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 LLVM
173 debugging information is versioned, allowing backwards compatibility in the
174 case that the core structures need to change in some way. Also, all debugging
175 information objects start with a tag to indicate what type of object it is.
176 The source-language is allowed to define its own objects, by using unreserved
177 tag numbers. We recommend using with tags in the range 0x1000 through 0x2000
178 (there is a defined ``enum DW_TAG_user_base = 0x1000``.)
180 The fields of debug descriptors used internally by LLVM are restricted to only
181 the simple data types ``i32``, ``i1``, ``float``, ``double``, ``mdstring`` and
191 <a name="LLVMDebugVersion">The first field of a descriptor is always an
192 ``i32`` containing a tag value identifying the content of the descriptor.
193 The remaining fields are specific to the descriptor. The values of tags are
194 loosely bound to the tag values of DWARF information entries. However, that
195 does not restrict the use of the information supplied to DWARF targets. To
196 facilitate versioning of debug information, the tag is augmented with the
197 current debug version (``LLVMDebugVersion = 8 << 16`` or 0x80000 or
200 The details of the various descriptors follow.
202 Compile unit descriptors
203 ^^^^^^^^^^^^^^^^^^^^^^^^
208 i32, ;; Tag = 17 + LLVMDebugVersion (DW_TAG_compile_unit)
209 i32, ;; Unused field.
210 i32, ;; DWARF language identifier (ex. DW_LANG_C89)
211 metadata, ;; Source file name
212 metadata, ;; Source file directory (includes trailing slash)
213 metadata ;; Producer (ex. "4.0.1 LLVM (LLVM research group)")
214 i1, ;; True if this is a main compile unit.
215 i1, ;; True if this is optimized.
217 i32 ;; Runtime version
218 metadata ;; List of enums types
219 metadata ;; List of retained types
220 metadata ;; List of subprograms
221 metadata ;; List of global variables
224 These descriptors contain a source language ID for the file (we use the DWARF
225 3.0 ID numbers, such as ``DW_LANG_C89``, ``DW_LANG_C_plus_plus``,
226 ``DW_LANG_Cobol74``, etc), three strings describing the filename, working
227 directory of the compiler, and an identifier string for the compiler that
230 Compile unit descriptors provide the root context for objects declared in a
231 specific compilation unit. File descriptors are defined using this context.
232 These descriptors are collected by a named metadata ``!llvm.dbg.cu``. They
233 keep track of subprograms, global variables and type information.
243 i32, ;; Tag = 41 + LLVMDebugVersion (DW_TAG_file_type)
244 metadata, ;; Source file name
245 metadata, ;; Source file directory (includes trailing slash)
249 These descriptors contain information for a file. Global variables and top
250 level functions would be defined using this context. File descriptors also
251 provide context for source line correspondence.
253 Each input file is encoded as a separate file descriptor in LLVM debugging
256 .. _format_global_variables:
258 Global variable descriptors
259 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
264 i32, ;; Tag = 52 + LLVMDebugVersion (DW_TAG_variable)
265 i32, ;; Unused field.
266 metadata, ;; Reference to context descriptor
268 metadata, ;; Display name (fully qualified C++ name)
269 metadata, ;; MIPS linkage name (for C++)
270 metadata, ;; Reference to file where defined
271 i32, ;; Line number where defined
272 metadata, ;; Reference to type descriptor
273 i1, ;; True if the global is local to compile unit (static)
274 i1, ;; True if the global is defined in the compile unit (not extern)
275 {}* ;; Reference to the global variable
278 These descriptors provide debug information about globals variables. They
279 provide details such as name, type and where the variable is defined. All
280 global variables are collected inside the named metadata ``!llvm.dbg.cu``.
282 .. _format_subprograms:
284 Subprogram descriptors
285 ^^^^^^^^^^^^^^^^^^^^^^
290 i32, ;; Tag = 46 + LLVMDebugVersion (DW_TAG_subprogram)
291 i32, ;; Unused field.
292 metadata, ;; Reference to context descriptor
294 metadata, ;; Display name (fully qualified C++ name)
295 metadata, ;; MIPS linkage name (for C++)
296 metadata, ;; Reference to file where defined
297 i32, ;; Line number where defined
298 metadata, ;; Reference to type descriptor
299 i1, ;; True if the global is local to compile unit (static)
300 i1, ;; True if the global is defined in the compile unit (not extern)
301 i32, ;; Virtuality, e.g. dwarf::DW_VIRTUALITY__virtual
302 i32, ;; Index into a virtual function
303 metadata, ;; indicates which base type contains the vtable pointer for the
305 i32, ;; Flags - Artifical, Private, Protected, Explicit, Prototyped.
307 Function * , ;; Pointer to LLVM function
308 metadata, ;; Lists function template parameters
309 metadata, ;; Function declaration descriptor
310 metadata, ;; List of function variables
311 i32 ;; Line number where the scope of the subprogram begins
314 These descriptors provide debug information about functions, methods and
315 subprograms. They provide details such as name, return types and the source
316 location where the subprogram is defined.
324 i32, ;; Tag = 11 + LLVMDebugVersion (DW_TAG_lexical_block)
325 metadata,;; Reference to context descriptor
327 i32, ;; Column number
328 metadata,;; Reference to source file
329 i32 ;; Unique ID to identify blocks from a template function
332 This descriptor provides debug information about nested blocks within a
333 subprogram. The line number and column numbers are used to dinstinguish two
334 lexical blocks at same depth.
339 i32, ;; Tag = 11 + LLVMDebugVersion (DW_TAG_lexical_block)
340 metadata ;; Reference to the scope we're annotating with a file change
341 metadata,;; Reference to the file the scope is enclosed in.
344 This descriptor provides a wrapper around a lexical scope to handle file
345 changes in the middle of a lexical block.
347 .. _format_basic_type:
349 Basic type descriptors
350 ^^^^^^^^^^^^^^^^^^^^^^
355 i32, ;; Tag = 36 + LLVMDebugVersion (DW_TAG_base_type)
356 metadata, ;; Reference to context
357 metadata, ;; Name (may be "" for anonymous types)
358 metadata, ;; Reference to file where defined (may be NULL)
359 i32, ;; Line number where defined (may be 0)
361 i64, ;; Alignment in bits
362 i64, ;; Offset in bits
364 i32 ;; DWARF type encoding
367 These descriptors define primitive types used in the code. Example ``int``,
368 ``bool`` and ``float``. The context provides the scope of the type, which is
369 usually the top level. Since basic types are not usually user defined the
370 context and line number can be left as NULL and 0. The size, alignment and
371 offset are expressed in bits and can be 64 bit values. The alignment is used
372 to round the offset when embedded in a :ref:`composite type
373 <format_composite_type>` (example to keep float doubles on 64 bit boundaries).
374 The offset is the bit offset if embedded in a :ref:`composite type
375 <format_composite_type>`.
377 The type encoding provides the details of the type. The values are typically
378 one of the following:
386 DW_ATE_signed_char = 6
388 DW_ATE_unsigned_char = 8
390 .. _format_derived_type:
392 Derived type descriptors
393 ^^^^^^^^^^^^^^^^^^^^^^^^
398 i32, ;; Tag (see below)
399 metadata, ;; Reference to context
400 metadata, ;; Name (may be "" for anonymous types)
401 metadata, ;; Reference to file where defined (may be NULL)
402 i32, ;; Line number where defined (may be 0)
404 i64, ;; Alignment in bits
405 i64, ;; Offset in bits
406 i32, ;; Flags to encode attributes, e.g. private
407 metadata, ;; Reference to type derived from
408 metadata, ;; (optional) Name of the Objective C property associated with
409 ;; Objective-C an ivar, or the type of which this
410 ;; pointer-to-member is pointing to members of.
411 metadata, ;; (optional) Name of the Objective C property getter selector.
412 metadata, ;; (optional) Name of the Objective C property setter selector.
413 i32 ;; (optional) Objective C property attributes.
416 These descriptors are used to define types derived from other types. The value
417 of the tag varies depending on the meaning. The following are possible tag
422 DW_TAG_formal_parameter = 5
424 DW_TAG_pointer_type = 15
425 DW_TAG_reference_type = 16
427 DW_TAG_ptr_to_member_type = 31
428 DW_TAG_const_type = 38
429 DW_TAG_volatile_type = 53
430 DW_TAG_restrict_type = 55
432 ``DW_TAG_member`` is used to define a member of a :ref:`composite type
433 <format_composite_type>` or :ref:`subprogram <format_subprograms>`. The type
434 of the member is the :ref:`derived type <format_derived_type>`.
