lexical scope ``!16``. The lexical scope itself
807 resides inside of subprogram ``!4`` described above.
809 The scope information attached with each instruction provides a straightforward
810 way to find instructions covered by a scope.
814 C/C++ front-end specific debug information
815 ==========================================
817 The C and C++ front-ends represent information about the program in a format
818 that is effectively identical to `DWARF 3.0
819 <http://www.eagercon.com/dwarf/dwarf3std.htm>`_ in terms of information
820 content. This allows code generators to trivially support native debuggers by
821 generating standard dwarf information, and contains enough information for
822 non-dwarf targets to translate it as needed.
824 This section describes the forms used to represent C and C++ programs. Other
825 languages could pattern themselves after this (which itself is tuned to
826 representing programs in the same way that DWARF 3 does), or they could choose
827 to provide completely different forms if they don't fit into the DWARF model.
828 As support for debugging information gets added to the various LLVM
829 source-language front-ends, the information used should be documented here.
831 The following sections provide examples of various C/C++ constructs and the
832 debug information that would best describe those constructs.
834 C/C++ source file information
835 -----------------------------
837 Given the source files ``MySource.cpp`` and ``MyHeader.h`` located in the
838 directory ``/Users/mine/sources``, the following code:
842 #include "MyHeader.h"
844 int main(int argc, char *argv[]) {
848 a C/C++ front-end would generate the following descriptors:
854 ;; Define the compile unit for the main source file "/Users/mine/sources/MySource.cpp".
858 metadata !1, ;; File/directory name
859 i32 4, ;; Language Id
860 metadata !"clang version 3.4 ",
861 i1 false, ;; Optimized compile unit
862 metadata !"", ;; Compiler flags
863 i32 0, ;; Runtime version
864 metadata !2, ;; Enumeration types
865 metadata !2, ;; Retained types
866 metadata !3, ;; Subprograms
867 metadata !2, ;; Global variables
868 metadata !2, ;; Imported entities (declarations and namespaces)
869 metadata !"" ;; Split debug filename
870 1, ;; Full debug info
874 ;; Define the file for the file "/Users/mine/sources/MySource.cpp".
877 metadata !"MySource.cpp",
878 metadata !"/Users/mine/sources"
886 ;; Define the file for the file "/Users/mine/sources/Myheader.h"
893 metadata !"./MyHeader.h",
894 metadata !"/Users/mine/sources",
899 ``llvm::Instruction`` provides easy access to metadata attached with an
900 instruction. One can extract line number information encoded in LLVM IR using
901 ``Instruction::getMetadata()`` and ``DILocation::getLineNumber()``.
905 if (MDNode *N = I->getMetadata("dbg")) { // Here I is an LLVM instruction
906 DILocation Loc(N); // DILocation is in DebugInfo.h
907 unsigned Line = Loc.getLineNumber();
908 StringRef File = Loc.getFilename();
909 StringRef Dir = Loc.getDirectory();
912 C/C++ global variable information
913 ---------------------------------
915 Given an integer global variable declared as follows:
921 a C/C++ front-end would generate the following descriptors:
926 ;; Define the global itself.
928 %MyGlobal = global int 100
931 ;; List of debug info of globals
935 ;; Define the compile unit.
940 metadata !"foo.cpp", ;; File
941 metadata !"/Volumes/Data/tmp", ;; Directory
942 metadata !"clang version 3.1 ", ;; Producer
943 i1 true, ;; Deprecated field
944 i1 false, ;; "isOptimized"?
945 metadata !"", ;; Flags
946 i32 0, ;; Runtime Version
947 metadata !1, ;; Enum Types
948 metadata !1, ;; Retained Types
949 metadata !1, ;; Subprograms
950 metadata !3, ;; Global Variables
951 metadata !1, ;; Imported entities
952 "", ;; Split debug filename
953 1, ;; Full debug info
954 } ; [ DW_TAG_compile_unit ]
956 ;; The Array of Global Variables
962 ;; Define the global variable itself.
968 metadata !"MyGlobal", ;; Name
969 metadata !"MyGlobal", ;; Display Name
970 metadata !"", ;; Linkage Name
974 i32 0, ;; IsLocalToUnit
975 i32 1, ;; IsDefinition
976 i32* @MyGlobal, ;; LLVM-IR Value
977 null ;; Static member declaration
978 } ; [ DW_TAG_variable ]
984 metadata !"foo.cpp", ;; File
985 metadata !"/Volumes/Data/tmp", ;; Directory
989 metadata !5 ;; Unused
990 } ; [ DW_TAG_file_type ]
999 metadata !"int", ;; Name
1001 i64 32, ;; Size in Bits
1002 i64 32, ;; Align in Bits
1006 } ; [ DW_TAG_base_type ]
1008 C/C++ function information
1009 --------------------------
1011 Given a function declared as follows:
1015 int main(int argc, char *argv[]) {
1019 a C/C++ front-end would generate the following descriptors:
1021 .. code-block:: llvm
1024 ;; Define the anchor for subprograms.
1028 metadata !1, ;; File
1029 metadata !1, ;; Context
1030 metadata !"main", ;; Name
1031 metadata !"main", ;; Display name
1032 metadata !"main", ;; Linkage name
1033 i32 1, ;; Line number
1034 metadata !4, ;; Type
1035 i1 false, ;; Is local
1036 i1 true, ;; Is definition
1037 i32 0, ;; Virtuality attribute, e.g. pure virtual function
1038 i32 0, ;; Index into virtual table for C++ methods
1039 i32 0, ;; Type that holds virtual table.
1041 i1 false, ;; True if this function is optimized
1042 Function *, ;; Pointer to llvm::Function
1043 null, ;; Function template parameters
1044 null, ;; List of function variables (emitted when optimizing)
1045 1 ;; Line number of the opening '{' of the function
1048 ;; Define the subprogram itself.
