1 ================================
2 Source Level Debugging with LLVM
3 ================================
5 .. sectionauthor:: Chris Lattner <sabre@nondot.org> and Jim Laskey <jlaskey@mac.com>
13 This document is the central repository for all information pertaining to debug
14 information in LLVM. It describes the :ref:`actual format that the LLVM debug
15 information takes <format>`, which is useful for those interested in creating
16 front-ends or dealing directly with the information. Further, this document
17 provides specific examples of what debug information for C/C++ looks like.
19 Philosophy behind LLVM debugging information
20 --------------------------------------------
22 The idea of the LLVM debugging information is to capture how the important
23 pieces of the source-language's Abstract Syntax Tree map onto LLVM code.
24 Several design aspects have shaped the solution that appears here. The
27 * Debugging information should have very little impact on the rest of the
28 compiler. No transformations, analyses, or code generators should need to
29 be modified because of debugging information.
31 * LLVM optimizations should interact in :ref:`well-defined and easily described
32 ways <intro_debugopt>` with the debugging information.
34 * Because LLVM is designed to support arbitrary programming languages,
35 LLVM-to-LLVM tools should not need to know anything about the semantics of
36 the source-level-language.
38 * Source-level languages are often **widely** different from one another.
39 LLVM should not put any restrictions of the flavor of the source-language,
40 and the debugging information should work with any language.
42 * With code generator support, it should be possible to use an LLVM compiler
43 to compile a program to native machine code and standard debugging
44 formats. This allows compatibility with traditional machine-code level
45 debuggers, like GDB or DBX.
47 The approach used by the LLVM implementation is to use a small set of
48 :ref:`intrinsic functions <format_common_intrinsics>` to define a mapping
49 between LLVM program objects and the source-level objects. The description of
50 the source-level program is maintained in LLVM metadata in an
51 :ref:`implementation-defined format <ccxx_frontend>` (the C/C++ front-end
52 currently uses working draft 7 of the `DWARF 3 standard
53 <http://www.eagercon.com/dwarf/dwarf3std.htm>`_).
55 When a program is being debugged, a debugger interacts with the user and turns
56 the stored debug information into source-language specific information. As
57 such, a debugger must be aware of the source-language, and is thus tied to a
58 specific language or family of languages.
60 Debug information consumers
61 ---------------------------
63 The role of debug information is to provide meta information normally stripped
64 away during the compilation process. This meta information provides an LLVM
65 user a relationship between generated code and the original program source
68 Currently, debug information is consumed by DwarfDebug to produce dwarf
69 information used by the gdb debugger. Other targets could use the same
70 information to produce stabs or other debug forms.
72 It would also be reasonable to use debug information to feed profiling tools
73 for analysis of generated code, or, tools for reconstructing the original
74 source from generated code.
76 TODO - expound a bit more.
80 Debugging optimized code
81 ------------------------
83 An extremely high priority of LLVM debugging information is to make it interact
84 well with optimizations and analysis. In particular, the LLVM debug
85 information provides the following guarantees:
87 * LLVM debug information **always provides information to accurately read
88 the source-level state of the program**, regardless of which LLVM
89 optimizations have been run, and without any modification to the
90 optimizations themselves. However, some optimizations may impact the
91 ability to modify the current state of the program with a debugger, such
92 as setting program variables, or calling functions that have been
95 * As desired, LLVM optimizations can be upgraded to be aware of the LLVM
96 debugging information, allowing them to update the debugging information
97 as they perform aggressive optimizations. This means that, with effort,
98 the LLVM optimizers could optimize debug code just as well as non-debug
101 * LLVM debug information does not prevent optimizations from
102 happening (for example inlining, basic block reordering/merging/cleanup,
103 tail duplication, etc).
105 * LLVM debug information is automatically optimized along with the rest of
106 the program, using existing facilities. For example, duplicate
107 information is automatically merged by the linker, and unused information
108 is automatically removed.
110 Basically, the debug information allows you to compile a program with
111 "``-O0 -g``" and get full debug information, allowing you to arbitrarily modify
112 the program as it executes from a debugger. Compiling a program with
113 "``-O3 -g``" gives you full debug information that is always available and
114 accurate for reading (e.g., you get accurate stack traces despite tail call
115 elimination and inlining), but you might lose the ability to modify the program
116 and call functions where were optimized out of the program, or inlined away
119 :ref:`LLVM test suite <test-suite-quickstart>` provides a framework to test
120 optimizer's handling of debugging information. It can be run like this:
124 % cd llvm/projects/test-suite/MultiSource/Benchmarks # or some other level
127 This will test impact of debugging information on optimization passes. If
128 debugging information influences optimization passes then it will be reported
129 as a failure. See :doc:`TestingGuide` for more information on LLVM test
130 infrastructure and how to run various tests.
134 Debugging information format
135 ============================
137 LLVM debugging information has been carefully designed to make it possible for
138 the optimizer to optimize the program and debugging information without
139 necessarily having to know anything about debugging information. In
140 particular, the use of metadata avoids duplicated debugging information from
141 the beginning, and the global dead code elimination pass automatically deletes
142 debugging information for a function if it decides to delete the function.
144 To do this, most of the debugging information (descriptors for types,
145 variables, functions, source files, etc) is inserted by the language front-end
146 in the form of LLVM metadata.
148 Debug information is designed to be agnostic about the target debugger and
149 debugging information representation (e.g. DWARF/Stabs/etc). It uses a generic
150 pass to decode the information that represents variables, types, functions,
151 namespaces, etc: this allows for arbitrary source-language semantics and
152 type-systems to be used, as long as there is a module written for the target
153 debugger to interpret the information.
155 To provide basic functionality, the LLVM debugger does have to make some
156 assumptions about the source-level language being debugged, though it keeps
157 these to a minimum. The only common features that the LLVM debugger assumes
158 exist are :ref:`source files <format_files>`, and :ref:`program objects
159 <format_global_variables>`. These abstract objects are used by a debugger to
160 form stack traces, show information about local variables, etc.
162 This section of the documentation first describes the representation aspects
163 common to any source-language. :ref:`ccxx_frontend` describes the data layout
164 conventions used by the C and C++ front-ends.
166 Debug information descriptors
167 -----------------------------
169 In consideration of the complexity and volume of debug information, LLVM
170 provides a specification for well formed debug descriptors.
172 Consumers of LLVM debug information expect the descriptors for program objects
173 to start in a canonical format, but the descriptors can include additional
174 information appended at the end that is source-language specific. All LLVM
175 debugging information is versioned, allowing backwards compatibility in the
176 case that the core structures need to change in some way. Also, all debugging
177 information objects start with a tag to indicate what type of object it is.
178 The source-language is allowed to define its own objects, by using unreserved
179 tag numbers. We recommend using with tags in the range 0x1000 through 0x2000
180 (there is a defined ``enum DW_TAG_user_base = 0x1000``.)
182 The fields of debug descriptors used internally by LLVM are restricted to only
183 the simple data types ``i32``, ``i1``, ``float``, ``double``, ``mdstring`` and
193 <a name="LLVMDebugVersion">The first field of a descriptor is always an
194 ``i32`` containing a tag value identifying the content of the descriptor.
195 The remaining fields are specific to the descriptor. The values of tags are
196 loosely bound to the tag values of DWARF information entries. However, that
197 does not restrict the use of the information supplied to DWARF targets. To
198 facilitate versioning of debug information, the tag is augmented with the
199 current debug version (``LLVMDebugVersion = 8 << 16`` or 0x80000 or
202 The details of the various descriptors follow.
204 Compile unit descriptors
205 ^^^^^^^^^^^^^^^^^^^^^^^^
210 i32, ;; Tag = 17 + LLVMDebugVersion (DW_TAG_compile_unit)
211 i32, ;; Unused field.
212 i32, ;; DWARF language identifier (ex. DW_LANG_C89)
213 metadata, ;; Source file name
214 metadata, ;; Source file directory (includes trailing slash)
215 metadata ;; Producer (ex. "4.0.1 LLVM (LLVM research group)")
216 i1, ;; True if this is a main compile unit.
217 i1, ;; True if this is optimized.
219 i32 ;; Runtime version
220 metadata ;; List of enums types
221 metadata ;; List of retained types
222 metadata ;; List of subprograms
223 metadata ;; List of global variables
226 These descriptors contain a source language ID for the file (we use the DWARF
227 3.0 ID numbers, such as ``DW_LANG_C89``, ``DW_LANG_C_plus_plus``,
228 ``DW_LANG_Cobol74``, etc), three strings describing the filename, working
229 directory of the compiler, and an identifier string for the compiler that
232 Compile unit descriptors provide the root context for objects declared in a
233 specific compilation unit. File descriptors are defined using this context.