435 ``DW_TAG_formal_parameter`` is used to define a member which is a formal
436 argument of a subprogram.
438 ``DW_TAG_typedef`` is used to provide a name for the derived type.
440 ``DW_TAG_pointer_type``, ``DW_TAG_reference_type``, ``DW_TAG_const_type``,
441 ``DW_TAG_volatile_type`` and ``DW_TAG_restrict_type`` are used to qualify the
442 :ref:`derived type <format_derived_type>`.
444 :ref:`Derived type <format_derived_type>` location can be determined from the
445 context and line number. The size, alignment and offset are expressed in bits
446 and can be 64 bit values. The alignment is used to round the offset when
447 embedded in a :ref:`composite type <format_composite_type>` (example to keep
448 float doubles on 64 bit boundaries.) The offset is the bit offset if embedded
449 in a :ref:`composite type <format_composite_type>`.
451 Note that the ``void *`` type is expressed as a type derived from NULL.
453 .. _format_composite_type:
455 Composite type descriptors
456 ^^^^^^^^^^^^^^^^^^^^^^^^^^
461 i32, ;; Tag (see below)
462 metadata, ;; Reference to context
463 metadata, ;; Name (may be "" for anonymous types)
464 metadata, ;; Reference to file where defined (may be NULL)
465 i32, ;; Line number where defined (may be 0)
467 i64, ;; Alignment in bits
468 i64, ;; Offset in bits
470 metadata, ;; Reference to type derived from
471 metadata, ;; Reference to array of member descriptors
472 i32 ;; Runtime languages
475 These descriptors are used to define types that are composed of 0 or more
476 elements. The value of the tag varies depending on the meaning. The following
477 are possible tag values:
481 DW_TAG_array_type = 1
482 DW_TAG_enumeration_type = 4
483 DW_TAG_structure_type = 19
484 DW_TAG_union_type = 23
485 DW_TAG_subroutine_type = 21
486 DW_TAG_inheritance = 28
488 The vector flag indicates that an array type is a native packed vector.
490 The members of array types (tag = ``DW_TAG_array_type``) are
491 :ref:`subrange descriptors <format_subrange>`, each
492 representing the range of subscripts at that level of indexing.
494 The members of enumeration types (tag = ``DW_TAG_enumeration_type``) are
495 :ref:`enumerator descriptors <format_enumerator>`, each representing the
496 definition of enumeration value for the set. All enumeration type descriptors
497 are collected inside the named metadata ``!llvm.dbg.cu``.
499 The members of structure (tag = ``DW_TAG_structure_type``) or union (tag =
500 ``DW_TAG_union_type``) types are any one of the :ref:`basic
501 <format_basic_type>`, :ref:`derived <format_derived_type>` or :ref:`composite
502 <format_composite_type>` type descriptors, each representing a field member of
503 the structure or union.
505 For C++ classes (tag = ``DW_TAG_structure_type``), member descriptors provide
506 information about base classes, static members and member functions. If a
507 member is a :ref:`derived type descriptor <format_derived_type>` and has a tag
508 of ``DW_TAG_inheritance``, then the type represents a base class. If the member
509 of is a :ref:`global variable descriptor <format_global_variables>` then it
510 represents a static member. And, if the member is a :ref:`subprogram
511 descriptor <format_subprograms>` then it represents a member function. For
512 static members and member functions, ``getName()`` returns the members link or
513 the C++ mangled name. ``getDisplayName()`` the simplied version of the name.
515 The first member of subroutine (tag = ``DW_TAG_subroutine_type``) type elements
516 is the return type for the subroutine. The remaining elements are the formal
517 arguments to the subroutine.
519 :ref:`Composite type <format_composite_type>` location can be determined from
520 the context and line number. The size, alignment and offset are expressed in
521 bits and can be 64 bit values. The alignment is used to round the offset when
522 embedded in a :ref:`composite type <format_composite_type>` (as an example, to
523 keep float doubles on 64 bit boundaries). The offset is the bit offset if
524 embedded in a :ref:`composite type <format_composite_type>`.
534 i32, ;; Tag = 33 + LLVMDebugVersion (DW_TAG_subrange_type)
539 These descriptors are used to define ranges of array subscripts for an array
540 :ref:`composite type <format_composite_type>`. The low value defines the lower
541 bounds typically zero for C/C++. The high value is the upper bounds. Values
542 are 64 bit. ``High - Low + 1`` is the size of the array. If ``Low > High``
543 the array bounds are not included in generated debugging information.
545 .. _format_enumerator:
547 Enumerator descriptors
548 ^^^^^^^^^^^^^^^^^^^^^^
553 i32, ;; Tag = 40 + LLVMDebugVersion (DW_TAG_enumerator)
558 These descriptors are used to define members of an enumeration :ref:`composite
559 type <format_composite_type>`, it associates the name to the value.
567 i32, ;; Tag (see below)
570 metadata, ;; Reference to file where defined
571 i32, ;; 24 bit - Line number where defined
572 ;; 8 bit - Argument number. 1 indicates 1st argument.
573 metadata, ;; Type descriptor
575 metadata ;; (optional) Reference to inline location
578 These descriptors are used to define variables local to a sub program. The
579 value of the tag depends on the usage of the variable:
583 DW_TAG_auto_variable = 256
584 DW_TAG_arg_variable = 257
586 An auto variable is any variable declared in the body of the function. An
587 argument variable is any variable that appears as a formal argument to the
590 The context is either the subprogram or block where the variable is defined.
591 Name the source variable name. Context and line indicate where the variable
592 was defined. Type descriptor defines the declared type of the variable.
594 .. _format_common_intrinsics:
596 Debugger intrinsic functions
597 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
599 LLVM uses several intrinsic functions (name prefixed with "``llvm.dbg``") to
600 provide debug information at various points in generated code.
607 void %llvm.dbg.declare(metadata, metadata)
609 This intrinsic provides information about a local element (e.g., variable).
610 The first argument is metadata holding the alloca for the variable. The second
611 argument is metadata containing a description of the variable.
618 void %llvm.dbg.value(metadata, i64, metadata)
620 This intrinsic provides information when a user source variable is set to a new
621 value. The first argument is the new value (wrapped as metadata). The second
622 argument is the offset in the user source variable where the new value is
623 written. The third argument is metadata containing a description of the user
626 Object lifetimes and scoping
627 ============================
629 In many languages, the local variables in functions can have their lifetimes or
630 scopes limited to a subset of a function. In the C family of languages, for
631 example, variables are only live (readable and writable) within the source
632 block that they are defined in. In functional languages, values are only
633 readable after they have been defined. Though this is a very obvious concept,
634 it is non-trivial to model in LLVM, because it has no notion of scoping in this
635 sense, and does not want to be tied to a language's scoping rules.
