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
218 These descriptors contain a source language ID for the file (we use the DWARF
219 3.0 ID numbers, such as ``DW_LANG_C89``, ``DW_LANG_C_plus_plus``,
220 ``DW_LANG_Cobol74``, etc), a reference to a metadata node containing a pair of
221 strings for the source file name and the working directory, as well as an
222 identifier string for the compiler that produced it.
224 Compile unit descriptors provide the root context for objects declared in a
225 specific compilation unit. File descriptors are defined using this context.
226 These descriptors are collected by a named metadata ``!llvm.dbg.cu``. They
227 keep track of subprograms, global variables, type information, and imported
228 entities (declarations and namespaces).
238 i32, ;; Tag = 41 (DW_TAG_file_type)
239 metadata, ;; Source directory (including trailing slash) & file pair
242 These descriptors contain information for a file. Global variables and top
243 level functions would be defined using this context. File descriptors also
244 provide context for source line correspondence.
246 Each input file is encoded as a separate file descriptor in LLVM debugging
249 .. _format_global_variables:
251 Global variable descriptors
252 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
257 i32, ;; Tag = 52 (DW_TAG_variable)
258 i32, ;; Unused field.
259 metadata, ;; Reference to context descriptor
261 metadata, ;; Display name (fully qualified C++ name)
262 metadata, ;; MIPS linkage name (for C++)
263 metadata, ;; Reference to file where defined
264 i32, ;; Line number where defined
265 metadata, ;; Reference to type descriptor
266 i1, ;; True if the global is local to compile unit (static)
267 i1, ;; True if the global is defined in the compile unit (not extern)
268 {}*, ;; Reference to the global variable
269 metadata, ;; The static member declaration, if any
272 These descriptors provide debug information about global variables. They
273 provide details such as name, type and where the variable is defined. All
274 global variables are collected inside the named metadata ``!llvm.dbg.cu``.
276 .. _format_subprograms:
278 Subprogram descriptors
279 ^^^^^^^^^^^^^^^^^^^^^^
284 i32, ;; Tag = 46 (DW_TAG_subprogram)
285 metadata, ;; Source directory (including trailing slash) & file pair
286 metadata, ;; Reference to context descriptor
288 metadata, ;; Display name (fully qualified C++ name)
289 metadata, ;; MIPS linkage name (for C++)
290 i32, ;; Line number where defined
291 metadata, ;; Reference to type descriptor
292 i1, ;; True if the global is local to compile unit (static)
293 i1, ;; True if the global is defined in the compile unit (not extern)
294 i32, ;; Virtuality, e.g. dwarf::DW_VIRTUALITY__virtual
295 i32, ;; Index into a virtual function
296 metadata, ;; indicates which base type contains the vtable pointer for the
298 i32, ;; Flags - Artificial, Private, Protected, Explicit, Prototyped.
300 {}*, ;; Reference to the LLVM function
301 metadata, ;; Lists function template parameters
302 metadata, ;; Function declaration descriptor
303 metadata, ;; List of function variables
304 i32 ;; Line number where the scope of the subprogram begins
307 These descriptors provide debug information about functions, methods and
308 subprograms. They provide details such as name, return types and the source
309 location where the subprogram is defined.
317 i32, ;; Tag = 11 (DW_TAG_lexical_block)
318 metadata, ;; Source directory (including trailing slash) & file pair
319 metadata, ;; Reference to context descriptor
321 i32, ;; Column number
322 i32, ;; DWARF path discriminator value
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
338 This descriptor provides a wrapper around a lexical scope to handle file
339 changes in the middle of a lexical block.
341 .. _format_basic_type:
343 Basic type descriptors
344 ^^^^^^^^^^^^^^^^^^^^^^
349 i32, ;; Tag = 36 (DW_TAG_base_type)
350 metadata, ;; Source directory (including trailing slash) & file pair (may be null)
351 metadata, ;; Reference to context
352 metadata, ;; Name (may be "" for anonymous types)
353 i32, ;; Line number where defined (may be 0)
355 i64, ;; Alignment in bits
356 i64, ;; Offset in bits
358 i32 ;; DWARF type encoding
361 These descriptors define primitive types used in the code. Example ``int``,
362 ``bool`` and ``float``. The context provides the scope of the type, which is
363 usually the top level. Since basic types are not usually user defined the
364 context and line number can be left as NULL and 0. The size, alignment and
365 offset are expressed in bits and can be 64 bit values. The alignment is used
366 to round the offset when embedded in a :ref:`composite type
367 <format_composite_type>` (example to keep float doubles on 64 bit boundaries).
368 The offset is the bit offset if embedded in a :ref:`composite type
369 <format_composite_type>`.
371 The type encoding provides the details of the type. The values are typically
372 one of the following:
380 DW_ATE_signed_char = 6
382 DW_ATE_unsigned_char = 8
384 .. _format_derived_type:
386 Derived type descriptors
387 ^^^^^^^^^^^^^^^^^^^^^^^^
392 i32, ;; Tag (see below)
393 metadata, ;; Source directory (including trailing slash) & file pair (may be null)
394 metadata, ;; Reference to context
395 metadata, ;; Name (may be "" for anonymous types)
396 i32, ;; Line number where defined (may be 0)
398 i64, ;; Alignment in bits
399 i64, ;; Offset in bits
400 i32, ;; Flags to encode attributes, e.g. private
401 metadata, ;; Reference to type derived from
402 metadata, ;; (optional) Name of the Objective C property associated with
403 ;; Objective-C an ivar, or the type of which this
404 ;; pointer-to-member is pointing to members of.
405 metadata, ;; (optional) Name of the Objective C property getter selector.
406 metadata, ;; (optional) Name of the Objective C property setter selector.
407 i32 ;; (optional) Objective C property attributes.
410 These descriptors are used to define types derived from other types. The value
411 of the tag varies depending on the meaning. The following are possible tag
416 DW_TAG_formal_parameter = 5
418 DW_TAG_pointer_type = 15
419 DW_TAG_reference_type = 16
421 DW_TAG_ptr_to_member_type = 31
422 DW_TAG_const_type = 38
423 DW_TAG_volatile_type = 53
424 DW_TAG_restrict_type = 55
426 ``DW_TAG_member`` is used to define a member of a :ref:`composite type
427 <format_composite_type>` or :ref:`subprogram <format_subprograms>`. The type
428 of the member is the :ref:`derived type <format_derived_type>`.
429 ``DW_TAG_formal_parameter`` is used to define a member which is a formal
430 argument of a subprogram.
432 ``DW_TAG_typedef`` is used to provide a name for the derived type.
434 ``DW_TAG_pointer_type``, ``DW_TAG_reference_type``, ``DW_TAG_const_type``,
435 ``DW_TAG_volatile_type`` and ``DW_TAG_restrict_type`` are used to qualify the
436 :ref:`derived type <format_derived_type>`.
438 :ref:`Derived type <format_derived_type>` location can be determined from the
439 context and line number. The size, alignment and offset are expressed in bits
440 and can be 64 bit values. The alignment is used to round the offset when
441 embedded in a :ref:`composite type <format_composite_type>` (example to keep
442 float doubles on 64 bit boundaries.) The offset is the bit offset if embedded
443 in a :ref:`composite type <format_composite_type>`.
445 Note that the ``void *`` type is expressed as a type derived from NULL.
447 .. _format_composite_type:
449 Composite type descriptors
450 ^^^^^^^^^^^^^^^^^^^^^^^^^^
455 i32, ;; Tag (see below)
456 metadata, ;; Source directory (including trailing slash) & file pair (may be null)
457 metadata, ;; Reference to context
458 metadata, ;; Name (may be "" for anonymous types)
459 i32, ;; Line number where defined (may be 0)
461 i64, ;; Alignment in bits
462 i64, ;; Offset in bits
464 metadata, ;; Reference to type derived from
465 metadata, ;; Reference to array of member descriptors
466 i32, ;; Runtime languages
467 metadata, ;; Base type containing the vtable pointer for this type
468 metadata, ;; Template parameters
469 metadata ;; A unique identifier for type uniquing purpose (may be null)
472 These descriptors are used to define types that are composed of 0 or more
473 elements. The value of the tag varies depending on the meaning. The following
474 are possible tag values:
478 DW_TAG_array_type = 1
479 DW_TAG_enumeration_type = 4
480 DW_TAG_structure_type = 19
481 DW_TAG_union_type = 23
482 DW_TAG_subroutine_type = 21
483 DW_TAG_inheritance = 28
485 The vector flag indicates that an array type is a native packed vector.
487 The members of array types (tag = ``DW_TAG_array_type``) are
488 :ref:`subrange descriptors <format_subrange>`, each
489 representing the range of subscripts at that level of indexing.