1050 define i32 @main(i32 %argc, i8** %argv) {
1057 The following are the basic type descriptors for C/C++ core types:
1062 .. code-block:: llvm
1068 metadata !"bool", ;; Name
1069 i32 0, ;; Line number
1070 i64 8, ;; Size in Bits
1071 i64 8, ;; Align in Bits
1072 i64 0, ;; Offset in Bits
1080 .. code-block:: llvm
1086 metadata !"char", ;; Name
1087 i32 0, ;; Line number
1088 i64 8, ;; Size in Bits
1089 i64 8, ;; Align in Bits
1090 i64 0, ;; Offset in Bits
1098 .. code-block:: llvm
1104 metadata !"unsigned char",
1105 i32 0, ;; Line number
1106 i64 8, ;; Size in Bits
1107 i64 8, ;; Align in Bits
1108 i64 0, ;; Offset in Bits
1116 .. code-block:: llvm
1122 metadata !"short int",
1123 i32 0, ;; Line number
1124 i64 16, ;; Size in Bits
1125 i64 16, ;; Align in Bits
1126 i64 0, ;; Offset in Bits
1134 .. code-block:: llvm
1140 metadata !"short unsigned int",
1141 i32 0, ;; Line number
1142 i64 16, ;; Size in Bits
1143 i64 16, ;; Align in Bits
1144 i64 0, ;; Offset in Bits
1152 .. code-block:: llvm
1158 metadata !"int", ;; Name
1159 i32 0, ;; Line number
1160 i64 32, ;; Size in Bits
1161 i64 32, ;; Align in Bits
1162 i64 0, ;; Offset in Bits
1170 .. code-block:: llvm
1176 metadata !"unsigned int",
1177 i32 0, ;; Line number
1178 i64 32, ;; Size in Bits
1179 i64 32, ;; Align in Bits
1180 i64 0, ;; Offset in Bits
1188 .. code-block:: llvm
1194 metadata !"long long int",
1195 i32 0, ;; Line number
1196 i64 64, ;; Size in Bits
1197 i64 64, ;; Align in Bits
1198 i64 0, ;; Offset in Bits
1206 .. code-block:: llvm
1212 metadata !"long long unsigned int",
1213 i32 0, ;; Line number
1214 i64 64, ;; Size in Bits
1215 i64 64, ;; Align in Bits
1216 i64 0, ;; Offset in Bits
1224 .. code-block:: llvm
1231 i32 0, ;; Line number
1232 i64 32, ;; Size in Bits
1233 i64 32, ;; Align in Bits
1234 i64 0, ;; Offset in Bits
1242 .. code-block:: llvm
1248 metadata !"double",;; Name
1249 i32 0, ;; Line number
1250 i64 64, ;; Size in Bits
1251 i64 64, ;; Align in Bits
1252 i64 0, ;; Offset in Bits
1260 Given the following as an example of C/C++ derived type:
1264 typedef const int *IntPtr;
1266 a C/C++ front-end would generate the following descriptors:
1268 .. code-block:: llvm
1271 ;; Define the typedef "IntPtr".
1275 metadata !3, ;; File
1276 metadata !1, ;; Context
1277 metadata !"IntPtr", ;; Name
1278 i32 0, ;; Line number
1279 i64 0, ;; Size in bits
1280 i64 0, ;; Align in bits
1281 i64 0, ;; Offset in bits
1283 metadata !4 ;; Derived From type
1286 ;; Define the pointer type.
1292 metadata !"", ;; Name
1293 i32 0, ;; Line number
1294 i64 64, ;; Size in bits
1295 i64 64, ;; Align in bits
1296 i64 0, ;; Offset in bits
1298 metadata !5 ;; Derived From type
1301 ;; Define the const type.
1307 metadata !"", ;; Name
1308 i32 0, ;; Line number
1309 i64 0, ;; Size in bits
1310 i64 0, ;; Align in bits
1311 i64 0, ;; Offset in bits
1313 metadata !6 ;; Derived From type
1316 ;; Define the int type.
1322 metadata !"int", ;; Name
1323 i32 0, ;; Line number
1324 i64 32, ;; Size in bits
1325 i64 32, ;; Align in bits
1326 i64 0, ;; Offset in bits
1331 C/C++ struct/union types
1332 ------------------------
1334 Given the following as an example of C/C++ struct type:
1344 a C/C++ front-end would generate the following descriptors:
1346 .. code-block:: llvm
1349 ;; Define basic type for unsigned int.
1355 metadata !"unsigned int",
1356 i32 0, ;; Line number
1357 i64 32, ;; Size in Bits
1358 i64 32, ;; Align in Bits
1359 i64 0, ;; Offset in Bits
1364 ;; Define composite type for struct Color.
1368 metadata !1, ;; Compile unit
1370 metadata !"Color", ;; Name
1371 i32 1, ;; Line number
1372 i64 96, ;; Size in bits
1373 i64 32, ;; Align in bits
1374 i64 0, ;; Offset in bits
1376 null, ;; Derived From
1377 metadata !3, ;; Elements
1378 i32 0, ;; Runtime Language
1379 null, ;; Base type containing the vtable pointer for this type
1380 null ;; Template parameters
1384 ;; Define the Red field.
1388 metadata !1, ;; File
1389 metadata !1, ;; Context
1390 metadata !"Red", ;; Name
1391 i32 2, ;; Line number
1392 i64 32, ;; Size in bits
1393 i64 32, ;; Align in bits
1394 i64 0, ;; Offset in bits
1396 metadata !5 ;; Derived From type
1400 ;; Define the Green field.