637 In order to handle this, the LLVM debug format uses the metadata attached to
638 llvm instructions to encode line number and scoping information. Consider the
639 following C fragment, for example:
653 Compiled to LLVM, this function would be represented like this:
657 define void @foo() nounwind ssp {
659 %X = alloca i32, align 4 ; <i32*> [#uses=4]
660 %Y = alloca i32, align 4 ; <i32*> [#uses=4]
661 %Z = alloca i32, align 4 ; <i32*> [#uses=3]
662 %0 = bitcast i32* %X to {}* ; <{}*> [#uses=1]
663 call void @llvm.dbg.declare(metadata !{i32 * %X}, metadata !0), !dbg !7
664 store i32 21, i32* %X, !dbg !8
665 %1 = bitcast i32* %Y to {}* ; <{}*> [#uses=1]
666 call void @llvm.dbg.declare(metadata !{i32 * %Y}, metadata !9), !dbg !10
667 store i32 22, i32* %Y, !dbg !11
668 %2 = bitcast i32* %Z to {}* ; <{}*> [#uses=1]
669 call void @llvm.dbg.declare(metadata !{i32 * %Z}, metadata !12), !dbg !14
670 store i32 23, i32* %Z, !dbg !15
671 %tmp = load i32* %X, !dbg !16 ; <i32> [#uses=1]
672 %tmp1 = load i32* %Y, !dbg !16 ; <i32> [#uses=1]
673 %add = add nsw i32 %tmp, %tmp1, !dbg !16 ; <i32> [#uses=1]
674 store i32 %add, i32* %Z, !dbg !16
675 %tmp2 = load i32* %Y, !dbg !17 ; <i32> [#uses=1]
676 store i32 %tmp2, i32* %X, !dbg !17
680 declare void @llvm.dbg.declare(metadata, metadata) nounwind readnone
682 !0 = metadata !{i32 459008, metadata !1, metadata !"X",
683 metadata !3, i32 2, metadata !6}; [ DW_TAG_auto_variable ]
684 !1 = metadata !{i32 458763, metadata !2}; [DW_TAG_lexical_block ]
685 !2 = metadata !{i32 458798, i32 0, metadata !3, metadata !"foo", metadata !"foo",
686 metadata !"foo", metadata !3, i32 1, metadata !4,
687 i1 false, i1 true}; [DW_TAG_subprogram ]
688 !3 = metadata !{i32 458769, i32 0, i32 12, metadata !"foo.c",
689 metadata !"/private/tmp", metadata !"clang 1.1", i1 true,
690 i1 false, metadata !"", i32 0}; [DW_TAG_compile_unit ]
691 !4 = metadata !{i32 458773, metadata !3, metadata !"", null, i32 0, i64 0, i64 0,
692 i64 0, i32 0, null, metadata !5, i32 0}; [DW_TAG_subroutine_type ]
693 !5 = metadata !{null}
694 !6 = metadata !{i32 458788, metadata !3, metadata !"int", metadata !3, i32 0,
695 i64 32, i64 32, i64 0, i32 0, i32 5}; [DW_TAG_base_type ]
696 !7 = metadata !{i32 2, i32 7, metadata !1, null}
697 !8 = metadata !{i32 2, i32 3, metadata !1, null}
698 !9 = metadata !{i32 459008, metadata !1, metadata !"Y", metadata !3, i32 3,
699 metadata !6}; [ DW_TAG_auto_variable ]
700 !10 = metadata !{i32 3, i32 7, metadata !1, null}
701 !11 = metadata !{i32 3, i32 3, metadata !1, null}
702 !12 = metadata !{i32 459008, metadata !13, metadata !"Z", metadata !3, i32 5,
703 metadata !6}; [ DW_TAG_auto_variable ]
704 !13 = metadata !{i32 458763, metadata !1}; [DW_TAG_lexical_block ]
705 !14 = metadata !{i32 5, i32 9, metadata !13, null}
706 !15 = metadata !{i32 5, i32 5, metadata !13, null}
707 !16 = metadata !{i32 6, i32 5, metadata !13, null}
708 !17 = metadata !{i32 8, i32 3, metadata !1, null}
709 !18 = metadata !{i32 9, i32 1, metadata !2, null}
711 This example illustrates a few important details about LLVM debugging
712 information. In particular, it shows how the ``llvm.dbg.declare`` intrinsic and
713 location information, which are attached to an instruction, are applied
714 together to allow a debugger to analyze the relationship between statements,
715 variable definitions, and the code used to implement the function.
719 call void @llvm.dbg.declare(metadata, metadata !0), !dbg !7
721 The first intrinsic ``%llvm.dbg.declare`` encodes debugging information for the
722 variable ``X``. The metadata ``!dbg !7`` attached to the intrinsic provides
723 scope information for the variable ``X``.
727 !7 = metadata !{i32 2, i32 7, metadata !1, null}
728 !1 = metadata !{i32 458763, metadata !2}; [DW_TAG_lexical_block ]
729 !2 = metadata !{i32 458798, i32 0, metadata !3, metadata !"foo",
730 metadata !"foo", metadata !"foo", metadata !3, i32 1,
731 metadata !4, i1 false, i1 true}; [DW_TAG_subprogram ]
733 Here ``!7`` is metadata providing location information. It has four fields:
734 line number, column number, scope, and original scope. The original scope
735 represents inline location if this instruction is inlined inside a caller, and
736 is null otherwise. In this example, scope is encoded by ``!1``. ``!1``
737 represents a lexical block inside the scope ``!2``, where ``!2`` is a
738 :ref:`subprogram descriptor <format_subprograms>`. This way the location
739 information attached to the intrinsics indicates that the variable ``X`` is
740 declared at line number 2 at a function level scope in function ``foo``.
742 Now lets take another example.
746 call void @llvm.dbg.declare(metadata, metadata !12), !dbg !14
748 The second intrinsic ``%llvm.dbg.declare`` encodes debugging information for
749 variable ``Z``. The metadata ``!dbg !14`` attached to the intrinsic provides
750 scope information for the variable ``Z``.
754 !13 = metadata !{i32 458763, metadata !1}; [DW_TAG_lexical_block ]
755 !14 = metadata !{i32 5, i32 9, metadata !13, null}
757 Here ``!14`` indicates that ``Z`` is declared at line number 5 and
758 column number 9 inside of lexical scope ``!13``. The lexical scope itself
759 resides inside of lexical scope ``!1`` described above.
761 The scope information attached with each instruction provides a straightforward
762 way to find instructions covered by a scope.
766 C/C++ front-end specific debug information
767 ==========================================
769 The C and C++ front-ends represent information about the program in a format
770 that is effectively identical to `DWARF 3.0
771 <http://www.eagercon.com/dwarf/dwarf3std.htm>`_ in terms of information
772 content. This allows code generators to trivially support native debuggers by
773 generating standard dwarf information, and contains enough information for
774 non-dwarf targets to translate it as needed.
776 This section describes the forms used to represent C and C++ programs. Other
777 languages could pattern themselves after this (which itself is tuned to
778 representing programs in the same way that DWARF 3 does), or they could choose
779 to provide completely different forms if they don't fit into the DWARF model.
780 As support for debugging information gets added to the various LLVM
781 source-language front-ends, the information used should be documented here.
783 The following sections provide examples of various C/C++ constructs and the
784 debug information that would best describe those constructs.
786 C/C++ source file information
787 -----------------------------
789 Given the source files ``MySource.cpp`` and ``MyHeader.h`` located in the
790 directory ``/Users/mine/sources``, the following code:
794 #include "MyHeader.h"
796 int main(int argc, char *argv[]) {
800 a C/C++ front-end would generate the following descriptors:
806 ;; Define the compile unit for the main source file "/Users/mine/sources/MySource.cpp".
811 i32 4, ;; Language Id
812 metadata !"MySource.cpp",
813 metadata !"/Users/mine/sources",
814 metadata !"4.2.1 (Based on Apple Inc. build 5649) (LLVM build 00)",
815 i1 true, ;; Main Compile Unit
816 i1 false, ;; Optimized compile unit
817 metadata !"", ;; Compiler flags
818 i32 0} ;; Runtime version
821 ;; Define the file for the file "/Users/mine/sources/MySource.cpp".
825 metadata !"MySource.cpp",
826 metadata !"/Users/mine/sources",
827 metadata !2 ;; Compile unit
831 ;; Define the file for the file "/Users/mine/sources/Myheader.h"
835 metadata !"Myheader.h"
836 metadata !"/Users/mine/sources",
837 metadata !2 ;; Compile unit
842 ``llvm::Instruction`` provides easy access to metadata attached with an
843 instruction. One can extract line number information encoded in LLVM IR using
844 ``Instruction::getMetadata()`` and ``DILocation::getLineNumber()``.
848 if (MDNode *N = I->getMetadata("dbg")) { // Here I is an LLVM instruction
849 DILocation Loc(N); // DILocation is in DebugInfo.h
850 unsigned Line = Loc.getLineNumber();
851 StringRef File = Loc.getFilename();
852 StringRef Dir = Loc.getDirectory();
855 C/C++ global variable information
856 ---------------------------------
858 Given an integer global variable declared as follows:
864 a C/C++ front-end would generate the following descriptors:
869 ;; Define the global itself.
871 %MyGlobal = global int 100
874 ;; List of debug info of globals
878 ;; Define the compile unit.
883 metadata !"foo.cpp", ;; File
884 metadata !"/Volumes/Data/tmp", ;; Directory
885 metadata !"clang version 3.1 ", ;; Producer
886 i1 true, ;; Deprecated field
887 i1 false, ;; "isOptimized"?
888 metadata !"", ;; Flags
889 i32 0, ;; Runtime Version
890 metadata !1, ;; Enum Types
891 metadata !1, ;; Retained Types
892 metadata !1, ;; Subprograms
893 metadata !3 ;; Global Variables
894 } ; [ DW_TAG_compile_unit ]
896 ;; The Array of Global Variables
906 ;; Define the global variable itself.