234 These descriptors are collected by a named metadata ``!llvm.dbg.cu``. Compile
235 unit descriptor keeps track of subprograms, global variables and type
246 i32, ;; Tag = 41 + LLVMDebugVersion (DW_TAG_file_type)
247 metadata, ;; Source file name
248 metadata, ;; Source file directory (includes trailing slash)
252 These descriptors contain information for a file. Global variables and top
253 level functions would be defined using this context. File descriptors also
254 provide context for source line correspondence.
256 Each input file is encoded as a separate file descriptor in LLVM debugging
259 .. _format_global_variables:
261 Global variable descriptors
262 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
267 i32, ;; Tag = 52 + LLVMDebugVersion (DW_TAG_variable)
268 i32, ;; Unused field.
269 metadata, ;; Reference to context descriptor
271 metadata, ;; Display name (fully qualified C++ name)
272 metadata, ;; MIPS linkage name (for C++)
273 metadata, ;; Reference to file where defined
274 i32, ;; Line number where defined
275 metadata, ;; Reference to type descriptor
276 i1, ;; True if the global is local to compile unit (static)
277 i1, ;; True if the global is defined in the compile unit (not extern)
278 {}* ;; Reference to the global variable
281 These descriptors provide debug information about globals variables. The
282 provide details such as name, type and where the variable is defined. All
283 global variables are collected inside the named metadata ``!llvm.dbg.cu``.
285 .. _format_subprograms:
287 Subprogram descriptors
288 ^^^^^^^^^^^^^^^^^^^^^^
293 i32, ;; Tag = 46 + LLVMDebugVersion (DW_TAG_subprogram)
294 i32, ;; Unused field.
295 metadata, ;; Reference to context descriptor
297 metadata, ;; Display name (fully qualified C++ name)
298 metadata, ;; MIPS linkage name (for C++)
299 metadata, ;; Reference to file where defined
300 i32, ;; Line number where defined
301 metadata, ;; Reference to type descriptor
302 i1, ;; True if the global is local to compile unit (static)
303 i1, ;; True if the global is defined in the compile unit (not extern)
304 i32, ;; Line number where the scope of the subprogram begins
305 i32, ;; Virtuality, e.g. dwarf::DW_VIRTUALITY__virtual
306 i32, ;; Index into a virtual function
307 metadata, ;; indicates which base type contains the vtable pointer for the
309 i32, ;; Flags - Artifical, Private, Protected, Explicit, Prototyped.
311 Function * , ;; Pointer to LLVM function
312 metadata, ;; Lists function template parameters
313 metadata, ;; Function declaration descriptor
314 metadata ;; List of function variables
317 These descriptors provide debug information about functions, methods and
318 subprograms. They provide details such as name, return types and the source
319 location where the subprogram is defined.
327 i32, ;; Tag = 11 + LLVMDebugVersion (DW_TAG_lexical_block)
328 metadata,;; Reference to context descriptor
330 i32, ;; Column number
331 metadata,;; Reference to source file
332 i32 ;; Unique ID to identify blocks from a template function
335 This descriptor provides debug information about nested blocks within a
336 subprogram. The line number and column numbers are used to dinstinguish two
337 lexical blocks at same depth.
342 i32, ;; Tag = 11 + LLVMDebugVersion (DW_TAG_lexical_block)
343 metadata ;; Reference to the scope we're annotating with a file change
344 metadata,;; Reference to the file the scope is enclosed in.
347 This descriptor provides a wrapper around a lexical scope to handle file
348 changes in the middle of a lexical block.
350 .. _format_basic_type:
352 Basic type descriptors
353 ^^^^^^^^^^^^^^^^^^^^^^
358 i32, ;; Tag = 36 + LLVMDebugVersion (DW_TAG_base_type)
359 metadata, ;; Reference to context
360 metadata, ;; Name (may be "" for anonymous types)
361 metadata, ;; Reference to file where defined (may be NULL)
362 i32, ;; Line number where defined (may be 0)
364 i64, ;; Alignment in bits
365 i64, ;; Offset in bits
367 i32 ;; DWARF type encoding
370 These descriptors define primitive types used in the code. Example ``int``,
371 ``bool`` and ``float``. The context provides the scope of the type, which is
372 usually the top level. Since basic types are not usually user defined the
373 context and line number can be left as NULL and 0. The size, alignment and
374 offset are expressed in bits and can be 64 bit values. The alignment is used
375 to round the offset when embedded in a :ref:`composite type
376 <format_composite_type>` (example to keep float doubles on 64 bit boundaries).
377 The offset is the bit offset if embedded in a :ref:`composite type
378 <format_composite_type>`.
380 The type encoding provides the details of the type. The values are typically
381 one of the following:
389 DW_ATE_signed_char = 6
391 DW_ATE_unsigned_char = 8
393 .. _format_derived_type:
395 Derived type descriptors
396 ^^^^^^^^^^^^^^^^^^^^^^^^
401 i32, ;; Tag (see below)
402 metadata, ;; Reference to context
403 metadata, ;; Name (may be "" for anonymous types)
404 metadata, ;; Reference to file where defined (may be NULL)
405 i32, ;; Line number where defined (may be 0)
407 i64, ;; Alignment in bits
408 i64, ;; Offset in bits
409 i32, ;; Flags to encode attributes, e.g. private
410 metadata, ;; Reference to type derived from
411 metadata, ;; (optional) Name of the Objective C property associated with
412 ;; Objective-C an ivar
413 metadata, ;; (optional) Name of the Objective C property getter selector.
414 metadata, ;; (optional) Name of the Objective C property setter selector.
415 i32 ;; (optional) Objective C property attributes.
418 These descriptors are used to define types derived from other types. The value
419 of the tag varies depending on the meaning. The following are possible tag
424 DW_TAG_formal_parameter = 5
426 DW_TAG_pointer_type = 15
427 DW_TAG_reference_type = 16
429 DW_TAG_const_type = 38
430 DW_TAG_volatile_type = 53
431 DW_TAG_restrict_type = 55
433 ``DW_TAG_member`` is used to define a member of a :ref:`composite type
434 <format_composite_type>` or :ref:`subprogram <format_subprograms>`. The type
435 of the member is the :ref:`derived type <format_derived_type>`.
436 ``DW_TAG_formal_parameter`` is used to define a member which is a formal
437 argument of a subprogram.
439 ``DW_TAG_typedef`` is used to provide a name for the derived type.
441 ``DW_TAG_pointer_type``, ``DW_TAG_reference_type``, ``DW_TAG_const_type``,
442 ``DW_TAG_volatile_type`` and ``DW_TAG_restrict_type`` are used to qualify the
443 :ref:`derived type <format_derived_type>`.
445 :ref:`Derived type <format_derived_type>` location can be determined from the
446 context and line number. The size, alignment and offset are expressed in bits
447 and can be 64 bit values. The alignment is used to round the offset when
448 embedded in a :ref:`composite type <format_composite_type>` (example to keep
449 float doubles on 64 bit boundaries.) The offset is the bit offset if embedded
450 in a :ref:`composite type <format_composite_type>`.
452 Note that the ``void *`` type is expressed as a type derived from NULL.
454 .. _format_composite_type:
456 Composite type descriptors
457 ^^^^^^^^^^^^^^^^^^^^^^^^^^
462 i32, ;; Tag (see below)
463 metadata, ;; Reference to context
464 metadata, ;; Name (may be "" for anonymous types)
465 metadata, ;; Reference to file where defined (may be NULL)
466 i32, ;; Line number where defined (may be 0)
468 i64, ;; Alignment in bits
469 i64, ;; Offset in bits
471 metadata, ;; Reference to type derived from
472 metadata, ;; Reference to array of member descriptors
473 i32 ;; Runtime languages
476 These descriptors are used to define types that are composed of 0 or more
477 elements. The value of the tag varies depending on the meaning. The following
478 are possible tag values:
482 DW_TAG_array_type = 1
483 DW_TAG_enumeration_type = 4
484 DW_TAG_structure_type = 19
485 DW_TAG_union_type = 23
486 DW_TAG_vector_type = 259
487 DW_TAG_subroutine_type = 21
488 DW_TAG_inheritance = 28
490 The vector flag indicates that an array type is a native packed vector.
492 The members of array types (tag = ``DW_TAG_array_type``) or vector types (tag =
493 ``DW_TAG_vector_type``) are :ref:`subrange descriptors <format_subrange>`, each
494 representing the range of subscripts at that level of indexing.