1404 metadata !1, ;; File
1405 metadata !1, ;; Context
1406 metadata !"Green", ;; Name
1407 i32 3, ;; Line number
1408 i64 32, ;; Size in bits
1409 i64 32, ;; Align in bits
1410 i64 32, ;; Offset in bits
1412 metadata !5 ;; Derived From type
1416 ;; Define the Blue field.
1420 metadata !1, ;; File
1421 metadata !1, ;; Context
1422 metadata !"Blue", ;; Name
1423 i32 4, ;; Line number
1424 i64 32, ;; Size in bits
1425 i64 32, ;; Align in bits
1426 i64 64, ;; Offset in bits
1428 metadata !5 ;; Derived From type
1432 ;; Define the array of fields used by the composite type Color.
1434 !3 = metadata !{metadata !4, metadata !6, metadata !7}
1436 C/C++ enumeration types
1437 -----------------------
1439 Given the following as an example of C/C++ enumeration type:
1449 a C/C++ front-end would generate the following descriptors:
1451 .. code-block:: llvm
1454 ;; Define composite type for enum Trees
1458 metadata !1, ;; File
1459 metadata !1, ;; Context
1460 metadata !"Trees", ;; Name
1461 i32 1, ;; Line number
1462 i64 32, ;; Size in bits
1463 i64 32, ;; Align in bits
1464 i64 0, ;; Offset in bits
1466 null, ;; Derived From type
1467 metadata !3, ;; Elements
1468 i32 0 ;; Runtime language
1472 ;; Define the array of enumerators used by composite type Trees.
1474 !3 = metadata !{metadata !4, metadata !5, metadata !6}
1477 ;; Define Spruce enumerator.
1479 !4 = metadata !{i32 786472, metadata !"Spruce", i64 100}
1482 ;; Define Oak enumerator.
1484 !5 = metadata !{i32 786472, metadata !"Oak", i64 200}
1487 ;; Define Maple enumerator.
1489 !6 = metadata !{i32 786472, metadata !"Maple", i64 300}
1491 Debugging information format
1492 ============================
1494 Debugging Information Extension for Objective C Properties
1495 ----------------------------------------------------------
1500 Objective C provides a simpler way to declare and define accessor methods using
1501 declared properties. The language provides features to declare a property and
1502 to let compiler synthesize accessor methods.
1504 The debugger lets developer inspect Objective C interfaces and their instance
1505 variables and class variables. However, the debugger does not know anything
1506 about the properties defined in Objective C interfaces. The debugger consumes
1507 information generated by compiler in DWARF format. The format does not support
1508 encoding of Objective C properties. This proposal describes DWARF extensions to
1509 encode Objective C properties, which the debugger can use to let developers
1510 inspect Objective C properties.
1515 Objective C properties exist separately from class members. A property can be
1516 defined only by "setter" and "getter" selectors, and be calculated anew on each
1517 access. Or a property can just be a direct access to some declared ivar.
1518 Finally it can have an ivar "automatically synthesized" for it by the compiler,
1519 in which case the property can be referred to in user code directly using the
1520 standard C dereference syntax as well as through the property "dot" syntax, but
1521 there is no entry in the ``@interface`` declaration corresponding to this ivar.
1523 To facilitate debugging, these properties we will add a new DWARF TAG into the
1524 ``DW_TAG_structure_type`` definition for the class to hold the description of a
1525 given property, and a set of DWARF attributes that provide said description.
1526 The property tag will also contain the name and declared type of the property.
1528 If there is a related ivar, there will also be a DWARF property attribute placed
1529 in the ``DW_TAG_member`` DIE for that ivar referring back to the property TAG
1530 for that property. And in the case where the compiler synthesizes the ivar
1531 directly, the compiler is expected to generate a ``DW_TAG_member`` for that
1532 ivar (with the ``DW_AT_artificial`` set to 1), whose name will be the name used
1533 to access this ivar directly in code, and with the property attribute pointing
1534 back to the property it is backing.
1536 The following examples will serve as illustration for our discussion:
1538 .. code-block:: objc
1550 @synthesize p2 = n2;
1553 This produces the following DWARF (this is a "pseudo dwarfdump" output):
1555 .. code-block:: none
1557 0x00000100: TAG_structure_type [7] *
1558 AT_APPLE_runtime_class( 0x10 )
1560 AT_decl_file( "Objc_Property.m" )
1563 0x00000110 TAG_APPLE_property
1565 AT_type ( {0x00000150} ( int ) )
1567 0x00000120: TAG_APPLE_property
1569 AT_type ( {0x00000150} ( int ) )
1571 0x00000130: TAG_member [8]
1573 AT_APPLE_property ( {0x00000110} "p1" )
1574 AT_type( {0x00000150} ( int ) )
1575 AT_artificial ( 0x1 )
1577 0x00000140: TAG_member [8]
1579 AT_APPLE_property ( {0x00000120} "p2" )
1580 AT_type( {0x00000150} ( int ) )
1582 0x00000150: AT_type( ( int ) )
1584 Note, the current convention is that the name of the ivar for an
1585 auto-synthesized property is the name of the property from which it derives
1586 with an underscore prepended, as is shown in the example. But we actually
1587 don't need to know this convention, since we are given the name of the ivar
1590 Also, it is common practice in ObjC to have different property declarations in
1591 the @interface and @implementation - e.g. to provide a read-only property in
1592 the interface,and a read-write interface in the implementation. In that case,
1593 the compiler should emit whichever property declaration will be in force in the
1594 current translation unit.
1596 Developers can decorate a property with attributes which are encoded using
1597 ``DW_AT_APPLE_property_attribute``.
1599 .. code-block:: objc
1601 @property (readonly, nonatomic) int pr;
1603 .. code-block:: none
1605 TAG_APPLE_property [8]
1607 AT_type ( {0x00000147} (int) )
1608 AT_APPLE_property_attribute (DW_APPLE_PROPERTY_readonly, DW_APPLE_PROPERTY_nonatomic)
1610 The setter and getter method names are attached to the property using
1611 ``DW_AT_APPLE_property_setter`` and ``DW_AT_APPLE_property_getter`` attributes.