491 The members of enumeration types (tag = ``DW_TAG_enumeration_type``) are
492 :ref:`enumerator descriptors <format_enumerator>`, each representing the
493 definition of enumeration value for the set. All enumeration type descriptors
494 are collected inside the named metadata ``!llvm.dbg.cu``.
496 The members of structure (tag = ``DW_TAG_structure_type``) or union (tag =
497 ``DW_TAG_union_type``) types are any one of the :ref:`basic
498 <format_basic_type>`, :ref:`derived <format_derived_type>` or :ref:`composite
499 <format_composite_type>` type descriptors, each representing a field member of
500 the structure or union.
502 For C++ classes (tag = ``DW_TAG_structure_type``), member descriptors provide
503 information about base classes, static members and member functions. If a
504 member is a :ref:`derived type descriptor <format_derived_type>` and has a tag
505 of ``DW_TAG_inheritance``, then the type represents a base class. If the member
506 of is a :ref:`global variable descriptor <format_global_variables>` then it
507 represents a static member. And, if the member is a :ref:`subprogram
508 descriptor <format_subprograms>` then it represents a member function. For
509 static members and member functions, ``getName()`` returns the members link or
510 the C++ mangled name. ``getDisplayName()`` the simplied version of the name.
512 The first member of subroutine (tag = ``DW_TAG_subroutine_type``) type elements
513 is the return type for the subroutine. The remaining elements are the formal
514 arguments to the subroutine.
516 :ref:`Composite type <format_composite_type>` location can be determined from
517 the context and line number. The size, alignment and offset are expressed in
518 bits and can be 64 bit values. The alignment is used to round the offset when
519 embedded in a :ref:`composite type <format_composite_type>` (as an example, to
520 keep float doubles on 64 bit boundaries). The offset is the bit offset if
521 embedded in a :ref:`composite type <format_composite_type>`.
531 i32, ;; Tag = 33 (DW_TAG_subrange_type)
536 These descriptors are used to define ranges of array subscripts for an array
537 :ref:`composite type <format_composite_type>`. The low value defines the lower
538 bounds typically zero for C/C++. The high value is the upper bounds. Values
539 are 64 bit. ``High - Low + 1`` is the size of the array. If ``Low > High``
540 the array bounds are not included in generated debugging information.
542 .. _format_enumerator:
544 Enumerator descriptors
545 ^^^^^^^^^^^^^^^^^^^^^^
550 i32, ;; Tag = 40 (DW_TAG_enumerator)
555 These descriptors are used to define members of an enumeration :ref:`composite
556 type <format_composite_type>`, it associates the name to the value.
564 i32, ;; Tag (see below)
567 metadata, ;; Reference to file where defined
568 i32, ;; 24 bit - Line number where defined
569 ;; 8 bit - Argument number. 1 indicates 1st argument.
570 metadata, ;; Reference to the type descriptor
572 metadata ;; (optional) Reference to inline location
573 metadata ;; (optional) Reference to a complex expression (see below)
576 These descriptors are used to define variables local to a sub program. The
577 value of the tag depends on the usage of the variable:
581 DW_TAG_auto_variable = 256
582 DW_TAG_arg_variable = 257
584 An auto variable is any variable declared in the body of the function. An
585 argument variable is any variable that appears as a formal argument to the
588 The context is either the subprogram or block where the variable is defined.
589 Name the source variable name. Context and line indicate where the variable
590 was defined. Type descriptor defines the declared type of the variable.
592 The ``OpPiece`` operator is used for (typically larger aggregate)
593 variables that are fragmented across several locations. It takes two
594 i32 arguments, an offset and a size in bytes to describe which piece
595 of the variable is at this location.
598 .. _format_common_intrinsics:
600 Debugger intrinsic functions
601 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
603 LLVM uses several intrinsic functions (name prefixed with "``llvm.dbg``") to
604 provide debug information at various points in generated code.
611 void %llvm.dbg.declare(metadata, metadata)
613 This intrinsic provides information about a local element (e.g., variable).
614 The first argument is metadata holding the alloca for the variable. The second
615 argument is metadata containing a description of the variable.
622 void %llvm.dbg.value(metadata, i64, metadata)
624 This intrinsic provides information when a user source variable is set to a new
625 value. The first argument is the new value (wrapped as metadata). The second
626 argument is the offset in the user source variable where the new value is
627 written. The third argument is metadata containing a description of the user
630 Object lifetimes and scoping
631 ============================
633 In many languages, the local variables in functions can have their lifetimes or
634 scopes limited to a subset of a function. In the C family of languages, for
635 example, variables are only live (readable and writable) within the source
636 block that they are defined in. In functional languages, values are only
637 readable after they have been defined. Though this is a very obvious concept,
638 it is non-trivial to model in LLVM, because it has no notion of scoping in this
639 sense, and does not want to be tied to a language's scoping rules.
641 In order to handle this, the LLVM debug format uses the metadata attached to
642 llvm instructions to encode line number and scoping information. Consider the
643 following C fragment, for example:
657 Compiled to LLVM, this function would be represented like this:
661 define void @foo() #0 {
663 %X = alloca i32, align 4
664 %Y = alloca i32, align 4
665 %Z = alloca i32, align 4
666 call void @llvm.dbg.declare(metadata !{i32* %X}, metadata !10), !dbg !12
667 ; [debug line = 2:7] [debug variable = X]
668 store i32 21, i32* %X, align 4, !dbg !12
669 call void @llvm.dbg.declare(metadata !{i32* %Y}, metadata !13), !dbg !14
670 ; [debug line = 3:7] [debug variable = Y]
671 store i32 22, i32* %Y, align 4, !dbg !14
672 call void @llvm.dbg.declare(metadata !{i32* %Z}, metadata !15), !dbg !17
673 ; [debug line = 5:9] [debug variable = Z]
674 store i32 23, i32* %Z, align 4, !dbg !17
675 %0 = load i32* %X, align 4, !dbg !18
677 store i32 %0, i32* %Z, align 4, !dbg !18
678 %1 = load i32* %Y, align 4, !dbg !19
680 store i32 %1, i32* %X, align 4, !dbg !19
684 ; Function Attrs: nounwind readnone
685 declare void @llvm.dbg.declare(metadata, metadata) #1
687 attributes #0 = { nounwind ssp uwtable "less-precise-fpmad"="false"
688 "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf"
689 "no-infs-fp-math"="false" "no-nans-fp-math"="false"
690 "stack-protector-buffer-size"="8" "unsafe-fp-math"="false"
691 "use-soft-float"="false" }
692 attributes #1 = { nounwind readnone }
695 !llvm.module.flags = !{!8}
698 !0 = metadata !{i32 786449, metadata !1, i32 12,
699 metadata !"clang version 3.4 (trunk 193128) (llvm/trunk 193139)",
700 i1 false, metadata !"", i32 0, metadata !2, metadata !2, metadata !3,
701 metadata !2, metadata !2, metadata !""} ; [ DW_TAG_compile_unit ] \
702 [/private/tmp/foo.c] \
704 !1 = metadata !{metadata !"t.c", metadata !"/private/tmp"}
705 !2 = metadata !{i32 0}
706 !3 = metadata !{metadata !4}
707 !4 = metadata !{i32 786478, metadata !1, metadata !5, metadata !"foo",
708 metadata !"foo", metadata !"", i32 1, metadata !6,
709 i1 false, i1 true, i32 0, i32 0, null, i32 0, i1 false,
710 void ()* @foo, null, null, metadata !2, i32 1}
711 ; [ DW_TAG_subprogram ] [line 1] [def] [foo]
712 !5 = metadata !{i32 786473, metadata !1} ; [ DW_TAG_file_type ] \
714 !6 = metadata !{i32 786453, i32 0, null, metadata !"", i32 0, i64 0, i64 0,
715 i64 0, i32 0, null, metadata !7, i32 0, null, null, null}
716 ; [ DW_TAG_subroutine_type ] \
717 [line 0, size 0, align 0, offset 0] [from ]
718 !7 = metadata !{null}
719 !8 = metadata !{i32 2, metadata !"Dwarf Version", i32 2}
720 !9 = metadata !{metadata !"clang version 3.4 (trunk 193128) (llvm/trunk 193139)"}
721 !10 = metadata !{i32 786688, metadata !4, metadata !"X", metadata !5, i32 2,
722 metadata !11, i32 0, i32 0} ; [ DW_TAG_auto_variable ] [X] \
724 !11 = metadata !{i32 786468, null, null, metadata !"int", i32 0, i64 32,
725 i64 32, i64 0, i32 0, i32 5} ; [ DW_TAG_base_type ] [int] \
726 [line 0, size 32, align 32, offset 0, enc DW_ATE_signed]
727 !12 = metadata !{i32 2, i32 0, metadata !4, null}
728 !13 = metadata !{i32 786688, metadata !4, metadata !"Y", metadata !5, i32 3,
729 metadata !11, i32 0, i32 0} ; [ DW_TAG_auto_variable ] [Y] \
731 !14 = metadata !{i32 3, i32 0, metadata !4, null}
732 !15 = metadata !{i32 786688, metadata !16, metadata !"Z", metadata !5, i32 5,
733 metadata !11, i32 0, i32 0} ; [ DW_TAG_auto_variable ] [Z] \
735 !16 = metadata !{i32 786443, metadata !1, metadata !4, i32 4, i32 0, i32 0,
737 ; [ DW_TAG_lexical_block ] [/private/tmp/t.c]
738 !17 = metadata !{i32 5, i32 0, metadata !16, null}
739 !18 = metadata !{i32 6, i32 0, metadata !16, null}
740 !19 = metadata !{i32 8, i32 0, metadata !4, null} ; [ DW_TAG_imported_declaration ]
741 !20 = metadata !{i32 9, i32 0, metadata !4, null}
743 This example illustrates a few important details about LLVM debugging
744 information. In particular, it shows how the ``llvm.dbg.declare`` intrinsic and
745 location information, which are attached to an instruction, are applied
746 together to allow a debugger to analyze the relationship between statements,
747 variable definitions, and the code used to implement the function.