496 The members of enumeration types (tag = ``DW_TAG_enumeration_type``) are
497 :ref:`enumerator descriptors <format_enumerator>`, each representing the
498 definition of enumeration value for the set. All enumeration type descriptors
499 are collected inside the named metadata ``!llvm.dbg.cu``.
501 The members of structure (tag = ``DW_TAG_structure_type``) or union (tag =
502 ``DW_TAG_union_type``) types are any one of the :ref:`basic
503 <format_basic_type>`, :ref:`derived <format_derived_type>` or :ref:`composite
504 <format_composite_type>` type descriptors, each representing a field member of
505 the structure or union.
507 For C++ classes (tag = ``DW_TAG_structure_type``), member descriptors provide
508 information about base classes, static members and member functions. If a
509 member is a :ref:`derived type descriptor <format_derived_type>` and has a tag
510 of ``DW_TAG_inheritance``, then the type represents a base class. If the member
511 of is a :ref:`global variable descriptor <format_global_variables>` then it
512 represents a static member. And, if the member is a :ref:`subprogram
513 descriptor <format_subprograms>` then it represents a member function. For
514 static members and member functions, ``getName()`` returns the members link or
515 the C++ mangled name. ``getDisplayName()`` the simplied version of the name.
517 The first member of subroutine (tag = ``DW_TAG_subroutine_type``) type elements
518 is the return type for the subroutine. The remaining elements are the formal
519 arguments to the subroutine.
521 :ref:`Composite type <format_composite_type>` location can be determined from
522 the context and line number. The size, alignment and offset are expressed in
523 bits and can be 64 bit values. The alignment is used to round the offset when
524 embedded in a :ref:`composite type <format_composite_type>` (as an example, to
525 keep float doubles on 64 bit boundaries). The offset is the bit offset if
526 embedded in a :ref:`composite type <format_composite_type>`.
536 i32, ;; Tag = 33 + LLVMDebugVersion (DW_TAG_subrange_type)
541 These descriptors are used to define ranges of array subscripts for an array
542 :ref:`composite type <format_composite_type>`. The low value defines the lower
543 bounds typically zero for C/C++. The high value is the upper bounds. Values
544 are 64 bit. ``High - Low + 1`` is the size of the array. If ``Low > High``
545 the array bounds are not included in generated debugging information.
547 .. _format_enumerator:
549 Enumerator descriptors
550 ^^^^^^^^^^^^^^^^^^^^^^
555 i32, ;; Tag = 40 + LLVMDebugVersion (DW_TAG_enumerator)
560 These descriptors are used to define members of an enumeration :ref:`composite
561 type <format_composite_type>`, it associates the name to the value.
569 i32, ;; Tag (see below)
572 metadata, ;; Reference to file where defined
573 i32, ;; 24 bit - Line number where defined
574 ;; 8 bit - Argument number. 1 indicates 1st argument.
575 metadata, ;; Type descriptor
577 metadata ;; (optional) Reference to inline location
580 These descriptors are used to define variables local to a sub program. The
581 value of the tag depends on the usage of the variable:
585 DW_TAG_auto_variable = 256
586 DW_TAG_arg_variable = 257
587 DW_TAG_return_variable = 258
589 An auto variable is any variable declared in the body of the function. An
590 argument variable is any variable that appears as a formal argument to the
591 function. A return variable is used to track the result of a function and has
592 no source correspondent.
594 The context is either the subprogram or block where the variable is defined.
595 Name the source variable name. Context and line indicate where the variable
596 was defined. Type descriptor defines the declared type of the variable.
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() nounwind ssp {
663 %X = alloca i32, align 4 ; <i32*> [#uses=4]
664 %Y = alloca i32, align 4 ; <i32*> [#uses=4]
665 %Z = alloca i32, align 4 ; <i32*> [#uses=3]
666 %0 = bitcast i32* %X to {}* ; <{}*> [#uses=1]
667 call void @llvm.dbg.declare(metadata !{i32 * %X}, metadata !0), !dbg !7
668 store i32 21, i32* %X, !dbg !8
669 %1 = bitcast i32* %Y to {}* ; <{}*> [#uses=1]
670 call void @llvm.dbg.declare(metadata !{i32 * %Y}, metadata !9), !dbg !10
671 store i32 22, i32* %Y, !dbg !11
672 %2 = bitcast i32* %Z to {}* ; <{}*> [#uses=1]
673 call void @llvm.dbg.declare(metadata !{i32 * %Z}, metadata !12), !dbg !14
674 store i32 23, i32* %Z, !dbg !15
675 %tmp = load i32* %X, !dbg !16 ; <i32> [#uses=1]
676 %tmp1 = load i32* %Y, !dbg !16 ; <i32> [#uses=1]
677 %add = add nsw i32 %tmp, %tmp1, !dbg !16 ; <i32> [#uses=1]
678 store i32 %add, i32* %Z, !dbg !16
679 %tmp2 = load i32* %Y, !dbg !17 ; <i32> [#uses=1]
680 store i32 %tmp2, i32* %X, !dbg !17
684 declare void @llvm.dbg.declare(metadata, metadata) nounwind readnone
686 !0 = metadata !{i32 459008, metadata !1, metadata !"X",
687 metadata !3, i32 2, metadata !6}; [ DW_TAG_auto_variable ]
688 !1 = metadata !{i32 458763, metadata !2}; [DW_TAG_lexical_block ]
689 !2 = metadata !{i32 458798, i32 0, metadata !3, metadata !"foo", metadata !"foo",
690 metadata !"foo", metadata !3, i32 1, metadata !4,
691 i1 false, i1 true}; [DW_TAG_subprogram ]
692 !3 = metadata !{i32 458769, i32 0, i32 12, metadata !"foo.c",
693 metadata !"/private/tmp", metadata !"clang 1.1", i1 true,
694 i1 false, metadata !"", i32 0}; [DW_TAG_compile_unit ]
695 !4 = metadata !{i32 458773, metadata !3, metadata !"", null, i32 0, i64 0, i64 0,
696 i64 0, i32 0, null, metadata !5, i32 0}; [DW_TAG_subroutine_type ]
697 !5 = metadata !{null}
698 !6 = metadata !{i32 458788, metadata !3, metadata !"int", metadata !3, i32 0,
699 i64 32, i64 32, i64 0, i32 0, i32 5}; [DW_TAG_base_type ]
700 !7 = metadata !{i32 2, i32 7, metadata !1, null}
701 !8 = metadata !{i32 2, i32 3, metadata !1, null}
702 !9 = metadata !{i32 459008, metadata !1, metadata !"Y", metadata !3, i32 3,
703 metadata !6}; [ DW_TAG_auto_variable ]
704 !10 = metadata !{i32 3, i32 7, metadata !1, null}
705 !11 = metadata !{i32 3, i32 3, metadata !1, null}
706 !12 = metadata !{i32 459008, metadata !13, metadata !"Z", metadata !3, i32 5,
707 metadata !6}; [ DW_TAG_auto_variable ]
708 !13 = metadata !{i32 458763, metadata !1}; [DW_TAG_lexical_block ]
709 !14 = metadata !{i32 5, i32 9, metadata !13, null}
710 !15 = metadata !{i32 5, i32 5, metadata !13, null}
711 !16 = metadata !{i32 6, i32 5, metadata !13, null}
712 !17 = metadata !{i32 8, i32 3, metadata !1, null}
713 !18 = metadata !{i32 9, i32 1, metadata !2, null}
715 This example illustrates a few important details about LLVM debugging
716 information. In particular, it shows how the ``llvm.dbg.declare`` intrinsic and
717 location information, which are attached to an instruction, are applied
718 together to allow a debugger to analyze the relationship between statements,
719 variable definitions, and the code used to implement the function.
723 call void @llvm.dbg.declare(metadata, metadata !0), !dbg !7
725 The first intrinsic ``%llvm.dbg.declare`` encodes debugging information for the
726 variable ``X``. The metadata ``!dbg !7`` attached to the intrinsic provides
727 scope information for the variable ``X``.
731 !7 = metadata !{i32 2, i32 7, metadata !1, null}
732 !1 = metadata !{i32 458763, metadata !2}; [DW_TAG_lexical_block ]
733 !2 = metadata !{i32 458798, i32 0, metadata !3, metadata !"foo",
734 metadata !"foo", metadata !"foo", metadata !3, i32 1,
735 metadata !4, i1 false, i1 true}; [DW_TAG_subprogram ]
737 Here ``!7`` is metadata providing location information. It has four fields:
738 line number, column number, scope, and original scope. The original scope
739 represents inline location if this instruction is inlined inside a caller, and
740 is null otherwise. In this example, scope is encoded by ``!1``. ``!1``
741 represents a lexical block inside the scope ``!2``, where ``!2`` is a
742 :ref:`subprogram descriptor <format_subprograms>`. This way the location
743 information attached to the intrinsics indicates that the variable ``X`` is
744 declared at line number 2 at a function level scope in function ``foo``.