912 metadata !"MyGlobal", ;; Name
913 metadata !"MyGlobal", ;; Display Name
914 metadata !"", ;; Linkage Name
918 i32 0, ;; IsLocalToUnit
919 i32 1, ;; IsDefinition
920 i32* @MyGlobal ;; LLVM-IR Value
921 } ; [ DW_TAG_variable ]
928 metadata !"foo.cpp", ;; File
929 metadata !"/Volumes/Data/tmp", ;; Directory
931 } ; [ DW_TAG_file_type ]
939 metadata !"int", ;; Name
942 i64 32, ;; Size in Bits
943 i64 32, ;; Align in Bits
947 } ; [ DW_TAG_base_type ]
949 C/C++ function information
950 --------------------------
952 Given a function declared as follows:
956 int main(int argc, char *argv[]) {
960 a C/C++ front-end would generate the following descriptors:
965 ;; Define the anchor for subprograms. Note that the second field of the
966 ;; anchor is 46, which is the same as the tag for subprograms
967 ;; (46 = DW_TAG_subprogram.)
972 metadata !1, ;; Context
973 metadata !"main", ;; Name
974 metadata !"main", ;; Display name
975 metadata !"main", ;; Linkage name
977 i32 1, ;; Line number
979 i1 false, ;; Is local
980 i1 true, ;; Is definition
981 i32 0, ;; Virtuality attribute, e.g. pure virtual function
982 i32 0, ;; Index into virtual table for C++ methods
983 i32 0, ;; Type that holds virtual table.
985 i1 false, ;; True if this function is optimized
986 Function *, ;; Pointer to llvm::Function
987 null ;; Function template parameters
990 ;; Define the subprogram itself.
992 define i32 @main(i32 %argc, i8** %argv) {
999 The following are the basic type descriptors for C/C++ core types:
1004 .. code-block:: llvm
1008 metadata !1, ;; Context
1009 metadata !"bool", ;; Name
1010 metadata !1, ;; File
1011 i32 0, ;; Line number
1012 i64 8, ;; Size in Bits
1013 i64 8, ;; Align in Bits
1014 i64 0, ;; Offset in Bits
1022 .. code-block:: llvm
1026 metadata !1, ;; Context
1027 metadata !"char", ;; Name
1028 metadata !1, ;; File
1029 i32 0, ;; Line number
1030 i64 8, ;; Size in Bits
1031 i64 8, ;; Align in Bits
1032 i64 0, ;; Offset in Bits
1040 .. code-block:: llvm
1044 metadata !1, ;; Context
1045 metadata !"unsigned char",
1046 metadata !1, ;; File
1047 i32 0, ;; Line number
1048 i64 8, ;; Size in Bits
1049 i64 8, ;; Align in Bits
1050 i64 0, ;; Offset in Bits
1058 .. code-block:: llvm
1062 metadata !1, ;; Context
1063 metadata !"short int",
1064 metadata !1, ;; File
1065 i32 0, ;; Line number
1066 i64 16, ;; Size in Bits
1067 i64 16, ;; Align in Bits
1068 i64 0, ;; Offset in Bits
1076 .. code-block:: llvm
1080 metadata !1, ;; Context
1081 metadata !"short unsigned int",
1082 metadata !1, ;; File
1083 i32 0, ;; Line number
1084 i64 16, ;; Size in Bits
1085 i64 16, ;; Align in Bits
1086 i64 0, ;; Offset in Bits
1094 .. code-block:: llvm
1098 metadata !1, ;; Context
1099 metadata !"int", ;; Name
1100 metadata !1, ;; File
1101 i32 0, ;; Line number
1102 i64 32, ;; Size in Bits
1103 i64 32, ;; Align in Bits
1104 i64 0, ;; Offset in Bits
1112 .. code-block:: llvm
1116 metadata !1, ;; Context
1117 metadata !"unsigned int",
1118 metadata !1, ;; File
1119 i32 0, ;; Line number
1120 i64 32, ;; Size in Bits
1121 i64 32, ;; Align in Bits
1122 i64 0, ;; Offset in Bits
1130 .. code-block:: llvm
1134 metadata !1, ;; Context
1135 metadata !"long long int",
1136 metadata !1, ;; File
1137 i32 0, ;; Line number
1138 i64 64, ;; Size in Bits
1139 i64 64, ;; Align in Bits
1140 i64 0, ;; Offset in Bits
1148 .. code-block:: llvm
1152 metadata !1, ;; Context
1153 metadata !"long long unsigned int",
1154 metadata !1, ;; File
1155 i32 0, ;; Line number
1156 i64 64, ;; Size in Bits
1157 i64 64, ;; Align in Bits
1158 i64 0, ;; Offset in Bits
1166 .. code-block:: llvm
1170 metadata !1, ;; Context
1172 metadata !1, ;; File
1173 i32 0, ;; Line number
1174 i64 32, ;; Size in Bits
1175 i64 32, ;; Align in Bits
1176 i64 0, ;; Offset in Bits
1184 .. code-block:: llvm
1188 metadata !1, ;; Context
1189 metadata !"double",;; Name
1190 metadata !1, ;; File
1191 i32 0, ;; Line number
1192 i64 64, ;; Size in Bits
1193 i64 64, ;; Align in Bits
1194 i64 0, ;; Offset in Bits
1202 Given the following as an example of C/C++ derived type:
1206 typedef const int *IntPtr;
1208 a C/C++ front-end would generate the following descriptors:
1210 .. code-block:: llvm
1213 ;; Define the typedef "IntPtr".
1217 metadata !1, ;; Context
1218 metadata !"IntPtr", ;; Name
1219 metadata !3, ;; File
1220 i32 0, ;; Line number
1221 i64 0, ;; Size in bits
1222 i64 0, ;; Align in bits
1223 i64 0, ;; Offset in bits
1225 metadata !4 ;; Derived From type
1228 ;; Define the pointer type.
1232 metadata !1, ;; Context
1233 metadata !"", ;; Name
1234 metadata !1, ;; File
1235 i32 0, ;; Line number
1236 i64 64, ;; Size in bits
1237 i64 64, ;; Align in bits
1238 i64 0, ;; Offset in bits
1240 metadata !5 ;; Derived From type
1243 ;; Define the const type.
1247 metadata !1, ;; Context
1248 metadata !"", ;; Name
1249 metadata !1, ;; File
1250 i32 0, ;; Line number
1251 i64 32, ;; Size in bits
1252 i64 32, ;; Align in bits
1253 i64 0, ;; Offset in bits
1255 metadata !6 ;; Derived From type
1258 ;; Define the int type.
1262 metadata !1, ;; Context
1263 metadata !"int", ;; Name
1264 metadata !1, ;; File
1265 i32 0, ;; Line number
1266 i64 32, ;; Size in bits
1267 i64 32, ;; Align in bits
1268 i64 0, ;; Offset in bits
1273 C/C++ struct/union types
1274 ------------------------
1276 Given the following as an example of C/C++ struct type:
1286 a C/C++ front-end would generate the following descriptors:
1288 .. code-block:: llvm
1291 ;; Define basic type for unsigned int.
1295 metadata !1, ;; Context
1296 metadata !"unsigned int",
1297 metadata !1, ;; File
1298 i32 0, ;; Line number
1299 i64 32, ;; Size in Bits
1300 i64 32, ;; Align in Bits
1301 i64 0, ;; Offset in Bits
1306 ;; Define composite type for struct Color.
1310 metadata !1, ;; Context
1311 metadata !"Color", ;; Name
1312 metadata !1, ;; Compile unit
1313 i32 1, ;; Line number
1314 i64 96, ;; Size in bits
1315 i64 32, ;; Align in bits
1316 i64 0, ;; Offset in bits
1318 null, ;; Derived From
1319 metadata !3, ;; Elements
1320 i32 0 ;; Runtime Language
1324 ;; Define the Red field.
1328 metadata !1, ;; Context
1329 metadata !"Red", ;; Name
1330 metadata !1, ;; File
1331 i32 2, ;; Line number
1332 i64 32, ;; Size in bits
1333 i64 32, ;; Align in bits
1334 i64 0, ;; Offset in bits
1336 metadata !5 ;; Derived From type
1340 ;; Define the Green field.
1344 metadata !1, ;; Context
1345 metadata !"Green", ;; Name
1346 metadata !1, ;; File
1347 i32 3, ;; Line number
1348 i64 32, ;; Size in bits
1349 i64 32, ;; Align in bits
1350 i64 32, ;; Offset in bits
1352 metadata !5 ;; Derived From type
1356 ;; Define the Blue field.