1613 .. code-block:: objc
1616 @property (setter=myOwnP3Setter:) int p3;
1617 -(void)myOwnP3Setter:(int)a;
1622 -(void)myOwnP3Setter:(int)a{ }
1625 The DWARF for this would be:
1627 .. code-block:: none
1629 0x000003bd: TAG_structure_type [7] *
1630 AT_APPLE_runtime_class( 0x10 )
1632 AT_decl_file( "Objc_Property.m" )
1635 0x000003cd TAG_APPLE_property
1637 AT_APPLE_property_setter ( "myOwnP3Setter:" )
1638 AT_type( {0x00000147} ( int ) )
1640 0x000003f3: TAG_member [8]
1642 AT_type ( {0x00000147} ( int ) )
1643 AT_APPLE_property ( {0x000003cd} )
1644 AT_artificial ( 0x1 )
1649 +-----------------------+--------+
1651 +=======================+========+
1652 | DW_TAG_APPLE_property | 0x4200 |
1653 +-----------------------+--------+
1655 New DWARF Attributes
1656 ^^^^^^^^^^^^^^^^^^^^
1658 +--------------------------------+--------+-----------+
1659 | Attribute | Value | Classes |
1660 +================================+========+===========+
1661 | DW_AT_APPLE_property | 0x3fed | Reference |
1662 +--------------------------------+--------+-----------+
1663 | DW_AT_APPLE_property_getter | 0x3fe9 | String |
1664 +--------------------------------+--------+-----------+
1665 | DW_AT_APPLE_property_setter | 0x3fea | String |
1666 +--------------------------------+--------+-----------+
1667 | DW_AT_APPLE_property_attribute | 0x3feb | Constant |
1668 +--------------------------------+--------+-----------+
1673 +--------------------------------+-------+
1675 +================================+=======+
1676 | DW_AT_APPLE_PROPERTY_readonly | 0x1 |
1677 +--------------------------------+-------+
1678 | DW_AT_APPLE_PROPERTY_readwrite | 0x2 |
1679 +--------------------------------+-------+
1680 | DW_AT_APPLE_PROPERTY_assign | 0x4 |
1681 +--------------------------------+-------+
1682 | DW_AT_APPLE_PROPERTY_retain | 0x8 |
1683 +--------------------------------+-------+
1684 | DW_AT_APPLE_PROPERTY_copy | 0x10 |
1685 +--------------------------------+-------+
1686 | DW_AT_APPLE_PROPERTY_nonatomic | 0x20 |
1687 +--------------------------------+-------+
1689 Name Accelerator Tables
1690 -----------------------
1695 The "``.debug_pubnames``" and "``.debug_pubtypes``" formats are not what a
1696 debugger needs. The "``pub``" in the section name indicates that the entries
1697 in the table are publicly visible names only. This means no static or hidden
1698 functions show up in the "``.debug_pubnames``". No static variables or private
1699 class variables are in the "``.debug_pubtypes``". Many compilers add different
1700 things to these tables, so we can't rely upon the contents between gcc, icc, or
1703 The typical query given by users tends not to match up with the contents of
1704 these tables. For example, the DWARF spec states that "In the case of the name
1705 of a function member or static data member of a C++ structure, class or union,
1706 the name presented in the "``.debug_pubnames``" section is not the simple name
1707 given by the ``DW_AT_name attribute`` of the referenced debugging information
1708 entry, but rather the fully qualified name of the data or function member."
1709 So the only names in these tables for complex C++ entries is a fully
1710 qualified name. Debugger users tend not to enter their search strings as
1711 "``a::b::c(int,const Foo&) const``", but rather as "``c``", "``b::c``" , or
1712 "``a::b::c``". So the name entered in the name table must be demangled in
1713 order to chop it up appropriately and additional names must be manually entered
1714 into the table to make it effective as a name lookup table for debuggers to
1717 All debuggers currently ignore the "``.debug_pubnames``" table as a result of
1718 its inconsistent and useless public-only name content making it a waste of
1719 space in the object file. These tables, when they are written to disk, are not
1720 sorted in any way, leaving every debugger to do its own parsing and sorting.
1721 These tables also include an inlined copy of the string values in the table
1722 itself making the tables much larger than they need to be on disk, especially
1723 for large C++ programs.
1725 Can't we just fix the sections by adding all of the names we need to this
1726 table? No, because that is not what the tables are defined to contain and we
1727 won't know the difference between the old bad tables and the new good tables.
1728 At best we could make our own renamed sections that contain all of the data we
1731 These tables are also insufficient for what a debugger like LLDB needs. LLDB
1732 uses clang for its expression parsing where LLDB acts as a PCH. LLDB is then
1733 often asked to look for type "``foo``" or namespace "``bar``", or list items in
1734 namespace "``baz``". Namespaces are not included in the pubnames or pubtypes
1735 tables. Since clang asks a lot of questions when it is parsing an expression,
1736 we need to be very fast when looking up names, as it happens a lot. Having new
1737 accelerator tables that are optimized for very quick lookups will benefit this
1738 type of debugging experience greatly.
1740 We would like to generate name lookup tables that can be mapped into memory
1741 from disk, and used as is, with little or no up-front parsing. We would also
1742 be able to control the exact content of these different tables so they contain
1743 exactly what we need. The Name Accelerator Tables were designed to fix these
1744 issues. In order to solve these issues we need to:
1746 * Have a format that can be mapped into memory from disk and used as is
1747 * Lookups should be very fast
1748 * Extensible table format so these tables can be made by many producers
1749 * Contain all of the names needed for typical lookups out of the box
1750 * Strict rules for the contents of tables
1752 Table size is important and the accelerator table format should allow the reuse
1753 of strings from common string tables so the strings for the names are not
1754 duplicated. We also want to make sure the table is ready to be used as-is by
1755 simply mapping the table into memory with minimal header parsing.