746 Now lets take another example.
750 call void @llvm.dbg.declare(metadata, metadata !12), !dbg !14
752 The second intrinsic ``%llvm.dbg.declare`` encodes debugging information for
753 variable ``Z``. The metadata ``!dbg !14`` attached to the intrinsic provides
754 scope information for the variable ``Z``.
758 !13 = metadata !{i32 458763, metadata !1}; [DW_TAG_lexical_block ]
759 !14 = metadata !{i32 5, i32 9, metadata !13, null}
761 Here ``!14`` indicates that ``Z`` is declared at line number 5 and
762 column number 9 inside of lexical scope ``!13``. The lexical scope itself
763 resides inside of lexical scope ``!1`` described above.
765 The scope information attached with each instruction provides a straightforward
766 way to find instructions covered by a scope.
770 C/C++ front-end specific debug information
771 ==========================================
773 The C and C++ front-ends represent information about the program in a format
774 that is effectively identical to `DWARF 3.0
775 <http://www.eagercon.com/dwarf/dwarf3std.htm>`_ in terms of information
776 content. This allows code generators to trivially support native debuggers by
777 generating standard dwarf information, and contains enough information for
778 non-dwarf targets to translate it as needed.
780 This section describes the forms used to represent C and C++ programs. Other
781 languages could pattern themselves after this (which itself is tuned to
782 representing programs in the same way that DWARF 3 does), or they could choose
783 to provide completely different forms if they don't fit into the DWARF model.
784 As support for debugging information gets added to the various LLVM
785 source-language front-ends, the information used should be documented here.
787 The following sections provide examples of various C/C++ constructs and the
788 debug information that would best describe those constructs.
790 C/C++ source file information
791 -----------------------------
793 Given the source files ``MySource.cpp`` and ``MyHeader.h`` located in the
794 directory ``/Users/mine/sources``, the following code:
798 #include "MyHeader.h"
800 int main(int argc, char *argv[]) {
804 a C/C++ front-end would generate the following descriptors:
810 ;; Define the compile unit for the main source file "/Users/mine/sources/MySource.cpp".
815 i32 4, ;; Language Id
816 metadata !"MySource.cpp",
817 metadata !"/Users/mine/sources",
818 metadata !"4.2.1 (Based on Apple Inc. build 5649) (LLVM build 00)",
819 i1 true, ;; Main Compile Unit
820 i1 false, ;; Optimized compile unit
821 metadata !"", ;; Compiler flags
822 i32 0} ;; Runtime version
825 ;; Define the file for the file "/Users/mine/sources/MySource.cpp".
829 metadata !"MySource.cpp",
830 metadata !"/Users/mine/sources",
831 metadata !2 ;; Compile unit
835 ;; Define the file for the file "/Users/mine/sources/Myheader.h"
839 metadata !"Myheader.h"
840 metadata !"/Users/mine/sources",
841 metadata !2 ;; Compile unit
846 ``llvm::Instruction`` provides easy access to metadata attached with an
847 instruction. One can extract line number information encoded in LLVM IR using
848 ``Instruction::getMetadata()`` and ``DILocation::getLineNumber()``.
852 if (MDNode *N = I->getMetadata("dbg")) { // Here I is an LLVM instruction
853 DILocation Loc(N); // DILocation is in DebugInfo.h
854 unsigned Line = Loc.getLineNumber();
855 StringRef File = Loc.getFilename();
856 StringRef Dir = Loc.getDirectory();
859 C/C++ global variable information
860 ---------------------------------
862 Given an integer global variable declared as follows:
868 a C/C++ front-end would generate the following descriptors:
873 ;; Define the global itself.
875 %MyGlobal = global int 100
878 ;; List of debug info of globals
882 ;; Define the compile unit.
887 metadata !"foo.cpp", ;; File
888 metadata !"/Volumes/Data/tmp", ;; Directory
889 metadata !"clang version 3.1 ", ;; Producer
890 i1 true, ;; Deprecated field
891 i1 false, ;; "isOptimized"?
892 metadata !"", ;; Flags
893 i32 0, ;; Runtime Version
894 metadata !1, ;; Enum Types
895 metadata !1, ;; Retained Types
896 metadata !1, ;; Subprograms
897 metadata !3 ;; Global Variables
898 } ; [ DW_TAG_compile_unit ]
900 ;; The Array of Global Variables
910 ;; Define the global variable itself.
916 metadata !"MyGlobal", ;; Name
917 metadata !"MyGlobal", ;; Display Name
918 metadata !"", ;; Linkage Name
922 i32 0, ;; IsLocalToUnit
923 i32 1, ;; IsDefinition
924 i32* @MyGlobal ;; LLVM-IR Value
925 } ; [ DW_TAG_variable ]
932 metadata !"foo.cpp", ;; File
933 metadata !"/Volumes/Data/tmp", ;; Directory
935 } ; [ DW_TAG_file_type ]
943 metadata !"int", ;; Name
946 i64 32, ;; Size in Bits
947 i64 32, ;; Align in Bits
951 } ; [ DW_TAG_base_type ]
953 C/C++ function information
954 --------------------------
956 Given a function declared as follows:
960 int main(int argc, char *argv[]) {
964 a C/C++ front-end would generate the following descriptors:
969 ;; Define the anchor for subprograms. Note that the second field of the
970 ;; anchor is 46, which is the same as the tag for subprograms
971 ;; (46 = DW_TAG_subprogram.)
976 metadata !1, ;; Context
977 metadata !"main", ;; Name
978 metadata !"main", ;; Display name
979 metadata !"main", ;; Linkage name
981 i32 1, ;; Line number
983 i1 false, ;; Is local
984 i1 true, ;; Is definition
985 i32 0, ;; Virtuality attribute, e.g. pure virtual function
986 i32 0, ;; Index into virtual table for C++ methods
987 i32 0, ;; Type that holds virtual table.
989 i1 false, ;; True if this function is optimized
990 Function *, ;; Pointer to llvm::Function
991 null ;; Function template parameters
994 ;; Define the subprogram itself.
996 define i32 @main(i32 %argc, i8** %argv) {
1003 The following are the basic type descriptors for C/C++ core types:
1008 .. code-block:: llvm
1012 metadata !1, ;; Context
1013 metadata !"bool", ;; Name
1014 metadata !1, ;; File
1015 i32 0, ;; Line number
1016 i64 8, ;; Size in Bits
1017 i64 8, ;; Align in Bits
1018 i64 0, ;; Offset in Bits
1026 .. code-block:: llvm
1030 metadata !1, ;; Context
1031 metadata !"char", ;; Name
1032 metadata !1, ;; File
1033 i32 0, ;; Line number
1034 i64 8, ;; Size in Bits
1035 i64 8, ;; Align in Bits
1036 i64 0, ;; Offset in Bits
1044 .. code-block:: llvm
1048 metadata !1, ;; Context
1049 metadata !"unsigned char",
1050 metadata !1, ;; File
1051 i32 0, ;; Line number
1052 i64 8, ;; Size in Bits
1053 i64 8, ;; Align in Bits
1054 i64 0, ;; Offset in Bits
1062 .. code-block:: llvm
1066 metadata !1, ;; Context
1067 metadata !"short int",
1068 metadata !1, ;; File
1069 i32 0, ;; Line number
1070 i64 16, ;; Size in Bits
1071 i64 16, ;; Align in Bits
1072 i64 0, ;; Offset in Bits
1080 .. code-block:: llvm
1084 metadata !1, ;; Context
1085 metadata !"short unsigned int",
1086 metadata !1, ;; File
1087 i32 0, ;; Line number
1088 i64 16, ;; Size in Bits
1089 i64 16, ;; Align in Bits
1090 i64 0, ;; Offset in Bits
1098 .. code-block:: llvm
1102 metadata !1, ;; Context
1103 metadata !"int", ;; Name
1104 metadata !1, ;; File
1105 i32 0, ;; Line number
1106 i64 32, ;; Size in Bits
1107 i64 32, ;; Align in Bits
1108 i64 0, ;; Offset in Bits
1116 .. code-block:: llvm
1120 metadata !1, ;; Context
1121 metadata !"unsigned int",
1122 metadata !1, ;; File
1123 i32 0, ;; Line number
1124 i64 32, ;; Size in Bits
1125 i64 32, ;; Align in Bits
1126 i64 0, ;; Offset in Bits
1134 .. code-block:: llvm
1138 metadata !1, ;; Context
1139 metadata !"long long int",
1140 metadata !1, ;; File
1141 i32 0, ;; Line number
1142 i64 64, ;; Size in Bits
1143 i64 64, ;; Align in Bits
1144 i64 0, ;; Offset in Bits
1152 .. code-block:: llvm
1156 metadata !1, ;; Context
1157 metadata !"long long unsigned int",
1158 metadata !1, ;; File
1159 i32 0, ;; Line number
1160 i64 64, ;; Size in Bits
1161 i64 64, ;; Align in Bits
1162 i64 0, ;; Offset in Bits
1170 .. code-block:: llvm
1174 metadata !1, ;; Context
1176 metadata !1, ;; File
1177 i32 0, ;; Line number
1178 i64 32, ;; Size in Bits
1179 i64 32, ;; Align in Bits
1180 i64 0, ;; Offset in Bits
1188 .. code-block:: llvm
1192 metadata !1, ;; Context
1193 metadata !"double",;; Name
1194 metadata !1, ;; File
1195 i32 0, ;; Line number
1196 i64 64, ;; Size in Bits
1197 i64 64, ;; Align in Bits
1198 i64 0, ;; Offset in Bits
1206 Given the following as an example of C/C++ derived type:
1210 typedef const int *IntPtr;
1212 a C/C++ front-end would generate the following descriptors:
1214 .. code-block:: llvm
1217 ;; Define the typedef "IntPtr".