1360 metadata !1, ;; Context
1361 metadata !"Blue", ;; Name
1362 metadata !1, ;; File
1363 i32 4, ;; Line number
1364 i64 32, ;; Size in bits
1365 i64 32, ;; Align in bits
1366 i64 64, ;; Offset in bits
1368 metadata !5 ;; Derived From type
1372 ;; Define the array of fields used by the composite type Color.
1374 !3 = metadata !{metadata !4, metadata !6, metadata !7}
1376 C/C++ enumeration types
1377 -----------------------
1379 Given the following as an example of C/C++ enumeration type:
1389 a C/C++ front-end would generate the following descriptors:
1391 .. code-block:: llvm
1394 ;; Define composite type for enum Trees
1398 metadata !1, ;; Context
1399 metadata !"Trees", ;; Name
1400 metadata !1, ;; File
1401 i32 1, ;; Line number
1402 i64 32, ;; Size in bits
1403 i64 32, ;; Align in bits
1404 i64 0, ;; Offset in bits
1406 null, ;; Derived From type
1407 metadata !3, ;; Elements
1408 i32 0 ;; Runtime language
1412 ;; Define the array of enumerators used by composite type Trees.
1414 !3 = metadata !{metadata !4, metadata !5, metadata !6}
1417 ;; Define Spruce enumerator.
1419 !4 = metadata !{i32 524328, metadata !"Spruce", i64 100}
1422 ;; Define Oak enumerator.
1424 !5 = metadata !{i32 524328, metadata !"Oak", i64 200}
1427 ;; Define Maple enumerator.
1429 !6 = metadata !{i32 524328, metadata !"Maple", i64 300}
1431 Debugging information format
1432 ============================
1434 Debugging Information Extension for Objective C Properties
1435 ----------------------------------------------------------
1440 Objective C provides a simpler way to declare and define accessor methods using
1441 declared properties. The language provides features to declare a property and
1442 to let compiler synthesize accessor methods.
1444 The debugger lets developer inspect Objective C interfaces and their instance
1445 variables and class variables. However, the debugger does not know anything
1446 about the properties defined in Objective C interfaces. The debugger consumes
1447 information generated by compiler in DWARF format. The format does not support
1448 encoding of Objective C properties. This proposal describes DWARF extensions to
1449 encode Objective C properties, which the debugger can use to let developers
1450 inspect Objective C properties.
1455 Objective C properties exist separately from class members. A property can be
1456 defined only by "setter" and "getter" selectors, and be calculated anew on each
1457 access. Or a property can just be a direct access to some declared ivar.
1458 Finally it can have an ivar "automatically synthesized" for it by the compiler,
1459 in which case the property can be referred to in user code directly using the
1460 standard C dereference syntax as well as through the property "dot" syntax, but
1461 there is no entry in the ``@interface`` declaration corresponding to this ivar.
1463 To facilitate debugging, these properties we will add a new DWARF TAG into the
1464 ``DW_TAG_structure_type`` definition for the class to hold the description of a
1465 given property, and a set of DWARF attributes that provide said description.
1466 The property tag will also contain the name and declared type of the property.
1468 If there is a related ivar, there will also be a DWARF property attribute placed
1469 in the ``DW_TAG_member`` DIE for that ivar referring back to the property TAG
1470 for that property. And in the case where the compiler synthesizes the ivar
1471 directly, the compiler is expected to generate a ``DW_TAG_member`` for that
1472 ivar (with the ``DW_AT_artificial`` set to 1), whose name will be the name used
1473 to access this ivar directly in code, and with the property attribute pointing
1474 back to the property it is backing.
1476 The following examples will serve as illustration for our discussion:
1478 .. code-block:: objc
1490 @synthesize p2 = n2;
1493 This produces the following DWARF (this is a "pseudo dwarfdump" output):
1495 .. code-block:: none
1497 0x00000100: TAG_structure_type [7] *
1498 AT_APPLE_runtime_class( 0x10 )
1500 AT_decl_file( "Objc_Property.m" )
1503 0x00000110 TAG_APPLE_property
1505 AT_type ( {0x00000150} ( int ) )
1507 0x00000120: TAG_APPLE_property
1509 AT_type ( {0x00000150} ( int ) )
1511 0x00000130: TAG_member [8]
1513 AT_APPLE_property ( {0x00000110} "p1" )
1514 AT_type( {0x00000150} ( int ) )
1515 AT_artificial ( 0x1 )
1517 0x00000140: TAG_member [8]
1519 AT_APPLE_property ( {0x00000120} "p2" )
1520 AT_type( {0x00000150} ( int ) )
1522 0x00000150: AT_type( ( int ) )
1524 Note, the current convention is that the name of the ivar for an
1525 auto-synthesized property is the name of the property from which it derives
1526 with an underscore prepended, as is shown in the example. But we actually
1527 don't need to know this convention, since we are given the name of the ivar
1530 Also, it is common practice in ObjC to have different property declarations in
1531 the @interface and @implementation - e.g. to provide a read-only property in
1532 the interface,and a read-write interface in the implementation. In that case,
1533 the compiler should emit whichever property declaration will be in force in the
1534 current translation unit.
1536 Developers can decorate a property with attributes which are encoded using
1537 ``DW_AT_APPLE_property_attribute``.
1539 .. code-block:: objc
1541 @property (readonly, nonatomic) int pr;
1543 .. code-block:: none
1545 TAG_APPLE_property [8]
1547 AT_type ( {0x00000147} (int) )
1548 AT_APPLE_property_attribute (DW_APPLE_PROPERTY_readonly, DW_APPLE_PROPERTY_nonatomic)
1550 The setter and getter method names are attached to the property using
1551 ``DW_AT_APPLE_property_setter`` and ``DW_AT_APPLE_property_getter`` attributes.
1553 .. code-block:: objc
1556 @property (setter=myOwnP3Setter:) int p3;
1557 -(void)myOwnP3Setter:(int)a;
1562 -(void)myOwnP3Setter:(int)a{ }
1565 The DWARF for this would be:
1567 .. code-block:: none
1569 0x000003bd: TAG_structure_type [7] *
1570 AT_APPLE_runtime_class( 0x10 )
1572 AT_decl_file( "Objc_Property.m" )
1575 0x000003cd TAG_APPLE_property
1577 AT_APPLE_property_setter ( "myOwnP3Setter:" )
1578 AT_type( {0x00000147} ( int ) )
1580 0x000003f3: TAG_member [8]
1582 AT_type ( {0x00000147} ( int ) )
1583 AT_APPLE_property ( {0x000003cd} )
1584 AT_artificial ( 0x1 )
1589 +-----------------------+--------+
1591 +=======================+========+
1592 | DW_TAG_APPLE_property | 0x4200 |
1593 +-----------------------+--------+
1595 New DWARF Attributes
1596 ^^^^^^^^^^^^^^^^^^^^
1598 +--------------------------------+--------+-----------+
1599 | Attribute | Value | Classes |
1600 +================================+========+===========+
1601 | DW_AT_APPLE_property | 0x3fed | Reference |
1602 +--------------------------------+--------+-----------+
1603 | DW_AT_APPLE_property_getter | 0x3fe9 | String |
1604 +--------------------------------+--------+-----------+
1605 | DW_AT_APPLE_property_setter | 0x3fea | String |
1606 +--------------------------------+--------+-----------+
1607 | DW_AT_APPLE_property_attribute | 0x3feb | Constant |
1608 +--------------------------------+--------+-----------+
1613 +--------------------------------+-------+
1615 +================================+=======+
1616 | DW_AT_APPLE_PROPERTY_readonly | 0x1 |
1617 +--------------------------------+-------+
1618 | DW_AT_APPLE_PROPERTY_readwrite | 0x2 |
1619 +--------------------------------+-------+
1620 | DW_AT_APPLE_PROPERTY_assign | 0x4 |
1621 +--------------------------------+-------+
1622 | DW_AT_APPLE_PROPERTY_retain | 0x8 |
1623 +--------------------------------+-------+
1624 | DW_AT_APPLE_PROPERTY_copy | 0x10 |
1625 +--------------------------------+-------+
1626 | DW_AT_APPLE_PROPERTY_nonatomic | 0x20 |
1627 +--------------------------------+-------+
1629 Name Accelerator Tables
1630 -----------------------
1635 The "``.debug_pubnames``" and "``.debug_pubtypes``" formats are not what a
1636 debugger needs. The "``pub``" in the section name indicates that the entries
1637 in the table are publicly visible names only. This means no static or hidden
1638 functions show up in the "``.debug_pubnames``". No static variables or private
1639 class variables are in the "``.debug_pubtypes``". Many compilers add different
1640 things to these tables, so we can't rely upon the contents between gcc, icc, or
1643 The typical query given by users tends not to match up with the contents of
1644 these tables. For example, the DWARF spec states that "In the case of the name
1645 of a function member or static data member of a C++ structure, class or union,
1646 the name presented in the "``.debug_pubnames``" section is not the simple name
1647 given by the ``DW_AT_name attribute`` of the referenced debugging information
1648 entry, but rather the fully qualified name of the data or function member."