1757 The name lookups need to be fast and optimized for the kinds of lookups that
1758 debuggers tend to do. Optimally we would like to touch as few parts of the
1759 mapped table as possible when doing a name lookup and be able to quickly find
1760 the name entry we are looking for, or discover there are no matches. In the
1761 case of debuggers we optimized for lookups that fail most of the time.
1763 Each table that is defined should have strict rules on exactly what is in the
1764 accelerator tables and documented so clients can rely on the content.
1769 Standard Hash Tables
1770 """"""""""""""""""""
1772 Typical hash tables have a header, buckets, and each bucket points to the
1775 .. code-block:: none
1785 The BUCKETS are an array of offsets to DATA for each hash:
1787 .. code-block:: none
1790 | 0x00001000 | BUCKETS[0]
1791 | 0x00002000 | BUCKETS[1]
1792 | 0x00002200 | BUCKETS[2]
1793 | 0x000034f0 | BUCKETS[3]
1795 | 0xXXXXXXXX | BUCKETS[n_buckets]
1798 So for ``bucket[3]`` in the example above, we have an offset into the table
1799 0x000034f0 which points to a chain of entries for the bucket. Each bucket must
1800 contain a next pointer, full 32 bit hash value, the string itself, and the data
1801 for the current string value.
1803 .. code-block:: none
1806 0x000034f0: | 0x00003500 | next pointer
1807 | 0x12345678 | 32 bit hash
1808 | "erase" | string value
1809 | data[n] | HashData for this bucket
1811 0x00003500: | 0x00003550 | next pointer
1812 | 0x29273623 | 32 bit hash
1813 | "dump" | string value
1814 | data[n] | HashData for this bucket
1816 0x00003550: | 0x00000000 | next pointer
1817 | 0x82638293 | 32 bit hash
1818 | "main" | string value
1819 | data[n] | HashData for this bucket
1822 The problem with this layout for debuggers is that we need to optimize for the
1823 negative lookup case where the symbol we're searching for is not present. So
1824 if we were to lookup "``printf``" in the table above, we would make a 32 hash
1825 for "``printf``", it might match ``bucket[3]``. We would need to go to the
1826 offset 0x000034f0 and start looking to see if our 32 bit hash matches. To do
1827 so, we need to read the next pointer, then read the hash, compare it, and skip
1828 to the next bucket. Each time we are skipping many bytes in memory and
1829 touching new cache pages just to do the compare on the full 32 bit hash. All
1830 of these accesses then tell us that we didn't have a match.
1835 To solve the issues mentioned above we have structured the hash tables a bit
1836 differently: a header, buckets, an array of all unique 32 bit hash values,
1837 followed by an array of hash value data offsets, one for each hash value, then
1838 the data for all hash values:
1840 .. code-block:: none
1854 The ``BUCKETS`` in the name tables are an index into the ``HASHES`` array. By
1855 making all of the full 32 bit hash values contiguous in memory, we allow
1856 ourselves to efficiently check for a match while touching as little memory as
1857 possible. Most often checking the 32 bit hash values is as far as the lookup
1858 goes. If it does match, it usually is a match with no collisions. So for a
1859 table with "``n_buckets``" buckets, and "``n_hashes``" unique 32 bit hash
1860 values, we can clarify the contents of the ``BUCKETS``, ``HASHES`` and
1863 .. code-block:: none
1865 .-------------------------.
1866 | HEADER.magic | uint32_t
1867 | HEADER.version | uint16_t
1868 | HEADER.hash_function | uint16_t
1869 | HEADER.bucket_count | uint32_t
1870 | HEADER.hashes_count | uint32_t
1871 | HEADER.header_data_len | uint32_t
1872 | HEADER_DATA | HeaderData
1873 |-------------------------|
1874 | BUCKETS | uint32_t[n_buckets] // 32 bit hash indexes
1875 |-------------------------|
1876 | HASHES | uint32_t[n_hashes] // 32 bit hash values
1877 |-------------------------|
1878 | OFFSETS | uint32_t[n_hashes] // 32 bit offsets to hash value data
1879 |-------------------------|
1881 `-------------------------'
1883 So taking the exact same data from the standard hash example above we end up
1886 .. code-block:: none
1896 | ... | BUCKETS[n_buckets]
1898 | 0x........ | HASHES[0]
1899 | 0x........ | HASHES[1]
1900 | 0x........ | HASHES[2]
1901 | 0x........ | HASHES[3]
1902 | 0x........ | HASHES[4]
1903 | 0x........ | HASHES[5]
1904 | 0x12345678 | HASHES[6] hash for BUCKETS[3]
1905 | 0x29273623 | HASHES[7] hash for BUCKETS[3]
1906 | 0x82638293 | HASHES[8] hash for BUCKETS[3]
1907 | 0x........ | HASHES[9]
1908 | 0x........ | HASHES[10]
1909 | 0x........ | HASHES[11]
1910 | 0x........ | HASHES[12]
1911 | 0x........ | HASHES[13]
1912 | 0x........ | HASHES[n_hashes]
1914 | 0x........ | OFFSETS[0]
1915 | 0x........ | OFFSETS[1]
1916 | 0x........ | OFFSETS[2]
1917 | 0x........ | OFFSETS[3]
1918 | 0x........ | OFFSETS[4]
1919 | 0x........ | OFFSETS[5]
1920 | 0x000034f0 | OFFSETS[6] offset for BUCKETS[3]
1921 | 0x00003500 | OFFSETS[7] offset for BUCKETS[3]
1922 | 0x00003550 | OFFSETS[8] offset for BUCKETS[3]
1923 | 0x........ | OFFSETS[9]
1924 | 0x........ | OFFSETS[10]
1925 | 0x........ | OFFSETS[11]
1926 | 0x........ | OFFSETS[12]
1927 | 0x........ | OFFSETS[13]
1928 | 0x........ | OFFSETS[n_hashes]
1936 0x000034f0: | 0x00001203 | .debug_str ("erase")
1937 | 0x00000004 | A 32 bit array count - number of HashData with name "erase"
1938 | 0x........ | HashData[0]