1221 metadata !1, ;; Context
1222 metadata !"IntPtr", ;; Name
1223 metadata !3, ;; File
1224 i32 0, ;; Line number
1225 i64 0, ;; Size in bits
1226 i64 0, ;; Align in bits
1227 i64 0, ;; Offset in bits
1229 metadata !4 ;; Derived From type
1232 ;; Define the pointer type.
1236 metadata !1, ;; Context
1237 metadata !"", ;; Name
1238 metadata !1, ;; File
1239 i32 0, ;; Line number
1240 i64 64, ;; Size in bits
1241 i64 64, ;; Align in bits
1242 i64 0, ;; Offset in bits
1244 metadata !5 ;; Derived From type
1247 ;; Define the const type.
1251 metadata !1, ;; Context
1252 metadata !"", ;; Name
1253 metadata !1, ;; File
1254 i32 0, ;; Line number
1255 i64 32, ;; Size in bits
1256 i64 32, ;; Align in bits
1257 i64 0, ;; Offset in bits
1259 metadata !6 ;; Derived From type
1262 ;; Define the int type.
1266 metadata !1, ;; Context
1267 metadata !"int", ;; Name
1268 metadata !1, ;; File
1269 i32 0, ;; Line number
1270 i64 32, ;; Size in bits
1271 i64 32, ;; Align in bits
1272 i64 0, ;; Offset in bits
1277 C/C++ struct/union types
1278 ------------------------
1280 Given the following as an example of C/C++ struct type:
1290 a C/C++ front-end would generate the following descriptors:
1292 .. code-block:: llvm
1295 ;; Define basic type for unsigned int.
1299 metadata !1, ;; Context
1300 metadata !"unsigned int",
1301 metadata !1, ;; File
1302 i32 0, ;; Line number
1303 i64 32, ;; Size in Bits
1304 i64 32, ;; Align in Bits
1305 i64 0, ;; Offset in Bits
1310 ;; Define composite type for struct Color.
1314 metadata !1, ;; Context
1315 metadata !"Color", ;; Name
1316 metadata !1, ;; Compile unit
1317 i32 1, ;; Line number
1318 i64 96, ;; Size in bits
1319 i64 32, ;; Align in bits
1320 i64 0, ;; Offset in bits
1322 null, ;; Derived From
1323 metadata !3, ;; Elements
1324 i32 0 ;; Runtime Language
1328 ;; Define the Red field.
1332 metadata !1, ;; Context
1333 metadata !"Red", ;; Name
1334 metadata !1, ;; File
1335 i32 2, ;; Line number
1336 i64 32, ;; Size in bits
1337 i64 32, ;; Align in bits
1338 i64 0, ;; Offset in bits
1340 metadata !5 ;; Derived From type
1344 ;; Define the Green field.
1348 metadata !1, ;; Context
1349 metadata !"Green", ;; Name
1350 metadata !1, ;; File
1351 i32 3, ;; Line number
1352 i64 32, ;; Size in bits
1353 i64 32, ;; Align in bits
1354 i64 32, ;; Offset in bits
1356 metadata !5 ;; Derived From type
1360 ;; Define the Blue field.
1364 metadata !1, ;; Context
1365 metadata !"Blue", ;; Name
1366 metadata !1, ;; File
1367 i32 4, ;; Line number
1368 i64 32, ;; Size in bits
1369 i64 32, ;; Align in bits
1370 i64 64, ;; Offset in bits
1372 metadata !5 ;; Derived From type
1376 ;; Define the array of fields used by the composite type Color.
1378 !3 = metadata !{metadata !4, metadata !6, metadata !7}
1380 C/C++ enumeration types
1381 -----------------------
1383 Given the following as an example of C/C++ enumeration type:
1393 a C/C++ front-end would generate the following descriptors:
1395 .. code-block:: llvm
1398 ;; Define composite type for enum Trees
1402 metadata !1, ;; Context
1403 metadata !"Trees", ;; Name
1404 metadata !1, ;; File
1405 i32 1, ;; Line number
1406 i64 32, ;; Size in bits
1407 i64 32, ;; Align in bits
1408 i64 0, ;; Offset in bits
1410 null, ;; Derived From type
1411 metadata !3, ;; Elements
1412 i32 0 ;; Runtime language
1416 ;; Define the array of enumerators used by composite type Trees.
1418 !3 = metadata !{metadata !4, metadata !5, metadata !6}
1421 ;; Define Spruce enumerator.
1423 !4 = metadata !{i32 524328, metadata !"Spruce", i64 100}
1426 ;; Define Oak enumerator.
1428 !5 = metadata !{i32 524328, metadata !"Oak", i64 200}
1431 ;; Define Maple enumerator.
1433 !6 = metadata !{i32 524328, metadata !"Maple", i64 300}
1435 Debugging information format
1436 ============================
1438 Debugging Information Extension for Objective C Properties
1439 ----------------------------------------------------------
1444 Objective C provides a simpler way to declare and define accessor methods using
1445 declared properties. The language provides features to declare a property and
1446 to let compiler synthesize accessor methods.
1448 The debugger lets developer inspect Objective C interfaces and their instance
1449 variables and class variables. However, the debugger does not know anything
1450 about the properties defined in Objective C interfaces. The debugger consumes
1451 information generated by compiler in DWARF format. The format does not support
1452 encoding of Objective C properties. This proposal describes DWARF extensions to
1453 encode Objective C properties, which the debugger can use to let developers
1454 inspect Objective C properties.
1459 Objective C properties exist separately from class members. A property can be
1460 defined only by "setter" and "getter" selectors, and be calculated anew on each
1461 access. Or a property can just be a direct access to some declared ivar.
1462 Finally it can have an ivar "automatically synthesized" for it by the compiler,
1463 in which case the property can be referred to in user code directly using the
1464 standard C dereference syntax as well as through the property "dot" syntax, but
1465 there is no entry in the ``@interface`` declaration corresponding to this ivar.
1467 To facilitate debugging, these properties we will add a new DWARF TAG into the
1468 ``DW_TAG_structure_type`` definition for the class to hold the description of a
1469 given property, and a set of DWARF attributes that provide said description.
1470 The property tag will also contain the name and declared type of the property.
1472 If there is a related ivar, there will also be a DWARF property attribute placed
1473 in the ``DW_TAG_member`` DIE for that ivar referring back to the property TAG
1474 for that property. And in the case where the compiler synthesizes the ivar
1475 directly, the compiler is expected to generate a ``DW_TAG_member`` for that
1476 ivar (with the ``DW_AT_artificial`` set to 1), whose name will be the name used
1477 to access this ivar directly in code, and with the property attribute pointing
1478 back to the property it is backing.