751 call void @llvm.dbg.declare(metadata !{i32* %X}, metadata !10), !dbg !12
752 ; [debug line = 2:7] [debug variable = X]
754 The first intrinsic ``%llvm.dbg.declare`` encodes debugging information for the
755 variable ``X``. The metadata ``!dbg !12`` attached to the intrinsic provides
756 scope information for the variable ``X``.
760 !12 = metadata !{i32 2, i32 0, metadata !4, null}
761 !4 = metadata !{i32 786478, metadata !1, metadata !5, metadata !"foo",
762 metadata !"foo", metadata !"", i32 1, metadata !6,
763 i1 false, i1 true, i32 0, i32 0, null, i32 0, i1 false,
764 void ()* @foo, null, null, metadata !2, i32 1}
765 ; [ DW_TAG_subprogram ] [line 1] [def] [foo]
767 Here ``!12`` is metadata providing location information. It has four fields:
768 line number, column number, scope, and original scope. The original scope
769 represents inline location if this instruction is inlined inside a caller, and
770 is null otherwise. In this example, scope is encoded by ``!4``, a
771 :ref:`subprogram descriptor <format_subprograms>`. This way the location
772 information attached to the intrinsics indicates that the variable ``X`` is
773 declared at line number 2 at a function level scope in function ``foo``.
775 Now lets take another example.
779 call void @llvm.dbg.declare(metadata !{i32* %Z}, metadata !15), !dbg !17
780 ; [debug line = 5:9] [debug variable = Z]
782 The third intrinsic ``%llvm.dbg.declare`` encodes debugging information for
783 variable ``Z``. The metadata ``!dbg !17`` attached to the intrinsic provides
784 scope information for the variable ``Z``.
788 !16 = metadata !{i32 786443, metadata !1, metadata !4, i32 4, i32 0, i32 0,
790 ; [ DW_TAG_lexical_block ] [/private/tmp/t.c]
791 !17 = metadata !{i32 5, i32 0, metadata !16, null}
793 Here ``!15`` indicates that ``Z`` is declared at line number 5 and
794 column number 0 inside of lexical scope ``!16``. The lexical scope itself
795 resides inside of subprogram ``!4`` described above.
797 The scope information attached with each instruction provides a straightforward
798 way to find instructions covered by a scope.
802 C/C++ front-end specific debug information
803 ==========================================
805 The C and C++ front-ends represent information about the program in a format
806 that is effectively identical to `DWARF 3.0
807 <http://www.eagercon.com/dwarf/dwarf3std.htm>`_ in terms of information
808 content. This allows code generators to trivially support native debuggers by
809 generating standard dwarf information, and contains enough information for
810 non-dwarf targets to translate it as needed.
812 This section describes the forms used to represent C and C++ programs. Other
813 languages could pattern themselves after this (which itself is tuned to
814 representing programs in the same way that DWARF 3 does), or they could choose
815 to provide completely different forms if they don't fit into the DWARF model.
816 As support for debugging information gets added to the various LLVM
817 source-language front-ends, the information used should be documented here.
819 The following sections provide examples of various C/C++ constructs and the
820 debug information that would best describe those constructs.
822 C/C++ source file information
823 -----------------------------
825 Given the source files ``MySource.cpp`` and ``MyHeader.h`` located in the
826 directory ``/Users/mine/sources``, the following code:
830 #include "MyHeader.h"
832 int main(int argc, char *argv[]) {
836 a C/C++ front-end would generate the following descriptors:
842 ;; Define the compile unit for the main source file "/Users/mine/sources/MySource.cpp".
846 metadata !1, ;; File/directory name
847 i32 4, ;; Language Id
848 metadata !"clang version 3.4 ",
849 i1 false, ;; Optimized compile unit
850 metadata !"", ;; Compiler flags
851 i32 0, ;; Runtime version
852 metadata !2, ;; Enumeration types
853 metadata !2, ;; Retained types
854 metadata !3, ;; Subprograms
855 metadata !2, ;; Global variables
856 metadata !2, ;; Imported entities (declarations and namespaces)
857 metadata !"" ;; Split debug filename
861 ;; Define the file for the file "/Users/mine/sources/MySource.cpp".
864 metadata !"MySource.cpp",
865 metadata !"/Users/mine/sources"
873 ;; Define the file for the file "/Users/mine/sources/Myheader.h"
880 metadata !"./MyHeader.h",
881 metadata !"/Users/mine/sources",
886 ``llvm::Instruction`` provides easy access to metadata attached with an
887 instruction. One can extract line number information encoded in LLVM IR using
888 ``Instruction::getMetadata()`` and ``DILocation::getLineNumber()``.
892 if (MDNode *N = I->getMetadata("dbg")) { // Here I is an LLVM instruction
893 DILocation Loc(N); // DILocation is in DebugInfo.h
894 unsigned Line = Loc.getLineNumber();
895 StringRef File = Loc.getFilename();
896 StringRef Dir = Loc.getDirectory();
899 C/C++ global variable information
900 ---------------------------------
902 Given an integer global variable declared as follows:
908 a C/C++ front-end would generate the following descriptors:
913 ;; Define the global itself.
915 %MyGlobal = global int 100
918 ;; List of debug info of globals
922 ;; Define the compile unit.
927 metadata !"foo.cpp", ;; File
928 metadata !"/Volumes/Data/tmp", ;; Directory
929 metadata !"clang version 3.1 ", ;; Producer
930 i1 true, ;; Deprecated field
931 i1 false, ;; "isOptimized"?
932 metadata !"", ;; Flags
933 i32 0, ;; Runtime Version
934 metadata !1, ;; Enum Types
935 metadata !1, ;; Retained Types
936 metadata !1, ;; Subprograms
937 metadata !3, ;; Global Variables
938 metadata !1, ;; Imported entities
939 "", ;; Split debug filename
940 } ; [ DW_TAG_compile_unit ]
942 ;; The Array of Global Variables
948 ;; Define the global variable itself.