1939 | 0x........ | HashData[1]
1940 | 0x........ | HashData[2]
1941 | 0x........ | HashData[3]
1942 | 0x00000000 | String offset into .debug_str (terminate data for hash)
1944 0x00003500: | 0x00001203 | String offset into .debug_str ("collision")
1945 | 0x00000002 | A 32 bit array count - number of HashData with name "collision"
1946 | 0x........ | HashData[0]
1947 | 0x........ | HashData[1]
1948 | 0x00001203 | String offset into .debug_str ("dump")
1949 | 0x00000003 | A 32 bit array count - number of HashData with name "dump"
1950 | 0x........ | HashData[0]
1951 | 0x........ | HashData[1]
1952 | 0x........ | HashData[2]
1953 | 0x00000000 | String offset into .debug_str (terminate data for hash)
1955 0x00003550: | 0x00001203 | String offset into .debug_str ("main")
1956 | 0x00000009 | A 32 bit array count - number of HashData with name "main"
1957 | 0x........ | HashData[0]
1958 | 0x........ | HashData[1]
1959 | 0x........ | HashData[2]
1960 | 0x........ | HashData[3]
1961 | 0x........ | HashData[4]
1962 | 0x........ | HashData[5]
1963 | 0x........ | HashData[6]
1964 | 0x........ | HashData[7]
1965 | 0x........ | HashData[8]
1966 | 0x00000000 | String offset into .debug_str (terminate data for hash)
1969 So we still have all of the same data, we just organize it more efficiently for
1970 debugger lookup. If we repeat the same "``printf``" lookup from above, we
1971 would hash "``printf``" and find it matches ``BUCKETS[3]`` by taking the 32 bit
1972 hash value and modulo it by ``n_buckets``. ``BUCKETS[3]`` contains "6" which
1973 is the index into the ``HASHES`` table. We would then compare any consecutive
1974 32 bit hashes values in the ``HASHES`` array as long as the hashes would be in
1975 ``BUCKETS[3]``. We do this by verifying that each subsequent hash value modulo
1976 ``n_buckets`` is still 3. In the case of a failed lookup we would access the
1977 memory for ``BUCKETS[3]``, and then compare a few consecutive 32 bit hashes
1978 before we know that we have no match. We don't end up marching through
1979 multiple words of memory and we really keep the number of processor data cache
1980 lines being accessed as small as possible.
1982 The string hash that is used for these lookup tables is the Daniel J.
1983 Bernstein hash which is also used in the ELF ``GNU_HASH`` sections. It is a
1984 very good hash for all kinds of names in programs with very few hash
1987 Empty buckets are designated by using an invalid hash index of ``UINT32_MAX``.
1992 These name hash tables are designed to be generic where specializations of the
1993 table get to define additional data that goes into the header ("``HeaderData``"),
1994 how the string value is stored ("``KeyType``") and the content of the data for each
2000 The header has a fixed part, and the specialized part. The exact format of the
2007 uint32_t magic; // 'HASH' magic value to allow endian detection
2008 uint16_t version; // Version number
2009 uint16_t hash_function; // The hash function enumeration that was used
2010 uint32_t bucket_count; // The number of buckets in this hash table
2011 uint32_t hashes_count; // The total number of unique hash values and hash data offsets in this table
2012 uint32_t header_data_len; // The bytes to skip to get to the hash indexes (buckets) for correct alignment
2013 // Specifically the length of the following HeaderData field - this does not
2014 // include the size of the preceding fields
2015 HeaderData header_data; // Implementation specific header data
2018 The header starts with a 32 bit "``magic``" value which must be ``'HASH'``
2019 encoded as an ASCII integer. This allows the detection of the start of the
2020 hash table and also allows the table's byte order to be determined so the table
2021 can be correctly extracted. The "``magic``" value is followed by a 16 bit
2022 ``version`` number which allows the table to be revised and modified in the
2023 future. The current version number is 1. ``hash_function`` is a ``uint16_t``
2024 enumeration that specifies which hash function was used to produce this table.
2025 The current values for the hash function enumerations include:
2029 enum HashFunctionType
2031 eHashFunctionDJB = 0u, // Daniel J Bernstein hash function
2034 ``bucket_count`` is a 32 bit unsigned integer that represents how many buckets
2035 are in the ``BUCKETS`` array. ``hashes_count`` is the number of unique 32 bit
2036 hash values that are in the ``HASHES`` array, and is the same number of offsets
2037 are contained in the ``OFFSETS`` array. ``header_data_len`` specifies the size
2038 in bytes of the ``HeaderData`` that is filled in by specialized versions of
2044 The header is followed by the buckets, hashes, offsets, and hash value data.
2050 uint32_t buckets[Header.bucket_count]; // An array of hash indexes into the "hashes[]" array below
2051 uint32_t hashes [Header.hashes_count]; // Every unique 32 bit hash for the entire table is in this table
2052 uint32_t offsets[Header.hashes_count]; // An offset that corresponds to each item in the "hashes[]" array above
2055 ``buckets`` is an array of 32 bit indexes into the ``hashes`` array. The
2056 ``hashes`` array contains all of the 32 bit hash values for all names in the
2057 hash table. Each hash in the ``hashes`` table has an offset in the ``offsets``
2058 array that points to the data for the hash value.