1480 The following examples will serve as illustration for our discussion:
1482 .. code-block:: objc
1494 @synthesize p2 = n2;
1497 This produces the following DWARF (this is a "pseudo dwarfdump" output):
1499 .. code-block:: none
1501 0x00000100: TAG_structure_type [7] *
1502 AT_APPLE_runtime_class( 0x10 )
1504 AT_decl_file( "Objc_Property.m" )
1507 0x00000110 TAG_APPLE_property
1509 AT_type ( {0x00000150} ( int ) )
1511 0x00000120: TAG_APPLE_property
1513 AT_type ( {0x00000150} ( int ) )
1515 0x00000130: TAG_member [8]
1517 AT_APPLE_property ( {0x00000110} "p1" )
1518 AT_type( {0x00000150} ( int ) )
1519 AT_artificial ( 0x1 )
1521 0x00000140: TAG_member [8]
1523 AT_APPLE_property ( {0x00000120} "p2" )
1524 AT_type( {0x00000150} ( int ) )
1526 0x00000150: AT_type( ( int ) )
1528 Note, the current convention is that the name of the ivar for an
1529 auto-synthesized property is the name of the property from which it derives
1530 with an underscore prepended, as is shown in the example. But we actually
1531 don't need to know this convention, since we are given the name of the ivar
1534 Also, it is common practice in ObjC to have different property declarations in
1535 the @interface and @implementation - e.g. to provide a read-only property in
1536 the interface,and a read-write interface in the implementation. In that case,
1537 the compiler should emit whichever property declaration will be in force in the
1538 current translation unit.
1540 Developers can decorate a property with attributes which are encoded using
1541 ``DW_AT_APPLE_property_attribute``.
1543 .. code-block:: objc
1545 @property (readonly, nonatomic) int pr;
1547 .. code-block:: none
1549 TAG_APPLE_property [8]
1551 AT_type ( {0x00000147} (int) )
1552 AT_APPLE_property_attribute (DW_APPLE_PROPERTY_readonly, DW_APPLE_PROPERTY_nonatomic)
1554 The setter and getter method names are attached to the property using
1555 ``DW_AT_APPLE_property_setter`` and ``DW_AT_APPLE_property_getter`` attributes.
1557 .. code-block:: objc
1560 @property (setter=myOwnP3Setter:) int p3;
1561 -(void)myOwnP3Setter:(int)a;
1566 -(void)myOwnP3Setter:(int)a{ }
1569 The DWARF for this would be:
1571 .. code-block:: none
1573 0x000003bd: TAG_structure_type [7] *
1574 AT_APPLE_runtime_class( 0x10 )
1576 AT_decl_file( "Objc_Property.m" )
1579 0x000003cd TAG_APPLE_property
1581 AT_APPLE_property_setter ( "myOwnP3Setter:" )
1582 AT_type( {0x00000147} ( int ) )
1584 0x000003f3: TAG_member [8]
1586 AT_type ( {0x00000147} ( int ) )
1587 AT_APPLE_property ( {0x000003cd} )
1588 AT_artificial ( 0x1 )
1593 +-----------------------+--------+
1595 +=======================+========+
1596 | DW_TAG_APPLE_property | 0x4200 |
1597 +-----------------------+--------+
1599 New DWARF Attributes
1600 ^^^^^^^^^^^^^^^^^^^^
1602 +--------------------------------+--------+-----------+
1603 | Attribute | Value | Classes |
1604 +================================+========+===========+
1605 | DW_AT_APPLE_property | 0x3fed | Reference |
1606 +--------------------------------+--------+-----------+
1607 | DW_AT_APPLE_property_getter | 0x3fe9 | String |
1608 +--------------------------------+--------+-----------+
1609 | DW_AT_APPLE_property_setter | 0x3fea | String |
1610 +--------------------------------+--------+-----------+
1611 | DW_AT_APPLE_property_attribute | 0x3feb | Constant |
1612 +--------------------------------+--------+-----------+
1617 +--------------------------------+-------+
1619 +================================+=======+
1620 | DW_AT_APPLE_PROPERTY_readonly | 0x1 |
1621 +--------------------------------+-------+
1622 | DW_AT_APPLE_PROPERTY_readwrite | 0x2 |
1623 +--------------------------------+-------+
1624 | DW_AT_APPLE_PROPERTY_assign | 0x4 |
1625 +--------------------------------+-------+
1626 | DW_AT_APPLE_PROPERTY_retain | 0x8 |
1627 +--------------------------------+-------+
1628 | DW_AT_APPLE_PROPERTY_copy | 0x10 |
1629 +--------------------------------+-------+
1630 | DW_AT_APPLE_PROPERTY_nonatomic | 0x20 |
1631 +--------------------------------+-------+
1633 Name Accelerator Tables
1634 -----------------------
1639 The "``.debug_pubnames``" and "``.debug_pubtypes``" formats are not what a
1640 debugger needs. The "``pub``" in the section name indicates that the entries
1641 in the table are publicly visible names only. This means no static or hidden
1642 functions show up in the "``.debug_pubnames``". No static variables or private
1643 class variables are in the "``.debug_pubtypes``". Many compilers add different
1644 things to these tables, so we can't rely upon the contents between gcc, icc, or
1647 The typical query given by users tends not to match up with the contents of
1648 these tables. For example, the DWARF spec states that "In the case of the name
1649 of a function member or static data member of a C++ structure, class or union,
1650 the name presented in the "``.debug_pubnames``" section is not the simple name
1651 given by the ``DW_AT_name attribute`` of the referenced debugging information
1652 entry, but rather the fully qualified name of the data or function member."
1653 So the only names in these tables for complex C++ entries is a fully
1654 qualified name. Debugger users tend not to enter their search strings as
1655 "``a::b::c(int,const Foo&) const``", but rather as "``c``", "``b::c``" , or
1656 "``a::b::c``". So the name entered in the name table must be demangled in
1657 order to chop it up appropriately and additional names must be manually entered
1658 into the table to make it effective as a name lookup table for debuggers to
1661 All debuggers currently ignore the "``.debug_pubnames``" table as a result of
1662 its inconsistent and useless public-only name content making it a waste of
1663 space in the object file. These tables, when they are written to disk, are not
1664 sorted in any way, leaving every debugger to do its own parsing and sorting.
1665 These tables also include an inlined copy of the string values in the table
1666 itself making the tables much larger than they need to be on disk, especially
1667 for large C++ programs.
1669 Can't we just fix the sections by adding all of the names we need to this
1670 table? No, because that is not what the tables are defined to contain and we
1671 won't know the difference between the old bad tables and the new good tables.
1672 At best we could make our own renamed sections that contain all of the data we
1675 These tables are also insufficient for what a debugger like LLDB needs. LLDB
1676 uses clang for its expression parsing where LLDB acts as a PCH. LLDB is then
1677 often asked to look for type "``foo``" or namespace "``bar``", or list items in
1678 namespace "``baz``". Namespaces are not included in the pubnames or pubtypes
1679 tables. Since clang asks a lot of questions when it is parsing an expression,
1680 we need to be very fast when looking up names, as it happens a lot. Having new
1681 accelerator tables that are optimized for very quick lookups will benefit this
1682 type of debugging experience greatly.
1684 We would like to generate name lookup tables that can be mapped into memory
1685 from disk, and used as is, with little or no up-front parsing. We would also
1686 be able to control the exact content of these different tables so they contain
1687 exactly what we need. The Name Accelerator Tables were designed to fix these
1688 issues. In order to solve these issues we need to:
1690 * Have a format that can be mapped into memory from disk and used as is
1691 * Lookups should be very fast
1692 * Extensible table format so these tables can be made by many producers
1693 * Contain all of the names needed for typical lookups out of the box
1694 * Strict rules for the contents of tables
1696 Table size is important and the accelerator table format should allow the reuse
1697 of strings from common string tables so the strings for the names are not
1698 duplicated. We also want to make sure the table is ready to be used as-is by
1699 simply mapping the table into memory with minimal header parsing.
1701 The name lookups need to be fast and optimized for the kinds of lookups that
1702 debuggers tend to do. Optimally we would like to touch as few parts of the
1703 mapped table as possible when doing a name lookup and be able to quickly find
1704 the name entry we are looking for, or discover there are no matches. In the
1705 case of debuggers we optimized for lookups that fail most of the time.
1707 Each table that is defined should have strict rules on exactly what is in the
1708 accelerator tables and documented so clients can rely on the content.
1713 Standard Hash Tables
1714 """"""""""""""""""""
1716 Typical hash tables have a header, buckets, and each bucket points to the
1719 .. code-block:: none
1729 The BUCKETS are an array of offsets to DATA for each hash:
1731 .. code-block:: none
1734 | 0x00001000 | BUCKETS[0]
1735 | 0x00002000 | BUCKETS[1]
1736 | 0x00002200 | BUCKETS[2]
1737 | 0x000034f0 | BUCKETS[3]
1739 | 0xXXXXXXXX | BUCKETS[n_buckets]
1742 So for ``bucket[3]`` in the example above, we have an offset into the table
1743 0x000034f0 which points to a chain of entries for the bucket. Each bucket must
1744 contain a next pointer, full 32 bit hash value, the string itself, and the data
1745 for the current string value.