954 metadata !"MyGlobal", ;; Name
955 metadata !"MyGlobal", ;; Display Name
956 metadata !"", ;; Linkage Name
960 i32 0, ;; IsLocalToUnit
961 i32 1, ;; IsDefinition
962 i32* @MyGlobal, ;; LLVM-IR Value
963 null ;; Static member declaration
964 } ; [ DW_TAG_variable ]
970 metadata !"foo.cpp", ;; File
971 metadata !"/Volumes/Data/tmp", ;; Directory
975 metadata !5 ;; Unused
976 } ; [ DW_TAG_file_type ]
985 metadata !"int", ;; Name
987 i64 32, ;; Size in Bits
988 i64 32, ;; Align in Bits
992 } ; [ DW_TAG_base_type ]
994 C/C++ function information
995 --------------------------
997 Given a function declared as follows:
1001 int main(int argc, char *argv[]) {
1005 a C/C++ front-end would generate the following descriptors:
1007 .. code-block:: llvm
1010 ;; Define the anchor for subprograms.
1014 metadata !1, ;; File
1015 metadata !1, ;; Context
1016 metadata !"main", ;; Name
1017 metadata !"main", ;; Display name
1018 metadata !"main", ;; Linkage name
1019 i32 1, ;; Line number
1020 metadata !4, ;; Type
1021 i1 false, ;; Is local
1022 i1 true, ;; Is definition
1023 i32 0, ;; Virtuality attribute, e.g. pure virtual function
1024 i32 0, ;; Index into virtual table for C++ methods
1025 i32 0, ;; Type that holds virtual table.
1027 i1 false, ;; True if this function is optimized
1028 Function *, ;; Pointer to llvm::Function
1029 null, ;; Function template parameters
1030 null, ;; List of function variables (emitted when optimizing)
1031 1 ;; Line number of the opening '{' of the function
1034 ;; Define the subprogram itself.
1036 define i32 @main(i32 %argc, i8** %argv) {
1043 The following are the basic type descriptors for C/C++ core types:
1048 .. code-block:: llvm
1054 metadata !"bool", ;; Name
1055 i32 0, ;; Line number
1056 i64 8, ;; Size in Bits
1057 i64 8, ;; Align in Bits
1058 i64 0, ;; Offset in Bits
1066 .. code-block:: llvm
1072 metadata !"char", ;; Name
1073 i32 0, ;; Line number
1074 i64 8, ;; Size in Bits
1075 i64 8, ;; Align in Bits
1076 i64 0, ;; Offset in Bits
1084 .. code-block:: llvm
1090 metadata !"unsigned char",
1091 i32 0, ;; Line number
1092 i64 8, ;; Size in Bits
1093 i64 8, ;; Align in Bits
1094 i64 0, ;; Offset in Bits
1102 .. code-block:: llvm
1108 metadata !"short int",
1109 i32 0, ;; Line number
1110 i64 16, ;; Size in Bits
1111 i64 16, ;; Align in Bits
1112 i64 0, ;; Offset in Bits
1120 .. code-block:: llvm
1126 metadata !"short unsigned int",
1127 i32 0, ;; Line number
1128 i64 16, ;; Size in Bits
1129 i64 16, ;; Align in Bits
1130 i64 0, ;; Offset in Bits
1138 .. code-block:: llvm
1144 metadata !"int", ;; Name
1145 i32 0, ;; Line number
1146 i64 32, ;; Size in Bits
1147 i64 32, ;; Align in Bits
1148 i64 0, ;; Offset in Bits
1156 .. code-block:: llvm
1162 metadata !"unsigned int",
1163 i32 0, ;; Line number
1164 i64 32, ;; Size in Bits
1165 i64 32, ;; Align in Bits
1166 i64 0, ;; Offset in Bits
1174 .. code-block:: llvm
1180 metadata !"long long int",
1181 i32 0, ;; Line number
1182 i64 64, ;; Size in Bits
1183 i64 64, ;; Align in Bits
1184 i64 0, ;; Offset in Bits
1192 .. code-block:: llvm
1198 metadata !"long long unsigned int",
1199 i32 0, ;; Line number
1200 i64 64, ;; Size in Bits
1201 i64 64, ;; Align in Bits
1202 i64 0, ;; Offset in Bits
1210 .. code-block:: llvm
1217 i32 0, ;; Line number
1218 i64 32, ;; Size in Bits
1219 i64 32, ;; Align in Bits
1220 i64 0, ;; Offset in Bits
1228 .. code-block:: llvm
1234 metadata !"double",;; Name
1235 i32 0, ;; Line number
1236 i64 64, ;; Size in Bits
1237 i64 64, ;; Align in Bits
1238 i64 0, ;; Offset in Bits
1246 Given the following as an example of C/C++ derived type:
1250 typedef const int *IntPtr;
1252 a C/C++ front-end would generate the following descriptors:
1254 .. code-block:: llvm
1257 ;; Define the typedef "IntPtr".
1261 metadata !3, ;; File
1262 metadata !1, ;; Context
1263 metadata !"IntPtr", ;; Name
1264 i32 0, ;; Line number
1265 i64 0, ;; Size in bits
1266 i64 0, ;; Align in bits
1267 i64 0, ;; Offset in bits
1269 metadata !4 ;; Derived From type
1272 ;; Define the pointer type.
1278 metadata !"", ;; Name
1279 i32 0, ;; Line number
1280 i64 64, ;; Size in bits
1281 i64 64, ;; Align in bits
1282 i64 0, ;; Offset in bits
1284 metadata !5 ;; Derived From type
1287 ;; Define the const type.
1293 metadata !"", ;; Name
1294 i32 0, ;; Line number
1295 i64 0, ;; Size in bits
1296 i64 0, ;; Align in bits
1297 i64 0, ;; Offset in bits
1299 metadata !6 ;; Derived From type
1302 ;; Define the int type.
1308 metadata !"int", ;; Name
1309 i32 0, ;; Line number
1310 i64 32, ;; Size in bits
1311 i64 32, ;; Align in bits
1312 i64 0, ;; Offset in bits
1317 C/C++ struct/union types
1318 ------------------------
1320 Given the following as an example of C/C++ struct type:
1330 a C/C++ front-end would generate the following descriptors:
1332 .. code-block:: llvm
1335 ;; Define basic type for unsigned int.
1341 metadata !"unsigned int",
1342 i32 0, ;; Line number
1343 i64 32, ;; Size in Bits
1344 i64 32, ;; Align in Bits
1345 i64 0, ;; Offset in Bits
1350 ;; Define composite type for struct Color.
1354 metadata !1, ;; Compile unit
1356 metadata !"Color", ;; Name
1357 i32 1, ;; Line number
1358 i64 96, ;; Size in bits
1359 i64 32, ;; Align in bits
1360 i64 0, ;; Offset in bits
1362 null, ;; Derived From
1363 metadata !3, ;; Elements
1364 i32 0, ;; Runtime Language
1365 null, ;; Base type containing the vtable pointer for this type
1366 null ;; Template parameters
1370 ;; Define the Red field.
1374 metadata !1, ;; File
1375 metadata !1, ;; Context
1376 metadata !"Red", ;; Name
1377 i32 2, ;; Line number
1378 i64 32, ;; Size in bits
1379 i64 32, ;; Align in bits
1380 i64 0, ;; Offset in bits
1382 metadata !5 ;; Derived From type
1386 ;; Define the Green field.
1390 metadata !1, ;; File
1391 metadata !1, ;; Context
1392 metadata !"Green", ;; Name
1393 i32 3, ;; Line number
1394 i64 32, ;; Size in bits
1395 i64 32, ;; Align in bits
1396 i64 32, ;; Offset in bits
1398 metadata !5 ;; Derived From type
1402 ;; Define the Blue field.
1406 metadata !1, ;; File
1407 metadata !1, ;; Context
1408 metadata !"Blue", ;; Name
1409 i32 4, ;; Line number
1410 i64 32, ;; Size in bits
1411 i64 32, ;; Align in bits
1412 i64 64, ;; Offset in bits
1414 metadata !5 ;; Derived From type
1418 ;; Define the array of fields used by the composite type Color.