2060 This table setup makes it very easy to repurpose these tables to contain
2061 different data, while keeping the lookup mechanism the same for all tables.
2062 This layout also makes it possible to save the table to disk and map it in
2063 later and do very efficient name lookups with little or no parsing.
2065 DWARF lookup tables can be implemented in a variety of ways and can store a lot
2066 of information for each name. We want to make the DWARF tables extensible and
2067 able to store the data efficiently so we have used some of the DWARF features
2068 that enable efficient data storage to define exactly what kind of data we store
2071 The ``HeaderData`` contains a definition of the contents of each HashData chunk.
2072 We might want to store an offset to all of the debug information entries (DIEs)
2073 for each name. To keep things extensible, we create a list of items, or
2074 Atoms, that are contained in the data for each name. First comes the type of
2075 the data in each atom:
2082 eAtomTypeDIEOffset = 1u, // DIE offset, check form for encoding
2083 eAtomTypeCUOffset = 2u, // DIE offset of the compiler unit header that contains the item in question
2084 eAtomTypeTag = 3u, // DW_TAG_xxx value, should be encoded as DW_FORM_data1 (if no tags exceed 255) or DW_FORM_data2
2085 eAtomTypeNameFlags = 4u, // Flags from enum NameFlags
2086 eAtomTypeTypeFlags = 5u, // Flags from enum TypeFlags
2089 The enumeration values and their meanings are:
2091 .. code-block:: none
2093 eAtomTypeNULL - a termination atom that specifies the end of the atom list
2094 eAtomTypeDIEOffset - an offset into the .debug_info section for the DWARF DIE for this name
2095 eAtomTypeCUOffset - an offset into the .debug_info section for the CU that contains the DIE
2096 eAtomTypeDIETag - The DW_TAG_XXX enumeration value so you don't have to parse the DWARF to see what it is
2097 eAtomTypeNameFlags - Flags for functions and global variables (isFunction, isInlined, isExternal...)
2098 eAtomTypeTypeFlags - Flags for types (isCXXClass, isObjCClass, ...)
2100 Then we allow each atom type to define the atom type and how the data for each
2101 atom type data is encoded:
2107 uint16_t type; // AtomType enum value
2108 uint16_t form; // DWARF DW_FORM_XXX defines
2111 The ``form`` type above is from the DWARF specification and defines the exact
2112 encoding of the data for the Atom type. See the DWARF specification for the
2113 ``DW_FORM_`` definitions.
2119 uint32_t die_offset_base;
2120 uint32_t atom_count;
2121 Atoms atoms[atom_count0];
2124 ``HeaderData`` defines the base DIE offset that should be added to any atoms
2125 that are encoded using the ``DW_FORM_ref1``, ``DW_FORM_ref2``,
2126 ``DW_FORM_ref4``, ``DW_FORM_ref8`` or ``DW_FORM_ref_udata``. It also defines
2127 what is contained in each ``HashData`` object -- ``Atom.form`` tells us how large
2128 each field will be in the ``HashData`` and the ``Atom.type`` tells us how this data
2129 should be interpreted.
2131 For the current implementations of the "``.apple_names``" (all functions +
2132 globals), the "``.apple_types``" (names of all types that are defined), and
2133 the "``.apple_namespaces``" (all namespaces), we currently set the ``Atom``
2138 HeaderData.atom_count = 1;
2139 HeaderData.atoms[0].type = eAtomTypeDIEOffset;
2140 HeaderData.atoms[0].form = DW_FORM_data4;
2142 This defines the contents to be the DIE offset (eAtomTypeDIEOffset) that is
2143 encoded as a 32 bit value (DW_FORM_data4). This allows a single name to have
2144 multiple matching DIEs in a single file, which could come up with an inlined
2145 function for instance. Future tables could include more information about the
2146 DIE such as flags indicating if the DIE is a function, method, block,
2149 The KeyType for the DWARF table is a 32 bit string table offset into the
2150 ".debug_str" table. The ".debug_str" is the string table for the DWARF which
2151 may already contain copies of all of the strings. This helps make sure, with
2152 help from the compiler, that we reuse the strings between all of the DWARF
2153 sections and keeps the hash table size down. Another benefit to having the
2154 compiler generate all strings as DW_FORM_strp in the debug info, is that
2155 DWARF parsing can be made much faster.
2157 After a lookup is made, we get an offset into the hash data. The hash data
2158 needs to be able to deal with 32 bit hash collisions, so the chunk of data
2159 at the offset in the hash data consists of a triple:
2164 uint32_t hash_data_count
2165 HashData[hash_data_count]
2167 If "str_offset" is zero, then the bucket contents are done. 99.9% of the
2168 hash data chunks contain a single item (no 32 bit hash collision):
2170 .. code-block:: none
2173 | 0x00001023 | uint32_t KeyType (.debug_str[0x0001023] => "main")
2174 | 0x00000004 | uint32_t HashData count
2175 | 0x........ | uint32_t HashData[0] DIE offset
2176 | 0x........ | uint32_t HashData[1] DIE offset
2177 | 0x........ | uint32_t HashData[2] DIE offset
2178 | 0x........ | uint32_t HashData[3] DIE offset
2179 | 0x00000000 | uint32_t KeyType (end of hash chain)
2182 If there are collisions, you will have multiple valid string offsets:
2184 .. code-block:: none
2187 | 0x00001023 | uint32_t KeyType (.debug_str[0x0001023] => "main")
2188 | 0x00000004 | uint32_t HashData count
2189 | 0x........ | uint32_t HashData[0] DIE offset
2190 | 0x........ | uint32_t HashData[1] DIE offset
2191 | 0x........ | uint32_t HashData[2] DIE offset
2192 | 0x........ | uint32_t HashData[3] DIE offset
2193 | 0x00002023 | uint32_t KeyType (.debug_str[0x0002023] => "print")
2194 | 0x00000002 | uint32_t HashData count
2195 | 0x........ | uint32_t HashData[0] DIE offset
2196 | 0x........ | uint32_t HashData[1] DIE offset
2197 | 0x00000000 | uint32_t KeyType (end of hash chain)
2200 Current testing with real world C++ binaries has shown that there is around 1
2201 32 bit hash collision per 100,000 name entries.