1649 So the only names in these tables for complex C++ entries is a fully
1650 qualified name. Debugger users tend not to enter their search strings as
1651 "``a::b::c(int,const Foo&) const``", but rather as "``c``", "``b::c``" , or
1652 "``a::b::c``". So the name entered in the name table must be demangled in
1653 order to chop it up appropriately and additional names must be manually entered
1654 into the table to make it effective as a name lookup table for debuggers to
1657 All debuggers currently ignore the "``.debug_pubnames``" table as a result of
1658 its inconsistent and useless public-only name content making it a waste of
1659 space in the object file. These tables, when they are written to disk, are not
1660 sorted in any way, leaving every debugger to do its own parsing and sorting.
1661 These tables also include an inlined copy of the string values in the table
1662 itself making the tables much larger than they need to be on disk, especially
1663 for large C++ programs.
1665 Can't we just fix the sections by adding all of the names we need to this
1666 table? No, because that is not what the tables are defined to contain and we
1667 won't know the difference between the old bad tables and the new good tables.
1668 At best we could make our own renamed sections that contain all of the data we
1671 These tables are also insufficient for what a debugger like LLDB needs. LLDB
1672 uses clang for its expression parsing where LLDB acts as a PCH. LLDB is then
1673 often asked to look for type "``foo``" or namespace "``bar``", or list items in
1674 namespace "``baz``". Namespaces are not included in the pubnames or pubtypes
1675 tables. Since clang asks a lot of questions when it is parsing an expression,
1676 we need to be very fast when looking up names, as it happens a lot. Having new
1677 accelerator tables that are optimized for very quick lookups will benefit this
1678 type of debugging experience greatly.
1680 We would like to generate name lookup tables that can be mapped into memory
1681 from disk, and used as is, with little or no up-front parsing. We would also
1682 be able to control the exact content of these different tables so they contain
1683 exactly what we need. The Name Accelerator Tables were designed to fix these
1684 issues. In order to solve these issues we need to:
1686 * Have a format that can be mapped into memory from disk and used as is
1687 * Lookups should be very fast
1688 * Extensible table format so these tables can be made by many producers
1689 * Contain all of the names needed for typical lookups out of the box
1690 * Strict rules for the contents of tables
1692 Table size is important and the accelerator table format should allow the reuse
1693 of strings from common string tables so the strings for the names are not
1694 duplicated. We also want to make sure the table is ready to be used as-is by
1695 simply mapping the table into memory with minimal header parsing.
1697 The name lookups need to be fast and optimized for the kinds of lookups that
1698 debuggers tend to do. Optimally we would like to touch as few parts of the
1699 mapped table as possible when doing a name lookup and be able to quickly find
1700 the name entry we are looking for, or discover there are no matches. In the
1701 case of debuggers we optimized for lookups that fail most of the time.
1703 Each table that is defined should have strict rules on exactly what is in the
1704 accelerator tables and documented so clients can rely on the content.
1709 Standard Hash Tables
1710 """"""""""""""""""""
1712 Typical hash tables have a header, buckets, and each bucket points to the
1715 .. code-block:: none
1725 The BUCKETS are an array of offsets to DATA for each hash:
1727 .. code-block:: none
1730 | 0x00001000 | BUCKETS[0]
1731 | 0x00002000 | BUCKETS[1]
1732 | 0x00002200 | BUCKETS[2]
1733 | 0x000034f0 | BUCKETS[3]
1735 | 0xXXXXXXXX | BUCKETS[n_buckets]
1738 So for ``bucket[3]`` in the example above, we have an offset into the table
1739 0x000034f0 which points to a chain of entries for the bucket. Each bucket must
1740 contain a next pointer, full 32 bit hash value, the string itself, and the data
1741 for the current string value.
1743 .. code-block:: none
1746 0x000034f0: | 0x00003500 | next pointer
1747 | 0x12345678 | 32 bit hash
1748 | "erase" | string value
1749 | data[n] | HashData for this bucket
1751 0x00003500: | 0x00003550 | next pointer
1752 | 0x29273623 | 32 bit hash
1753 | "dump" | string value
1754 | data[n] | HashData for this bucket
1756 0x00003550: | 0x00000000 | next pointer
1757 | 0x82638293 | 32 bit hash
1758 | "main" | string value
1759 | data[n] | HashData for this bucket
1762 The problem with this layout for debuggers is that we need to optimize for the
1763 negative lookup case where the symbol we're searching for is not present. So
1764 if we were to lookup "``printf``" in the table above, we would make a 32 hash
1765 for "``printf``", it might match ``bucket[3]``. We would need to go to the
1766 offset 0x000034f0 and start looking to see if our 32 bit hash matches. To do
1767 so, we need to read the next pointer, then read the hash, compare it, and skip
1768 to the next bucket. Each time we are skipping many bytes in memory and
1769 touching new cache pages just to do the compare on the full 32 bit hash. All
1770 of these accesses then tell us that we didn't have a match.
1775 To solve the issues mentioned above we have structured the hash tables a bit
1776 differently: a header, buckets, an array of all unique 32 bit hash values,
1777 followed by an array of hash value data offsets, one for each hash value, then
1778 the data for all hash values:
1780 .. code-block:: none
1794 The ``BUCKETS`` in the name tables are an index into the ``HASHES`` array. By
1795 making all of the full 32 bit hash values contiguous in memory, we allow
1796 ourselves to efficiently check for a match while touching as little memory as
1797 possible. Most often checking the 32 bit hash values is as far as the lookup
1798 goes. If it does match, it usually is a match with no collisions. So for a
1799 table with "``n_buckets``" buckets, and "``n_hashes``" unique 32 bit hash
1800 values, we can clarify the contents of the ``BUCKETS``, ``HASHES`` and
1803 .. code-block:: none
1805 .-------------------------.