1420 !3 = metadata !{metadata !4, metadata !6, metadata !7}
1422 C/C++ enumeration types
1423 -----------------------
1425 Given the following as an example of C/C++ enumeration type:
1435 a C/C++ front-end would generate the following descriptors:
1437 .. code-block:: llvm
1440 ;; Define composite type for enum Trees
1444 metadata !1, ;; File
1445 metadata !1, ;; Context
1446 metadata !"Trees", ;; Name
1447 i32 1, ;; Line number
1448 i64 32, ;; Size in bits
1449 i64 32, ;; Align in bits
1450 i64 0, ;; Offset in bits
1452 null, ;; Derived From type
1453 metadata !3, ;; Elements
1454 i32 0 ;; Runtime language
1458 ;; Define the array of enumerators used by composite type Trees.
1460 !3 = metadata !{metadata !4, metadata !5, metadata !6}
1463 ;; Define Spruce enumerator.
1465 !4 = metadata !{i32 786472, metadata !"Spruce", i64 100}
1468 ;; Define Oak enumerator.
1470 !5 = metadata !{i32 786472, metadata !"Oak", i64 200}
1473 ;; Define Maple enumerator.
1475 !6 = metadata !{i32 786472, metadata !"Maple", i64 300}
1477 Debugging information format
1478 ============================
1480 Debugging Information Extension for Objective C Properties
1481 ----------------------------------------------------------
1486 Objective C provides a simpler way to declare and define accessor methods using
1487 declared properties. The language provides features to declare a property and
1488 to let compiler synthesize accessor methods.
1490 The debugger lets developer inspect Objective C interfaces and their instance
1491 variables and class variables. However, the debugger does not know anything
1492 about the properties defined in Objective C interfaces. The debugger consumes
1493 information generated by compiler in DWARF format. The format does not support
1494 encoding of Objective C properties. This proposal describes DWARF extensions to
1495 encode Objective C properties, which the debugger can use to let developers
1496 inspect Objective C properties.
1501 Objective C properties exist separately from class members. A property can be
1502 defined only by "setter" and "getter" selectors, and be calculated anew on each
1503 access. Or a property can just be a direct access to some declared ivar.
1504 Finally it can have an ivar "automatically synthesized" for it by the compiler,
1505 in which case the property can be referred to in user code directly using the
1506 standard C dereference syntax as well as through the property "dot" syntax, but
1507 there is no entry in the ``@interface`` declaration corresponding to this ivar.
1509 To facilitate debugging, these properties we will add a new DWARF TAG into the
1510 ``DW_TAG_structure_type`` definition for the class to hold the description of a
1511 given property, and a set of DWARF attributes that provide said description.
1512 The property tag will also contain the name and declared type of the property.
1514 If there is a related ivar, there will also be a DWARF property attribute placed
1515 in the ``DW_TAG_member`` DIE for that ivar referring back to the property TAG
1516 for that property. And in the case where the compiler synthesizes the ivar
1517 directly, the compiler is expected to generate a ``DW_TAG_member`` for that
1518 ivar (with the ``DW_AT_artificial`` set to 1), whose name will be the name used
1519 to access this ivar directly in code, and with the property attribute pointing
1520 back to the property it is backing.
1522 The following examples will serve as illustration for our discussion:
1524 .. code-block:: objc
1536 @synthesize p2 = n2;
1539 This produces the following DWARF (this is a "pseudo dwarfdump" output):
1541 .. code-block:: none
1543 0x00000100: TAG_structure_type [7] *
1544 AT_APPLE_runtime_class( 0x10 )
1546 AT_decl_file( "Objc_Property.m" )
1549 0x00000110 TAG_APPLE_property
1551 AT_type ( {0x00000150} ( int ) )
1553 0x00000120: TAG_APPLE_property
1555 AT_type ( {0x00000150} ( int ) )
1557 0x00000130: TAG_member [8]
1559 AT_APPLE_property ( {0x00000110} "p1" )
1560 AT_type( {0x00000150} ( int ) )
1561 AT_artificial ( 0x1 )
1563 0x00000140: TAG_member [8]
1565 AT_APPLE_property ( {0x00000120} "p2" )
1566 AT_type( {0x00000150} ( int ) )
1568 0x00000150: AT_type( ( int ) )
1570 Note, the current convention is that the name of the ivar for an
1571 auto-synthesized property is the name of the property from which it derives
1572 with an underscore prepended, as is shown in the example. But we actually
1573 don't need to know this convention, since we are given the name of the ivar
1576 Also, it is common practice in ObjC to have different property declarations in
1577 the @interface and @implementation - e.g. to provide a read-only property in
1578 the interface,and a read-write interface in the implementation. In that case,
1579 the compiler should emit whichever property declaration will be in force in the
1580 current translation unit.
1582 Developers can decorate a property with attributes which are encoded using
1583 ``DW_AT_APPLE_property_attribute``.
1585 .. code-block:: objc
1587 @property (readonly, nonatomic) int pr;
1589 .. code-block:: none
1591 TAG_APPLE_property [8]
1593 AT_type ( {0x00000147} (int) )
1594 AT_APPLE_property_attribute (DW_APPLE_PROPERTY_readonly, DW_APPLE_PROPERTY_nonatomic)
1596 The setter and getter method names are attached to the property using
1597 ``DW_AT_APPLE_property_setter`` and ``DW_AT_APPLE_property_getter`` attributes.
1599 .. code-block:: objc
1602 @property (setter=myOwnP3Setter:) int p3;
1603 -(void)myOwnP3Setter:(int)a;
1608 -(void)myOwnP3Setter:(int)a{ }
1611 The DWARF for this would be:
1613 .. code-block:: none
1615 0x000003bd: TAG_structure_type [7] *
1616 AT_APPLE_runtime_class( 0x10 )
1618 AT_decl_file( "Objc_Property.m" )
1621 0x000003cd TAG_APPLE_property
1623 AT_APPLE_property_setter ( "myOwnP3Setter:" )
1624 AT_type( {0x00000147} ( int ) )
1626 0x000003f3: TAG_member [8]
1628 AT_type ( {0x00000147} ( int ) )
1629 AT_APPLE_property ( {0x000003cd} )
1630 AT_artificial ( 0x1 )
1635 +-----------------------+--------+
1637 +=======================+========+
1638 | DW_TAG_APPLE_property | 0x4200 |
1639 +-----------------------+--------+
1641 New DWARF Attributes
1642 ^^^^^^^^^^^^^^^^^^^^
1644 +--------------------------------+--------+-----------+
1645 | Attribute | Value | Classes |
1646 +================================+========+===========+
1647 | DW_AT_APPLE_property | 0x3fed | Reference |
1648 +--------------------------------+--------+-----------+
1649 | DW_AT_APPLE_property_getter | 0x3fe9 | String |
1650 +--------------------------------+--------+-----------+
1651 | DW_AT_APPLE_property_setter | 0x3fea | String |
1652 +--------------------------------+--------+-----------+
1653 | DW_AT_APPLE_property_attribute | 0x3feb | Constant |
1654 +--------------------------------+--------+-----------+
1659 +--------------------------------+-------+
1661 +================================+=======+
1662 | DW_AT_APPLE_PROPERTY_readonly | 0x1 |
1663 +--------------------------------+-------+
1664 | DW_AT_APPLE_PROPERTY_readwrite | 0x2 |
1665 +--------------------------------+-------+
1666 | DW_AT_APPLE_PROPERTY_assign | 0x4 |
1667 +--------------------------------+-------+
1668 | DW_AT_APPLE_PROPERTY_retain | 0x8 |
1669 +--------------------------------+-------+
1670 | DW_AT_APPLE_PROPERTY_copy | 0x10 |
1671 +--------------------------------+-------+
1672 | DW_AT_APPLE_PROPERTY_nonatomic | 0x20 |
1673 +--------------------------------+-------+
1675 Name Accelerator Tables
1676 -----------------------
1681 The "``.debug_pubnames``" and "``.debug_pubtypes``" formats are not what a
1682 debugger needs. The "``pub``" in the section name indicates that the entries
1683 in the table are publicly visible names only. This means no static or hidden
1684 functions show up in the "``.debug_pubnames``". No static variables or private
1685 class variables are in the "``.debug_pubtypes``". Many compilers add different
1686 things to these tables, so we can't rely upon the contents between gcc, icc, or
1689 The typical query given by users tends not to match up with the contents of
1690 these tables. For example, the DWARF spec states that "In the case of the name
1691 of a function member or static data member of a C++ structure, class or union,
1692 the name presented in the "``.debug_pubnames``" section is not the simple name
1693 given by the ``DW_AT_name attribute`` of the referenced debugging information
1694 entry, but rather the fully qualified name of the data or function member."