1747 .. code-block:: none
1750 0x000034f0: | 0x00003500 | next pointer
1751 | 0x12345678 | 32 bit hash
1752 | "erase" | string value
1753 | data[n] | HashData for this bucket
1755 0x00003500: | 0x00003550 | next pointer
1756 | 0x29273623 | 32 bit hash
1757 | "dump" | string value
1758 | data[n] | HashData for this bucket
1760 0x00003550: | 0x00000000 | next pointer
1761 | 0x82638293 | 32 bit hash
1762 | "main" | string value
1763 | data[n] | HashData for this bucket
1766 The problem with this layout for debuggers is that we need to optimize for the
1767 negative lookup case where the symbol we're searching for is not present. So
1768 if we were to lookup "``printf``" in the table above, we would make a 32 hash
1769 for "``printf``", it might match ``bucket[3]``. We would need to go to the
1770 offset 0x000034f0 and start looking to see if our 32 bit hash matches. To do
1771 so, we need to read the next pointer, then read the hash, compare it, and skip
1772 to the next bucket. Each time we are skipping many bytes in memory and
1773 touching new cache pages just to do the compare on the full 32 bit hash. All
1774 of these accesses then tell us that we didn't have a match.
1779 To solve the issues mentioned above we have structured the hash tables a bit
1780 differently: a header, buckets, an array of all unique 32 bit hash values,
1781 followed by an array of hash value data offsets, one for each hash value, then
1782 the data for all hash values:
1784 .. code-block:: none
1798 The ``BUCKETS`` in the name tables are an index into the ``HASHES`` array. By
1799 making all of the full 32 bit hash values contiguous in memory, we allow
1800 ourselves to efficiently check for a match while touching as little memory as
1801 possible. Most often checking the 32 bit hash values is as far as the lookup
1802 goes. If it does match, it usually is a match with no collisions. So for a
1803 table with "``n_buckets``" buckets, and "``n_hashes``" unique 32 bit hash
1804 values, we can clarify the contents of the ``BUCKETS``, ``HASHES`` and
1807 .. code-block:: none
1809 .-------------------------.
1810 | HEADER.magic | uint32_t
1811 | HEADER.version | uint16_t
1812 | HEADER.hash_function | uint16_t
1813 | HEADER.bucket_count | uint32_t
1814 | HEADER.hashes_count | uint32_t
1815 | HEADER.header_data_len | uint32_t
1816 | HEADER_DATA | HeaderData
1817 |-------------------------|
1818 | BUCKETS | uint32_t[bucket_count] // 32 bit hash indexes
1819 |-------------------------|
1820 | HASHES | uint32_t[hashes_count] // 32 bit hash values
1821 |-------------------------|
1822 | OFFSETS | uint32_t[hashes_count] // 32 bit offsets to hash value data
1823 |-------------------------|
1825 `-------------------------'
1827 So taking the exact same data from the standard hash example above we end up
1830 .. code-block:: none
1840 | ... | BUCKETS[n_buckets]
1842 | 0x........ | HASHES[0]
1843 | 0x........ | HASHES[1]
1844 | 0x........ | HASHES[2]
1845 | 0x........ | HASHES[3]
1846 | 0x........ | HASHES[4]
1847 | 0x........ | HASHES[5]
1848 | 0x12345678 | HASHES[6] hash for BUCKETS[3]
1849 | 0x29273623 | HASHES[7] hash for BUCKETS[3]
1850 | 0x82638293 | HASHES[8] hash for BUCKETS[3]
1851 | 0x........ | HASHES[9]
1852 | 0x........ | HASHES[10]
1853 | 0x........ | HASHES[11]
1854 | 0x........ | HASHES[12]
1855 | 0x........ | HASHES[13]
1856 | 0x........ | HASHES[n_hashes]
1858 | 0x........ | OFFSETS[0]
1859 | 0x........ | OFFSETS[1]
1860 | 0x........ | OFFSETS[2]
1861 | 0x........ | OFFSETS[3]
1862 | 0x........ | OFFSETS[4]
1863 | 0x........ | OFFSETS[5]
1864 | 0x000034f0 | OFFSETS[6] offset for BUCKETS[3]
1865 | 0x00003500 | OFFSETS[7] offset for BUCKETS[3]
1866 | 0x00003550 | OFFSETS[8] offset for BUCKETS[3]
1867 | 0x........ | OFFSETS[9]
1868 | 0x........ | OFFSETS[10]
1869 | 0x........ | OFFSETS[11]
1870 | 0x........ | OFFSETS[12]
1871 | 0x........ | OFFSETS[13]
1872 | 0x........ | OFFSETS[n_hashes]
1880 0x000034f0: | 0x00001203 | .debug_str ("erase")
1881 | 0x00000004 | A 32 bit array count - number of HashData with name "erase"
1882 | 0x........ | HashData[0]
1883 | 0x........ | HashData[1]
1884 | 0x........ | HashData[2]
1885 | 0x........ | HashData[3]
1886 | 0x00000000 | String offset into .debug_str (terminate data for hash)
1888 0x00003500: | 0x00001203 | String offset into .debug_str ("collision")
1889 | 0x00000002 | A 32 bit array count - number of HashData with name "collision"
1890 | 0x........ | HashData[0]
1891 | 0x........ | HashData[1]
1892 | 0x00001203 | String offset into .debug_str ("dump")
1893 | 0x00000003 | A 32 bit array count - number of HashData with name "dump"
1894 | 0x........ | HashData[0]
1895 | 0x........ | HashData[1]
1896 | 0x........ | HashData[2]
1897 | 0x00000000 | String offset into .debug_str (terminate data for hash)
1899 0x00003550: | 0x00001203 | String offset into .debug_str ("main")
1900 | 0x00000009 | A 32 bit array count - number of HashData with name "main"
1901 | 0x........ | HashData[0]
1902 | 0x........ | HashData[1]
1903 | 0x........ | HashData[2]
1904 | 0x........ | HashData[3]
1905 | 0x........ | HashData[4]
1906 | 0x........ | HashData[5]
1907 | 0x........ | HashData[6]
1908 | 0x........ | HashData[7]
1909 | 0x........ | HashData[8]
1910 | 0x00000000 | String offset into .debug_str (terminate data for hash)
1913 So we still have all of the same data, we just organize it more efficiently for
1914 debugger lookup. If we repeat the same "``printf``" lookup from above, we
1915 would hash "``printf``" and find it matches ``BUCKETS[3]`` by taking the 32 bit
1916 hash value and modulo it by ``n_buckets``. ``BUCKETS[3]`` contains "6" which
1917 is the index into the ``HASHES`` table. We would then compare any consecutive
1918 32 bit hashes values in the ``HASHES`` array as long as the hashes would be in
1919 ``BUCKETS[3]``. We do this by verifying that each subsequent hash value modulo
1920 ``n_buckets`` is still 3. In the case of a failed lookup we would access the
1921 memory for ``BUCKETS[3]``, and then compare a few consecutive 32 bit hashes
1922 before we know that we have no match. We don't end up marching through
1923 multiple words of memory and we really keep the number of processor data cache
1924 lines being accessed as small as possible.
1926 The string hash that is used for these lookup tables is the Daniel J.
1927 Bernstein hash which is also used in the ELF ``GNU_HASH`` sections. It is a
1928 very good hash for all kinds of names in programs with very few hash
1931 Empty buckets are designated by using an invalid hash index of ``UINT32_MAX``.
1936 These name hash tables are designed to be generic where specializations of the
1937 table get to define additional data that goes into the header ("``HeaderData``"),
1938 how the string value is stored ("``KeyType``") and the content of the data for each
1944 The header has a fixed part, and the specialized part. The exact format of the
1951 uint32_t magic; // 'HASH' magic value to allow endian detection
1952 uint16_t version; // Version number
1953 uint16_t hash_function; // The hash function enumeration that was used
1954 uint32_t bucket_count; // The number of buckets in this hash table
1955 uint32_t hashes_count; // The total number of unique hash values and hash data offsets in this table
1956 uint32_t header_data_len; // The bytes to skip to get to the hash indexes (buckets) for correct alignment
1957 // Specifically the length of the following HeaderData field - this does not
1958 // include the size of the preceding fields
1959 HeaderData header_data; // Implementation specific header data
1962 The header starts with a 32 bit "``magic``" value which must be ``'HASH'``
1963 encoded as an ASCII integer. This allows the detection of the start of the
1964 hash table and also allows the table's byte order to be determined so the table
1965 can be correctly extracted. The "``magic``" value is followed by a 16 bit
1966 ``version`` number which allows the table to be revised and modified in the
1967 future. The current version number is 1. ``hash_function`` is a ``uint16_t``
1968 enumeration that specifies which hash function was used to produce this table.