1806 | HEADER.magic | uint32_t
1807 | HEADER.version | uint16_t
1808 | HEADER.hash_function | uint16_t
1809 | HEADER.bucket_count | uint32_t
1810 | HEADER.hashes_count | uint32_t
1811 | HEADER.header_data_len | uint32_t
1812 | HEADER_DATA | HeaderData
1813 |-------------------------|
1814 | BUCKETS | uint32_t[bucket_count] // 32 bit hash indexes
1815 |-------------------------|
1816 | HASHES | uint32_t[hashes_count] // 32 bit hash values
1817 |-------------------------|
1818 | OFFSETS | uint32_t[hashes_count] // 32 bit offsets to hash value data
1819 |-------------------------|
1821 `-------------------------'
1823 So taking the exact same data from the standard hash example above we end up
1826 .. code-block:: none
1836 | ... | BUCKETS[n_buckets]
1838 | 0x........ | HASHES[0]
1839 | 0x........ | HASHES[1]
1840 | 0x........ | HASHES[2]
1841 | 0x........ | HASHES[3]
1842 | 0x........ | HASHES[4]
1843 | 0x........ | HASHES[5]
1844 | 0x12345678 | HASHES[6] hash for BUCKETS[3]
1845 | 0x29273623 | HASHES[7] hash for BUCKETS[3]
1846 | 0x82638293 | HASHES[8] hash for BUCKETS[3]
1847 | 0x........ | HASHES[9]
1848 | 0x........ | HASHES[10]
1849 | 0x........ | HASHES[11]
1850 | 0x........ | HASHES[12]
1851 | 0x........ | HASHES[13]
1852 | 0x........ | HASHES[n_hashes]
1854 | 0x........ | OFFSETS[0]
1855 | 0x........ | OFFSETS[1]
1856 | 0x........ | OFFSETS[2]
1857 | 0x........ | OFFSETS[3]
1858 | 0x........ | OFFSETS[4]
1859 | 0x........ | OFFSETS[5]
1860 | 0x000034f0 | OFFSETS[6] offset for BUCKETS[3]
1861 | 0x00003500 | OFFSETS[7] offset for BUCKETS[3]
1862 | 0x00003550 | OFFSETS[8] offset for BUCKETS[3]
1863 | 0x........ | OFFSETS[9]
1864 | 0x........ | OFFSETS[10]
1865 | 0x........ | OFFSETS[11]
1866 | 0x........ | OFFSETS[12]
1867 | 0x........ | OFFSETS[13]
1868 | 0x........ | OFFSETS[n_hashes]
1876 0x000034f0: | 0x00001203 | .debug_str ("erase")
1877 | 0x00000004 | A 32 bit array count - number of HashData with name "erase"
1878 | 0x........ | HashData[0]
1879 | 0x........ | HashData[1]
1880 | 0x........ | HashData[2]
1881 | 0x........ | HashData[3]
1882 | 0x00000000 | String offset into .debug_str (terminate data for hash)
1884 0x00003500: | 0x00001203 | String offset into .debug_str ("collision")
1885 | 0x00000002 | A 32 bit array count - number of HashData with name "collision"
1886 | 0x........ | HashData[0]
1887 | 0x........ | HashData[1]
1888 | 0x00001203 | String offset into .debug_str ("dump")
1889 | 0x00000003 | A 32 bit array count - number of HashData with name "dump"
1890 | 0x........ | HashData[0]
1891 | 0x........ | HashData[1]
1892 | 0x........ | HashData[2]
1893 | 0x00000000 | String offset into .debug_str (terminate data for hash)
1895 0x00003550: | 0x00001203 | String offset into .debug_str ("main")
1896 | 0x00000009 | A 32 bit array count - number of HashData with name "main"
1897 | 0x........ | HashData[0]
1898 | 0x........ | HashData[1]
1899 | 0x........ | HashData[2]
1900 | 0x........ | HashData[3]
1901 | 0x........ | HashData[4]
1902 | 0x........ | HashData[5]
1903 | 0x........ | HashData[6]
1904 | 0x........ | HashData[7]
1905 | 0x........ | HashData[8]
1906 | 0x00000000 | String offset into .debug_str (terminate data for hash)
1909 So we still have all of the same data, we just organize it more efficiently for
1910 debugger lookup. If we repeat the same "``printf``" lookup from above, we
1911 would hash "``printf``" and find it matches ``BUCKETS[3]`` by taking the 32 bit
1912 hash value and modulo it by ``n_buckets``. ``BUCKETS[3]`` contains "6" which
1913 is the index into the ``HASHES`` table. We would then compare any consecutive
1914 32 bit hashes values in the ``HASHES`` array as long as the hashes would be in
1915 ``BUCKETS[3]``. We do this by verifying that each subsequent hash value modulo
1916 ``n_buckets`` is still 3. In the case of a failed lookup we would access the
1917 memory for ``BUCKETS[3]``, and then compare a few consecutive 32 bit hashes
1918 before we know that we have no match. We don't end up marching through
1919 multiple words of memory and we really keep the number of processor data cache
1920 lines being accessed as small as possible.
1922 The string hash that is used for these lookup tables is the Daniel J.
1923 Bernstein hash which is also used in the ELF ``GNU_HASH`` sections. It is a
1924 very good hash for all kinds of names in programs with very few hash
1927 Empty buckets are designated by using an invalid hash index of ``UINT32_MAX``.
1932 These name hash tables are designed to be generic where specializations of the
1933 table get to define additional data that goes into the header ("``HeaderData``"),
1934 how the string value is stored ("``KeyType``") and the content of the data for each
1940 The header has a fixed part, and the specialized part. The exact format of the
1947 uint32_t magic; // 'HASH' magic value to allow endian detection
1948 uint16_t version; // Version number
1949 uint16_t hash_function; // The hash function enumeration that was used
1950 uint32_t bucket_count; // The number of buckets in this hash table
1951 uint32_t hashes_count; // The total number of unique hash values and hash data offsets in this table
1952 uint32_t header_data_len; // The bytes to skip to get to the hash indexes (buckets) for correct alignment
1953 // Specifically the length of the following HeaderData field - this does not
1954 // include the size of the preceding fields
1955 HeaderData header_data; // Implementation specific header data
1958 The header starts with a 32 bit "``magic``" value which must be ``'HASH'``
1959 encoded as an ASCII integer. This allows the detection of the start of the
1960 hash table and also allows the table's byte order to be determined so the table
1961 can be correctly extracted. The "``magic``" value is followed by a 16 bit
1962 ``version`` number which allows the table to be revised and modified in the
1963 future. The current version number is 1. ``hash_function`` is a ``uint16_t``
1964 enumeration that specifies which hash function was used to produce this table.
1965 The current values for the hash function enumerations include:
1969 enum HashFunctionType
1971 eHashFunctionDJB = 0u, // Daniel J Bernstein hash function
1974 ``bucket_count`` is a 32 bit unsigned integer that represents how many buckets
1975 are in the ``BUCKETS`` array. ``hashes_count`` is the number of unique 32 bit
1976 hash values that are in the ``HASHES`` array, and is the same number of offsets
1977 are contained in the ``OFFSETS`` array. ``header_data_len`` specifies the size
1978 in bytes of the ``HeaderData`` that is filled in by specialized versions of
1984 The header is followed by the buckets, hashes, offsets, and hash value data.
1990 uint32_t buckets[Header.bucket_count]; // An array of hash indexes into the "hashes[]" array below
1991 uint32_t hashes [Header.hashes_count]; // Every unique 32 bit hash for the entire table is in this table
1992 uint32_t offsets[Header.hashes_count]; // An offset that corresponds to each item in the "hashes[]" array above
1995 ``buckets`` is an array of 32 bit indexes into the ``hashes`` array. The
1996 ``hashes`` array contains all of the 32 bit hash values for all names in the
1997 hash table. Each hash in the ``hashes`` table has an offset in the ``offsets``
1998 array that points to the data for the hash value.
2000 This table setup makes it very easy to repurpose these tables to contain
2001 different data, while keeping the lookup mechanism the same for all tables.
2002 This layout also makes it possible to save the table to disk and map it in
2003 later and do very efficient name lookups with little or no parsing.
2005 DWARF lookup tables can be implemented in a variety of ways and can store a lot
2006 of information for each name. We want to make the DWARF tables extensible and
2007 able to store the data efficiently so we have used some of the DWARF features
2008 that enable efficient data storage to define exactly what kind of data we store
2011 The ``HeaderData`` contains a definition of the contents of each HashData chunk.
2012 We might want to store an offset to all of the debug information entries (DIEs)
2013 for each name. To keep things extensible, we create a list of items, or
2014 Atoms, that are contained in the data for each name. First comes the type of
2015 the data in each atom:
2022 eAtomTypeDIEOffset = 1u, // DIE offset, check form for encoding
2023 eAtomTypeCUOffset = 2u, // DIE offset of the compiler unit header that contains the item in question
2024 eAtomTypeTag = 3u, // DW_TAG_xxx value, should be encoded as DW_FORM_data1 (if no tags exceed 255) or DW_FORM_data2
2025 eAtomTypeNameFlags = 4u, // Flags from enum NameFlags
2026 eAtomTypeTypeFlags = 5u, // Flags from enum TypeFlags
2029 The enumeration values and their meanings are:
2031 .. code-block:: none
2033 eAtomTypeNULL - a termination atom that specifies the end of the atom list
2034 eAtomTypeDIEOffset - an offset into the .debug_info section for the DWARF DIE for this name
2035 eAtomTypeCUOffset - an offset into the .debug_info section for the CU that contains the DIE
2036 eAtomTypeDIETag - The DW_TAG_XXX enumeration value so you don't have to parse the DWARF to see what it is
2037 eAtomTypeNameFlags - Flags for functions and global variables (isFunction, isInlined, isExternal...)
2038 eAtomTypeTypeFlags - Flags for types (isCXXClass, isObjCClass, ...)
2040 Then we allow each atom type to define the atom type and how the data for each
2041 atom type data is encoded:
2047 uint16_t type; // AtomType enum value
2048 uint16_t form; // DWARF DW_FORM_XXX defines
2051 The ``form`` type above is from the DWARF specification and defines the exact
2052 encoding of the data for the Atom type. See the DWARF specification for the
2053 ``DW_FORM_`` definitions.
2059 uint32_t die_offset_base;
2060 uint32_t atom_count;
2061 Atoms atoms[atom_count0];
2064 ``HeaderData`` defines the base DIE offset that should be added to any atoms
2065 that are encoded using the ``DW_FORM_ref1``, ``DW_FORM_ref2``,
2066 ``DW_FORM_ref4``, ``DW_FORM_ref8`` or ``DW_FORM_ref_udata``. It also defines
2067 what is contained in each ``HashData`` object -- ``Atom.form`` tells us how large
2068 each field will be in the ``HashData`` and the ``Atom.type`` tells us how this data
2069 should be interpreted.