1695 So the only names in these tables for complex C++ entries is a fully
1696 qualified name. Debugger users tend not to enter their search strings as
1697 "``a::b::c(int,const Foo&) const``", but rather as "``c``", "``b::c``" , or
1698 "``a::b::c``". So the name entered in the name table must be demangled in
1699 order to chop it up appropriately and additional names must be manually entered
1700 into the table to make it effective as a name lookup table for debuggers to
1703 All debuggers currently ignore the "``.debug_pubnames``" table as a result of
1704 its inconsistent and useless public-only name content making it a waste of
1705 space in the object file. These tables, when they are written to disk, are not
1706 sorted in any way, leaving every debugger to do its own parsing and sorting.
1707 These tables also include an inlined copy of the string values in the table
1708 itself making the tables much larger than they need to be on disk, especially
1709 for large C++ programs.
1711 Can't we just fix the sections by adding all of the names we need to this
1712 table? No, because that is not what the tables are defined to contain and we
1713 won't know the difference between the old bad tables and the new good tables.
1714 At best we could make our own renamed sections that contain all of the data we
1717 These tables are also insufficient for what a debugger like LLDB needs. LLDB
1718 uses clang for its expression parsing where LLDB acts as a PCH. LLDB is then
1719 often asked to look for type "``foo``" or namespace "``bar``", or list items in
1720 namespace "``baz``". Namespaces are not included in the pubnames or pubtypes
1721 tables. Since clang asks a lot of questions when it is parsing an expression,
1722 we need to be very fast when looking up names, as it happens a lot. Having new
1723 accelerator tables that are optimized for very quick lookups will benefit this
1724 type of debugging experience greatly.
1726 We would like to generate name lookup tables that can be mapped into memory
1727 from disk, and used as is, with little or no up-front parsing. We would also
1728 be able to control the exact content of these different tables so they contain
1729 exactly what we need. The Name Accelerator Tables were designed to fix these
1730 issues. In order to solve these issues we need to:
1732 * Have a format that can be mapped into memory from disk and used as is
1733 * Lookups should be very fast
1734 * Extensible table format so these tables can be made by many producers
1735 * Contain all of the names needed for typical lookups out of the box
1736 * Strict rules for the contents of tables
1738 Table size is important and the accelerator table format should allow the reuse
1739 of strings from common string tables so the strings for the names are not
1740 duplicated. We also want to make sure the table is ready to be used as-is by
1741 simply mapping the table into memory with minimal header parsing.
1743 The name lookups need to be fast and optimized for the kinds of lookups that
1744 debuggers tend to do. Optimally we would like to touch as few parts of the
1745 mapped table as possible when doing a name lookup and be able to quickly find
1746 the name entry we are looking for, or discover there are no matches. In the
1747 case of debuggers we optimized for lookups that fail most of the time.
1749 Each table that is defined should have strict rules on exactly what is in the
1750 accelerator tables and documented so clients can rely on the content.
1755 Standard Hash Tables
1756 """"""""""""""""""""
1758 Typical hash tables have a header, buckets, and each bucket points to the
1761 .. code-block:: none
1771 The BUCKETS are an array of offsets to DATA for each hash:
1773 .. code-block:: none
1776 | 0x00001000 | BUCKETS[0]
1777 | 0x00002000 | BUCKETS[1]
1778 | 0x00002200 | BUCKETS[2]
1779 | 0x000034f0 | BUCKETS[3]
1781 | 0xXXXXXXXX | BUCKETS[n_buckets]
1784 So for ``bucket[3]`` in the example above, we have an offset into the table
1785 0x000034f0 which points to a chain of entries for the bucket. Each bucket must
1786 contain a next pointer, full 32 bit hash value, the string itself, and the data
1787 for the current string value.
1789 .. code-block:: none
1792 0x000034f0: | 0x00003500 | next pointer
1793 | 0x12345678 | 32 bit hash
1794 | "erase" | string value
1795 | data[n] | HashData for this bucket
1797 0x00003500: | 0x00003550 | next pointer
1798 | 0x29273623 | 32 bit hash
1799 | "dump" | string value
1800 | data[n] | HashData for this bucket
1802 0x00003550: | 0x00000000 | next pointer
1803 | 0x82638293 | 32 bit hash
1804 | "main" | string value
1805 | data[n] | HashData for this bucket
1808 The problem with this layout for debuggers is that we need to optimize for the
1809 negative lookup case where the symbol we're searching for is not present. So
1810 if we were to lookup "``printf``" in the table above, we would make a 32 hash
1811 for "``printf``", it might match ``bucket[3]``. We would need to go to the
1812 offset 0x000034f0 and start looking to see if our 32 bit hash matches. To do
1813 so, we need to read the next pointer, then read the hash, compare it, and skip
1814 to the next bucket. Each time we are skipping many bytes in memory and
1815 touching new cache pages just to do the compare on the full 32 bit hash. All
1816 of these accesses then tell us that we didn't have a match.
1821 To solve the issues mentioned above we have structured the hash tables a bit
1822 differently: a header, buckets, an array of all unique 32 bit hash values,
1823 followed by an array of hash value data offsets, one for each hash value, then
1824 the data for all hash values:
1826 .. code-block:: none
1840 The ``BUCKETS`` in the name tables are an index into the ``HASHES`` array. By
1841 making all of the full 32 bit hash values contiguous in memory, we allow
1842 ourselves to efficiently check for a match while touching as little memory as
1843 possible. Most often checking the 32 bit hash values is as far as the lookup
1844 goes. If it does match, it usually is a match with no collisions. So for a
1845 table with "``n_buckets``" buckets, and "``n_hashes``" unique 32 bit hash
1846 values, we can clarify the contents of the ``BUCKETS``, ``HASHES`` and
1849 .. code-block:: none
1851 .-------------------------.