2206 As we said, we want to strictly define exactly what is included in the
2207 different tables. For DWARF, we have 3 tables: "``.apple_names``",
2208 "``.apple_types``", and "``.apple_namespaces``".
2210 "``.apple_names``" sections should contain an entry for each DWARF DIE whose
2211 ``DW_TAG`` is a ``DW_TAG_label``, ``DW_TAG_inlined_subroutine``, or
2212 ``DW_TAG_subprogram`` that has address attributes: ``DW_AT_low_pc``,
2213 ``DW_AT_high_pc``, ``DW_AT_ranges`` or ``DW_AT_entry_pc``. It also contains
2214 ``DW_TAG_variable`` DIEs that have a ``DW_OP_addr`` in the location (global and
2215 static variables). All global and static variables should be included,
2216 including those scoped within functions and classes. For example using the
2228 Both of the static ``var`` variables would be included in the table. All
2229 functions should emit both their full names and their basenames. For C or C++,
2230 the full name is the mangled name (if available) which is usually in the
2231 ``DW_AT_MIPS_linkage_name`` attribute, and the ``DW_AT_name`` contains the
2232 function basename. If global or static variables have a mangled name in a
2233 ``DW_AT_MIPS_linkage_name`` attribute, this should be emitted along with the
2234 simple name found in the ``DW_AT_name`` attribute.
2236 "``.apple_types``" sections should contain an entry for each DWARF DIE whose
2241 * DW_TAG_enumeration_type
2242 * DW_TAG_pointer_type
2243 * DW_TAG_reference_type
2244 * DW_TAG_string_type
2245 * DW_TAG_structure_type
2246 * DW_TAG_subroutine_type
2249 * DW_TAG_ptr_to_member_type
2251 * DW_TAG_subrange_type
2257 * DW_TAG_packed_type
2258 * DW_TAG_volatile_type
2259 * DW_TAG_restrict_type
2260 * DW_TAG_interface_type
2261 * DW_TAG_unspecified_type
2262 * DW_TAG_shared_type
2264 Only entries with a ``DW_AT_name`` attribute are included, and the entry must
2265 not be a forward declaration (``DW_AT_declaration`` attribute with a non-zero
2266 value). For example, using the following code:
2276 We get a few type DIEs:
2278 .. code-block:: none
2280 0x00000067: TAG_base_type [5]
2281 AT_encoding( DW_ATE_signed )
2283 AT_byte_size( 0x04 )
2285 0x0000006e: TAG_pointer_type [6]
2286 AT_type( {0x00000067} ( int ) )
2287 AT_byte_size( 0x08 )
2289 The DW_TAG_pointer_type is not included because it does not have a ``DW_AT_name``.
2291 "``.apple_namespaces``" section should contain all ``DW_TAG_namespace`` DIEs.
2292 If we run into a namespace that has no name this is an anonymous namespace, and
2293 the name should be output as "``(anonymous namespace)``" (without the quotes).
2294 Why? This matches the output of the ``abi::cxa_demangle()`` that is in the
2295 standard C++ library that demangles mangled names.
2298 Language Extensions and File Format Changes
2299 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2301 Objective-C Extensions
2302 """"""""""""""""""""""
2304 "``.apple_objc``" section should contain all ``DW_TAG_subprogram`` DIEs for an
2305 Objective-C class. The name used in the hash table is the name of the
2306 Objective-C class itself. If the Objective-C class has a category, then an
2307 entry is made for both the class name without the category, and for the class
2308 name with the category. So if we have a DIE at offset 0x1234 with a name of
2309 method "``-[NSString(my_additions) stringWithSpecialString:]``", we would add
2310 an entry for "``NSString``" that points to DIE 0x1234, and an entry for
2311 "``NSString(my_additions)``" that points to 0x1234. This allows us to quickly
2312 track down all Objective-C methods for an Objective-C class when doing
2313 expressions. It is needed because of the dynamic nature of Objective-C where
2314 anyone can add methods to a class. The DWARF for Objective-C methods is also
2315 emitted differently from C++ classes where the methods are not usually
2316 contained in the class definition, they are scattered about across one or more
2317 compile units. Categories can also be defined in different shared libraries.
2318 So we need to be able to quickly find all of the methods and class functions
2319 given the Objective-C class name, or quickly find all methods and class
2320 functions for a class + category name. This table does not contain any
2321 selector names, it just maps Objective-C class names (or class names +
2322 category) to all of the methods and class functions. The selectors are added
2323 as function basenames in the "``.debug_names``" section.
2325 In the "``.apple_names``" section for Objective-C functions, the full name is
2326 the entire function name with the brackets ("``-[NSString
2327 stringWithCString:]``") and the basename is the selector only
2328 ("``stringWithCString:``").
2333 The sections names for the apple hash tables are for non-mach-o files. For
2334 mach-o files, the sections should be contained in the ``__DWARF`` segment with
2337 * "``.apple_names``" -> "``__apple_names``"
2338 * "``.apple_types``" -> "``__apple_types``"
2339 * "``.apple_namespaces``" -> "``__apple_namespac``" (16 character limit)
2340 * "``.apple_objc``" -> "``__apple_objc``"