2071 For the current implementations of the "``.apple_names``" (all functions +
2072 globals), the "``.apple_types``" (names of all types that are defined), and
2073 the "``.apple_namespaces``" (all namespaces), we currently set the ``Atom``
2078 HeaderData.atom_count = 1;
2079 HeaderData.atoms[0].type = eAtomTypeDIEOffset;
2080 HeaderData.atoms[0].form = DW_FORM_data4;
2082 This defines the contents to be the DIE offset (eAtomTypeDIEOffset) that is
2083 encoded as a 32 bit value (DW_FORM_data4). This allows a single name to have
2084 multiple matching DIEs in a single file, which could come up with an inlined
2085 function for instance. Future tables could include more information about the
2086 DIE such as flags indicating if the DIE is a function, method, block,
2089 The KeyType for the DWARF table is a 32 bit string table offset into the
2090 ".debug_str" table. The ".debug_str" is the string table for the DWARF which
2091 may already contain copies of all of the strings. This helps make sure, with
2092 help from the compiler, that we reuse the strings between all of the DWARF
2093 sections and keeps the hash table size down. Another benefit to having the
2094 compiler generate all strings as DW_FORM_strp in the debug info, is that
2095 DWARF parsing can be made much faster.
2097 After a lookup is made, we get an offset into the hash data. The hash data
2098 needs to be able to deal with 32 bit hash collisions, so the chunk of data
2099 at the offset in the hash data consists of a triple:
2104 uint32_t hash_data_count
2105 HashData[hash_data_count]
2107 If "str_offset" is zero, then the bucket contents are done. 99.9% of the
2108 hash data chunks contain a single item (no 32 bit hash collision):
2110 .. code-block:: none
2113 | 0x00001023 | uint32_t KeyType (.debug_str[0x0001023] => "main")
2114 | 0x00000004 | uint32_t HashData count
2115 | 0x........ | uint32_t HashData[0] DIE offset
2116 | 0x........ | uint32_t HashData[1] DIE offset
2117 | 0x........ | uint32_t HashData[2] DIE offset
2118 | 0x........ | uint32_t HashData[3] DIE offset
2119 | 0x00000000 | uint32_t KeyType (end of hash chain)
2122 If there are collisions, you will have multiple valid string offsets:
2124 .. code-block:: none
2127 | 0x00001023 | uint32_t KeyType (.debug_str[0x0001023] => "main")
2128 | 0x00000004 | uint32_t HashData count
2129 | 0x........ | uint32_t HashData[0] DIE offset
2130 | 0x........ | uint32_t HashData[1] DIE offset
2131 | 0x........ | uint32_t HashData[2] DIE offset
2132 | 0x........ | uint32_t HashData[3] DIE offset
2133 | 0x00002023 | uint32_t KeyType (.debug_str[0x0002023] => "print")
2134 | 0x00000002 | uint32_t HashData count
2135 | 0x........ | uint32_t HashData[0] DIE offset
2136 | 0x........ | uint32_t HashData[1] DIE offset
2137 | 0x00000000 | uint32_t KeyType (end of hash chain)
2140 Current testing with real world C++ binaries has shown that there is around 1
2141 32 bit hash collision per 100,000 name entries.
2146 As we said, we want to strictly define exactly what is included in the
2147 different tables. For DWARF, we have 3 tables: "``.apple_names``",
2148 "``.apple_types``", and "``.apple_namespaces``".
2150 "``.apple_names``" sections should contain an entry for each DWARF DIE whose
2151 ``DW_TAG`` is a ``DW_TAG_label``, ``DW_TAG_inlined_subroutine``, or
2152 ``DW_TAG_subprogram`` that has address attributes: ``DW_AT_low_pc``,
2153 ``DW_AT_high_pc``, ``DW_AT_ranges`` or ``DW_AT_entry_pc``. It also contains
2154 ``DW_TAG_variable`` DIEs that have a ``DW_OP_addr`` in the location (global and
2155 static variables). All global and static variables should be included,
2156 including those scoped within functions and classes. For example using the
2168 Both of the static ``var`` variables would be included in the table. All
2169 functions should emit both their full names and their basenames. For C or C++,
2170 the full name is the mangled name (if available) which is usually in the
2171 ``DW_AT_MIPS_linkage_name`` attribute, and the ``DW_AT_name`` contains the
2172 function basename. If global or static variables have a mangled name in a
2173 ``DW_AT_MIPS_linkage_name`` attribute, this should be emitted along with the
2174 simple name found in the ``DW_AT_name`` attribute.
2176 "``.apple_types``" sections should contain an entry for each DWARF DIE whose
2181 * DW_TAG_enumeration_type
2182 * DW_TAG_pointer_type
2183 * DW_TAG_reference_type
2184 * DW_TAG_string_type
2185 * DW_TAG_structure_type
2186 * DW_TAG_subroutine_type
2189 * DW_TAG_ptr_to_member_type
2191 * DW_TAG_subrange_type
2197 * DW_TAG_packed_type
2198 * DW_TAG_volatile_type
2199 * DW_TAG_restrict_type
2200 * DW_TAG_interface_type
2201 * DW_TAG_unspecified_type
2202 * DW_TAG_shared_type
2204 Only entries with a ``DW_AT_name`` attribute are included, and the entry must
2205 not be a forward declaration (``DW_AT_declaration`` attribute with a non-zero
2206 value). For example, using the following code:
2216 We get a few type DIEs:
2218 .. code-block:: none
2220 0x00000067: TAG_base_type [5]
2221 AT_encoding( DW_ATE_signed )
2223 AT_byte_size( 0x04 )
2225 0x0000006e: TAG_pointer_type [6]
2226 AT_type( {0x00000067} ( int ) )
2227 AT_byte_size( 0x08 )
2229 The DW_TAG_pointer_type is not included because it does not have a ``DW_AT_name``.
2231 "``.apple_namespaces``" section should contain all ``DW_TAG_namespace`` DIEs.
2232 If we run into a namespace that has no name this is an anonymous namespace, and
2233 the name should be output as "``(anonymous namespace)``" (without the quotes).
2234 Why? This matches the output of the ``abi::cxa_demangle()`` that is in the
2235 standard C++ library that demangles mangled names.
2238 Language Extensions and File Format Changes
2239 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2241 Objective-C Extensions
2242 """"""""""""""""""""""
2244 "``.apple_objc``" section should contain all ``DW_TAG_subprogram`` DIEs for an
2245 Objective-C class. The name used in the hash table is the name of the
2246 Objective-C class itself. If the Objective-C class has a category, then an
2247 entry is made for both the class name without the category, and for the class
2248 name with the category. So if we have a DIE at offset 0x1234 with a name of
2249 method "``-[NSString(my_additions) stringWithSpecialString:]``", we would add
2250 an entry for "``NSString``" that points to DIE 0x1234, and an entry for
2251 "``NSString(my_additions)``" that points to 0x1234. This allows us to quickly
2252 track down all Objective-C methods for an Objective-C class when doing
2253 expressions. It is needed because of the dynamic nature of Objective-C where
2254 anyone can add methods to a class. The DWARF for Objective-C methods is also
2255 emitted differently from C++ classes where the methods are not usually
2256 contained in the class definition, they are scattered about across one or more
2257 compile units. Categories can also be defined in different shared libraries.
2258 So we need to be able to quickly find all of the methods and class functions
2259 given the Objective-C class name, or quickly find all methods and class
2260 functions for a class + category name. This table does not contain any
2261 selector names, it just maps Objective-C class names (or class names +
2262 category) to all of the methods and class functions. The selectors are added
2263 as function basenames in the "``.debug_names``" section.
2265 In the "``.apple_names``" section for Objective-C functions, the full name is
2266 the entire function name with the brackets ("``-[NSString
2267 stringWithCString:]``") and the basename is the selector only
2268 ("``stringWithCString:``").
2273 The sections names for the apple hash tables are for non mach-o files. For
2274 mach-o files, the sections should be contained in the ``__DWARF`` segment with
2277 * "``.apple_names``" -> "``__apple_names``"
2278 * "``.apple_types``" -> "``__apple_types``"
2279 * "``.apple_namespaces``" -> "``__apple_namespac``" (16 character limit)
2280 * "``.apple_objc``" -> "``__apple_objc``"