1852 | HEADER.magic | uint32_t
1853 | HEADER.version | uint16_t
1854 | HEADER.hash_function | uint16_t
1855 | HEADER.bucket_count | uint32_t
1856 | HEADER.hashes_count | uint32_t
1857 | HEADER.header_data_len | uint32_t
1858 | HEADER_DATA | HeaderData
1859 |-------------------------|
1860 | BUCKETS | uint32_t[n_buckets] // 32 bit hash indexes
1861 |-------------------------|
1862 | HASHES | uint32_t[n_hashes] // 32 bit hash values
1863 |-------------------------|
1864 | OFFSETS | uint32_t[n_hashes] // 32 bit offsets to hash value data
1865 |-------------------------|
1867 `-------------------------'
1869 So taking the exact same data from the standard hash example above we end up
1872 .. code-block:: none
1882 | ... | BUCKETS[n_buckets]
1884 | 0x........ | HASHES[0]
1885 | 0x........ | HASHES[1]
1886 | 0x........ | HASHES[2]
1887 | 0x........ | HASHES[3]
1888 | 0x........ | HASHES[4]
1889 | 0x........ | HASHES[5]
1890 | 0x12345678 | HASHES[6] hash for BUCKETS[3]
1891 | 0x29273623 | HASHES[7] hash for BUCKETS[3]
1892 | 0x82638293 | HASHES[8] hash for BUCKETS[3]
1893 | 0x........ | HASHES[9]
1894 | 0x........ | HASHES[10]
1895 | 0x........ | HASHES[11]
1896 | 0x........ | HASHES[12]
1897 | 0x........ | HASHES[13]
1898 | 0x........ | HASHES[n_hashes]
1900 | 0x........ | OFFSETS[0]
1901 | 0x........ | OFFSETS[1]
1902 | 0x........ | OFFSETS[2]
1903 | 0x........ | OFFSETS[3]
1904 | 0x........ | OFFSETS[4]
1905 | 0x........ | OFFSETS[5]
1906 | 0x000034f0 | OFFSETS[6] offset for BUCKETS[3]
1907 | 0x00003500 | OFFSETS[7] offset for BUCKETS[3]
1908 | 0x00003550 | OFFSETS[8] offset for BUCKETS[3]
1909 | 0x........ | OFFSETS[9]
1910 | 0x........ | OFFSETS[10]
1911 | 0x........ | OFFSETS[11]
1912 | 0x........ | OFFSETS[12]
1913 | 0x........ | OFFSETS[13]
1914 | 0x........ | OFFSETS[n_hashes]
1922 0x000034f0: | 0x00001203 | .debug_str ("erase")
1923 | 0x00000004 | A 32 bit array count - number of HashData with name "erase"
1924 | 0x........ | HashData[0]
1925 | 0x........ | HashData[1]
1926 | 0x........ | HashData[2]
1927 | 0x........ | HashData[3]
1928 | 0x00000000 | String offset into .debug_str (terminate data for hash)
1930 0x00003500: | 0x00001203 | String offset into .debug_str ("collision")
1931 | 0x00000002 | A 32 bit array count - number of HashData with name "collision"
1932 | 0x........ | HashData[0]
1933 | 0x........ | HashData[1]
1934 | 0x00001203 | String offset into .debug_str ("dump")
1935 | 0x00000003 | A 32 bit array count - number of HashData with name "dump"
1936 | 0x........ | HashData[0]
1937 | 0x........ | HashData[1]
1938 | 0x........ | HashData[2]
1939 | 0x00000000 | String offset into .debug_str (terminate data for hash)
1941 0x00003550: | 0x00001203 | String offset into .debug_str ("main")
1942 | 0x00000009 | A 32 bit array count - number of HashData with name "main"
1943 | 0x........ | HashData[0]
1944 | 0x........ | HashData[1]
1945 | 0x........ | HashData[2]
1946 | 0x........ | HashData[3]
1947 | 0x........ | HashData[4]
1948 | 0x........ | HashData[5]
1949 | 0x........ | HashData[6]
1950 | 0x........ | HashData[7]
1951 | 0x........ | HashData[8]
1952 | 0x00000000 | String offset into .debug_str (terminate data for hash)
1955 So we still have all of the same data, we just organize it more efficiently for
1956 debugger lookup. If we repeat the same "``printf``" lookup from above, we
1957 would hash "``printf``" and find it matches ``BUCKETS[3]`` by taking the 32 bit
1958 hash value and modulo it by ``n_buckets``. ``BUCKETS[3]`` contains "6" which
1959 is the index into the ``HASHES`` table. We would then compare any consecutive
1960 32 bit hashes values in the ``HASHES`` array as long as the hashes would be in
1961 ``BUCKETS[3]``. We do this by verifying that each subsequent hash value modulo
1962 ``n_buckets`` is still 3. In the case of a failed lookup we would access the
1963 memory for ``BUCKETS[3]``, and then compare a few consecutive 32 bit hashes
1964 before we know that we have no match. We don't end up marching through
1965 multiple words of memory and we really keep the number of processor data cache
1966 lines being accessed as small as possible.
1968 The string hash that is used for these lookup tables is the Daniel J.
1969 Bernstein hash which is also used in the ELF ``GNU_HASH`` sections. It is a
1970 very good hash for all kinds of names in programs with very few hash
1973 Empty buckets are designated by using an invalid hash index of ``UINT32_MAX``.
1978 These name hash tables are designed to be generic where specializations of the
1979 table get to define additional data that goes into the header ("``HeaderData``"),
1980 how the string value is stored ("``KeyType``") and the content of the data for each
1986 The header has a fixed part, and the specialized part. The exact format of the
1993 uint32_t magic; // 'HASH' magic value to allow endian detection
1994 uint16_t version; // Version number
1995 uint16_t hash_function; // The hash function enumeration that was used
1996 uint32_t bucket_count; // The number of buckets in this hash table
1997 uint32_t hashes_count; // The total number of unique hash values and hash data offsets in this table
1998 uint32_t header_data_len; // The bytes to skip to get to the hash indexes (buckets) for correct alignment
1999 // Specifically the length of the following HeaderData field - this does not
2000 // include the size of the preceding fields
2001 HeaderData header_data; // Implementation specific header data
2004 The header starts with a 32 bit "``magic``" value which must be ``'HASH'``
2005 encoded as an ASCII integer. This allows the detection of the start of the
2006 hash table and also allows the table's byte order to be determined so the table
2007 can be correctly extracted. The "``magic``" value is followed by a 16 bit
2008 ``version`` number which allows the table to be revised and modified in the
2009 future. The current version number is 1. ``hash_function`` is a ``uint16_t``
2010 enumeration that specifies which hash function was used to produce this table.
2011 The current values for the hash function enumerations include:
2015 enum HashFunctionType
2017 eHashFunctionDJB = 0u, // Daniel J Bernstein hash function
2020 ``bucket_count`` is a 32 bit unsigned integer that represents how many buckets
2021 are in the ``BUCKETS`` array. ``hashes_count`` is the number of unique 32 bit
2022 hash values that are in the ``HASHES`` array, and is the same number of offsets
2023 are contained in the ``OFFSETS`` array. ``header_data_len`` specifies the size
2024 in bytes of the ``HeaderData`` that is filled in by specialized versions of
2030 The header is followed by the buckets, hashes, offsets, and hash value data.
2036 uint32_t buckets[Header.bucket_count]; // An array of hash indexes into the "hashes[]" array below
2037 uint32_t hashes [Header.hashes_count]; // Every unique 32 bit hash for the entire table is in this table
2038 uint32_t offsets[Header.hashes_count]; // An offset that corresponds to each item in the "hashes[]" array above
2041 ``buckets`` is an array of 32 bit indexes into the ``hashes`` array. The
2042 ``hashes`` array contains all of the 32 bit hash values for all names in the
2043 hash table. Each hash in the ``hashes`` table has an offset in the ``offsets``
2044 array that points to the data for the hash value.
2046 This table setup makes it very easy to repurpose these tables to contain
2047 different data, while keeping the lookup mechanism the same for all tables.
2048 This layout also makes it possible to save the table to disk and map it in
2049 later and do very efficient name lookups with little or no parsing.
2051 DWARF lookup tables can be implemented in a variety of ways and can store a lot
2052 of information for each name. We want to make the DWARF tables extensible and
2053 able to store the data efficiently so we have used some of the DWARF features
2054 that enable efficient data storage to define exactly what kind of data we store
2057 The ``HeaderData`` contains a definition of the contents of each HashData chunk.
2058 We might want to store an offset to all of the debug information entries (DIEs)
2059 for each name. To keep things extensible, we create a list of items, or
2060 Atoms, that are contained in the data for each name. First comes the type of
2061 the data in each atom:
2068 eAtomTypeDIEOffset = 1u, // DIE offset, check form for encoding
2069 eAtomTypeCUOffset = 2u, // DIE offset of the compiler unit header that contains the item in question
2070 eAtomTypeTag = 3u, // DW_TAG_xxx value, should be encoded as DW_FORM_data1 (if no tags exceed 255) or DW_FORM_data2
2071 eAtomTypeNameFlags = 4u, // Flags from enum NameFlags
2072 eAtomTypeTypeFlags = 5u, // Flags from enum TypeFlags
2075 The enumeration values and their meanings are:
2077 .. code-block:: none
2079 eAtomTypeNULL - a termination atom that specifies the end of the atom list
2080 eAtomTypeDIEOffset - an offset into the .debug_info section for the DWARF DIE for this name
2081 eAtomTypeCUOffset - an offset into the .debug_info section for the CU that contains the DIE
2082 eAtomTypeDIETag - The DW_TAG_XXX enumeration value so you don't have to parse the DWARF to see what it is
2083 eAtomTypeNameFlags - Flags for functions and global variables (isFunction, isInlined, isExternal...)
2084 eAtomTypeTypeFlags - Flags for types (isCXXClass, isObjCClass, ...)
2086 Then we allow each atom type to define the atom type and how the data for each
2087 atom type data is encoded:
2093 uint16_t type; // AtomType enum value
2094 uint16_t form; // DWARF DW_FORM_XXX defines
2097 The ``form`` type above is from the DWARF specification and defines the exact
2098 encoding of the data for the Atom type. See the DWARF specification for the
2099 ``DW_FORM_`` definitions.
2105 uint32_t die_offset_base;
2106 uint32_t atom_count;
2107 Atoms atoms[atom_count0];
2110 ``HeaderData`` defines the base DIE offset that should be added to any atoms
2111 that are encoded using the ``DW_FORM_ref1``, ``DW_FORM_ref2``,
2112 ``DW_FORM_ref4``, ``DW_FORM_ref8`` or ``DW_FORM_ref_udata``. It also defines
2113 what is contained in each ``HashData`` object -- ``Atom.form`` tells us how large
2114 each field will be in the ``HashData`` and the ``Atom.type`` tells us how this data
2115 should be interpreted.
2117 For the current implementations of the "``.apple_names``" (all functions +
2118 globals), the "``.apple_types``" (names of all types that are defined), and
2119 the "``.apple_namespaces``" (all namespaces), we currently set the ``Atom``
2124 HeaderData.atom_count = 1;
2125 HeaderData.atoms[0].type = eAtomTypeDIEOffset;
2126 HeaderData.atoms[0].form = DW_FORM_data4;
2128 This defines the contents to be the DIE offset (eAtomTypeDIEOffset) that is
2129 encoded as a 32 bit value (DW_FORM_data4). This allows a single name to have
2130 multiple matching DIEs in a single file, which could come up with an inlined
2131 function for instance. Future tables could include more information about the
2132 DIE such as flags indicating if the DIE is a function, method, block,
2135 The KeyType for the DWARF table is a 32 bit string table offset into the
2136 ".debug_str" table. The ".debug_str" is the string table for the DWARF which
2137 may already contain copies of all of the strings. This helps make sure, with
2138 help from the compiler, that we reuse the strings between all of the DWARF
2139 sections and keeps the hash table size down. Another benefit to having the
2140 compiler generate all strings as DW_FORM_strp in the debug info, is that
2141 DWARF parsing can be made much faster.
2143 After a lookup is made, we get an offset into the hash data. The hash data
2144 needs to be able to deal with 32 bit hash collisions, so the chunk of data
2145 at the offset in the hash data consists of a triple:
2150 uint32_t hash_data_count
2151 HashData[hash_data_count]
2153 If "str_offset" is zero, then the bucket contents are done. 99.9% of the
2154 hash data chunks contain a single item (no 32 bit hash collision):
2156 .. code-block:: none
2159 | 0x00001023 | uint32_t KeyType (.debug_str[0x0001023] => "main")
2160 | 0x00000004 | uint32_t HashData count
2161 | 0x........ | uint32_t HashData[0] DIE offset
2162 | 0x........ | uint32_t HashData[1] DIE offset
2163 | 0x........ | uint32_t HashData[2] DIE offset
2164 | 0x........ | uint32_t HashData[3] DIE offset
2165 | 0x00000000 | uint32_t KeyType (end of hash chain)
2168 If there are collisions, you will have multiple valid string offsets:
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 | 0x00002023 | uint32_t KeyType (.debug_str[0x0002023] => "print")
2180 | 0x00000002 | uint32_t HashData count
2181 | 0x........ | uint32_t HashData[0] DIE offset
2182 | 0x........ | uint32_t HashData[1] DIE offset
2183 | 0x00000000 | uint32_t KeyType (end of hash chain)
2186 Current testing with real world C++ binaries has shown that there is around 1
2187 32 bit hash collision per 100,000 name entries.
2192 As we said, we want to strictly define exactly what is included in the
2193 different tables. For DWARF, we have 3 tables: "``.apple_names``",
2194 "``.apple_types``", and "``.apple_namespaces``".
2196 "``.apple_names``" sections should contain an entry for each DWARF DIE whose
2197 ``DW_TAG`` is a ``DW_TAG_label``, ``DW_TAG_inlined_subroutine``, or
2198 ``DW_TAG_subprogram`` that has address attributes: ``DW_AT_low_pc``,
2199 ``DW_AT_high_pc``, ``DW_AT_ranges`` or ``DW_AT_entry_pc``. It also contains
2200 ``DW_TAG_variable`` DIEs that have a ``DW_OP_addr`` in the location (global and
2201 static variables). All global and static variables should be included,
2202 including those scoped within functions and classes. For example using the
2214 Both of the static ``var`` variables would be included in the table. All
2215 functions should emit both their full names and their basenames. For C or C++,
2216 the full name is the mangled name (if available) which is usually in the
2217 ``DW_AT_MIPS_linkage_name`` attribute, and the ``DW_AT_name`` contains the
2218 function basename. If global or static variables have a mangled name in a
2219 ``DW_AT_MIPS_linkage_name`` attribute, this should be emitted along with the
2220 simple name found in the ``DW_AT_name`` attribute.
2222 "``.apple_types``" sections should contain an entry for each DWARF DIE whose
2227 * DW_TAG_enumeration_type
2228 * DW_TAG_pointer_type
2229 * DW_TAG_reference_type
2230 * DW_TAG_string_type
2231 * DW_TAG_structure_type
2232 * DW_TAG_subroutine_type
2235 * DW_TAG_ptr_to_member_type
2237 * DW_TAG_subrange_type
2243 * DW_TAG_packed_type
2244 * DW_TAG_volatile_type
2245 * DW_TAG_restrict_type
2246 * DW_TAG_interface_type
2247 * DW_TAG_unspecified_type
2248 * DW_TAG_shared_type
2250 Only entries with a ``DW_AT_name`` attribute are included, and the entry must
2251 not be a forward declaration (``DW_AT_declaration`` attribute with a non-zero
2252 value). For example, using the following code:
2262 We get a few type DIEs:
2264 .. code-block:: none
2266 0x00000067: TAG_base_type [5]
2267 AT_encoding( DW_ATE_signed )
2269 AT_byte_size( 0x04 )
2271 0x0000006e: TAG_pointer_type [6]
2272 AT_type( {0x00000067} ( int ) )
2273 AT_byte_size( 0x08 )
2275 The DW_TAG_pointer_type is not included because it does not have a ``DW_AT_name``.
2277 "``.apple_namespaces``" section should contain all ``DW_TAG_namespace`` DIEs.
2278 If we run into a namespace that has no name this is an anonymous namespace, and
2279 the name should be output as "``(anonymous namespace)``" (without the quotes).
2280 Why? This matches the output of the ``abi::cxa_demangle()`` that is in the
2281 standard C++ library that demangles mangled names.
2284 Language Extensions and File Format Changes
2285 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2287 Objective-C Extensions
2288 """"""""""""""""""""""
2290 "``.apple_objc``" section should contain all ``DW_TAG_subprogram`` DIEs for an
2291 Objective-C class. The name used in the hash table is the name of the
2292 Objective-C class itself. If the Objective-C class has a category, then an
2293 entry is made for both the class name without the category, and for the class
2294 name with the category. So if we have a DIE at offset 0x1234 with a name of
2295 method "``-[NSString(my_additions) stringWithSpecialString:]``", we would add
2296 an entry for "``NSString``" that points to DIE 0x1234, and an entry for
2297 "``NSString(my_additions)``" that points to 0x1234. This allows us to quickly
2298 track down all Objective-C methods for an Objective-C class when doing
2299 expressions. It is needed because of the dynamic nature of Objective-C where
2300 anyone can add methods to a class. The DWARF for Objective-C methods is also
2301 emitted differently from C++ classes where the methods are not usually
2302 contained in the class definition, they are scattered about across one or more
2303 compile units. Categories can also be defined in different shared libraries.
2304 So we need to be able to quickly find all of the methods and class functions
2305 given the Objective-C class name, or quickly find all methods and class
2306 functions for a class + category name. This table does not contain any
2307 selector names, it just maps Objective-C class names (or class names +
2308 category) to all of the methods and class functions. The selectors are added
2309 as function basenames in the "``.debug_names``" section.
2311 In the "``.apple_names``" section for Objective-C functions, the full name is
2312 the entire function name with the brackets ("``-[NSString
2313 stringWithCString:]``") and the basename is the selector only
2314 ("``stringWithCString:``").
2319 The sections names for the apple hash tables are for non-mach-o files. For
2320 mach-o files, the sections should be contained in the ``__DWARF`` segment with
2323 * "``.apple_names``" -> "``__apple_names``"
2324 * "``.apple_types``" -> "``__apple_types``"
2325 * "``.apple_namespaces``" -> "``__apple_namespac``" (16 character limit)
2326 * "``.apple_objc``" -> "``__apple_objc``"