1969 The current values for the hash function enumerations include:
1973 enum HashFunctionType
1975 eHashFunctionDJB = 0u, // Daniel J Bernstein hash function
1978 ``bucket_count`` is a 32 bit unsigned integer that represents how many buckets
1979 are in the ``BUCKETS`` array. ``hashes_count`` is the number of unique 32 bit
1980 hash values that are in the ``HASHES`` array, and is the same number of offsets
1981 are contained in the ``OFFSETS`` array. ``header_data_len`` specifies the size
1982 in bytes of the ``HeaderData`` that is filled in by specialized versions of
1988 The header is followed by the buckets, hashes, offsets, and hash value data.
1994 uint32_t buckets[Header.bucket_count]; // An array of hash indexes into the "hashes[]" array below
1995 uint32_t hashes [Header.hashes_count]; // Every unique 32 bit hash for the entire table is in this table
1996 uint32_t offsets[Header.hashes_count]; // An offset that corresponds to each item in the "hashes[]" array above
1999 ``buckets`` is an array of 32 bit indexes into the ``hashes`` array. The
2000 ``hashes`` array contains all of the 32 bit hash values for all names in the
2001 hash table. Each hash in the ``hashes`` table has an offset in the ``offsets``
2002 array that points to the data for the hash value.
2004 This table setup makes it very easy to repurpose these tables to contain
2005 different data, while keeping the lookup mechanism the same for all tables.
2006 This layout also makes it possible to save the table to disk and map it in
2007 later and do very efficient name lookups with little or no parsing.
2009 DWARF lookup tables can be implemented in a variety of ways and can store a lot
2010 of information for each name. We want to make the DWARF tables extensible and
2011 able to store the data efficiently so we have used some of the DWARF features
2012 that enable efficient data storage to define exactly what kind of data we store
2015 The ``HeaderData`` contains a definition of the contents of each HashData chunk.
2016 We might want to store an offset to all of the debug information entries (DIEs)
2017 for each name. To keep things extensible, we create a list of items, or
2018 Atoms, that are contained in the data for each name. First comes the type of
2019 the data in each atom:
2026 eAtomTypeDIEOffset = 1u, // DIE offset, check form for encoding
2027 eAtomTypeCUOffset = 2u, // DIE offset of the compiler unit header that contains the item in question
2028 eAtomTypeTag = 3u, // DW_TAG_xxx value, should be encoded as DW_FORM_data1 (if no tags exceed 255) or DW_FORM_data2
2029 eAtomTypeNameFlags = 4u, // Flags from enum NameFlags
2030 eAtomTypeTypeFlags = 5u, // Flags from enum TypeFlags
2033 The enumeration values and their meanings are:
2035 .. code-block:: none
2037 eAtomTypeNULL - a termination atom that specifies the end of the atom list
2038 eAtomTypeDIEOffset - an offset into the .debug_info section for the DWARF DIE for this name
2039 eAtomTypeCUOffset - an offset into the .debug_info section for the CU that contains the DIE
2040 eAtomTypeDIETag - The DW_TAG_XXX enumeration value so you don't have to parse the DWARF to see what it is
2041 eAtomTypeNameFlags - Flags for functions and global variables (isFunction, isInlined, isExternal...)
2042 eAtomTypeTypeFlags - Flags for types (isCXXClass, isObjCClass, ...)
2044 Then we allow each atom type to define the atom type and how the data for each
2045 atom type data is encoded:
2051 uint16_t type; // AtomType enum value
2052 uint16_t form; // DWARF DW_FORM_XXX defines
2055 The ``form`` type above is from the DWARF specification and defines the exact
2056 encoding of the data for the Atom type. See the DWARF specification for the
2057 ``DW_FORM_`` definitions.
2063 uint32_t die_offset_base;
2064 uint32_t atom_count;
2065 Atoms atoms[atom_count0];
2068 ``HeaderData`` defines the base DIE offset that should be added to any atoms
2069 that are encoded using the ``DW_FORM_ref1``, ``DW_FORM_ref2``,
2070 ``DW_FORM_ref4``, ``DW_FORM_ref8`` or ``DW_FORM_ref_udata``. It also defines
2071 what is contained in each ``HashData`` object -- ``Atom.form`` tells us how large
2072 each field will be in the ``HashData`` and the ``Atom.type`` tells us how this data
2073 should be interpreted.
2075 For the current implementations of the "``.apple_names``" (all functions +
2076 globals), the "``.apple_types``" (names of all types that are defined), and
2077 the "``.apple_namespaces``" (all namespaces), we currently set the ``Atom``
2082 HeaderData.atom_count = 1;
2083 HeaderData.atoms[0].type = eAtomTypeDIEOffset;
2084 HeaderData.atoms[0].form = DW_FORM_data4;
2086 This defines the contents to be the DIE offset (eAtomTypeDIEOffset) that is
2087 encoded as a 32 bit value (DW_FORM_data4). This allows a single name to have
2088 multiple matching DIEs in a single file, which could come up with an inlined
2089 function for instance. Future tables could include more information about the
2090 DIE such as flags indicating if the DIE is a function, method, block,
2093 The KeyType for the DWARF table is a 32 bit string table offset into the
2094 ".debug_str" table. The ".debug_str" is the string table for the DWARF which
2095 may already contain copies of all of the strings. This helps make sure, with
2096 help from the compiler, that we reuse the strings between all of the DWARF
2097 sections and keeps the hash table size down. Another benefit to having the
2098 compiler generate all strings as DW_FORM_strp in the debug info, is that
2099 DWARF parsing can be made much faster.
2101 After a lookup is made, we get an offset into the hash data. The hash data
2102 needs to be able to deal with 32 bit hash collisions, so the chunk of data
2103 at the offset in the hash data consists of a triple:
2108 uint32_t hash_data_count
2109 HashData[hash_data_count]
2111 If "str_offset" is zero, then the bucket contents are done. 99.9% of the
2112 hash data chunks contain a single item (no 32 bit hash collision):
2114 .. code-block:: none
2117 | 0x00001023 | uint32_t KeyType (.debug_str[0x0001023] => "main")
2118 | 0x00000004 | uint32_t HashData count
2119 | 0x........ | uint32_t HashData[0] DIE offset
2120 | 0x........ | uint32_t HashData[1] DIE offset
2121 | 0x........ | uint32_t HashData[2] DIE offset
2122 | 0x........ | uint32_t HashData[3] DIE offset
2123 | 0x00000000 | uint32_t KeyType (end of hash chain)
2126 If there are collisions, you will have multiple valid string offsets:
2128 .. code-block:: none
2131 | 0x00001023 | uint32_t KeyType (.debug_str[0x0001023] => "main")
2132 | 0x00000004 | uint32_t HashData count
2133 | 0x........ | uint32_t HashData[0] DIE offset
2134 | 0x........ | uint32_t HashData[1] DIE offset
2135 | 0x........ | uint32_t HashData[2] DIE offset
2136 | 0x........ | uint32_t HashData[3] DIE offset
2137 | 0x00002023 | uint32_t KeyType (.debug_str[0x0002023] => "print")
2138 | 0x00000002 | uint32_t HashData count
2139 | 0x........ | uint32_t HashData[0] DIE offset
2140 | 0x........ | uint32_t HashData[1] DIE offset
2141 | 0x00000000 | uint32_t KeyType (end of hash chain)
2144 Current testing with real world C++ binaries has shown that there is around 1
2145 32 bit hash collision per 100,000 name entries.
2150 As we said, we want to strictly define exactly what is included in the
2151 different tables. For DWARF, we have 3 tables: "``.apple_names``",
2152 "``.apple_types``", and "``.apple_namespaces``".
2154 "``.apple_names``" sections should contain an entry for each DWARF DIE whose
2155 ``DW_TAG`` is a ``DW_TAG_label``, ``DW_TAG_inlined_subroutine``, or
2156 ``DW_TAG_subprogram`` that has address attributes: ``DW_AT_low_pc``,
2157 ``DW_AT_high_pc``, ``DW_AT_ranges`` or ``DW_AT_entry_pc``. It also contains
2158 ``DW_TAG_variable`` DIEs that have a ``DW_OP_addr`` in the location (global and
2159 static variables). All global and static variables should be included,
2160 including those scoped within functions and classes. For example using the
2172 Both of the static ``var`` variables would be included in the table. All
2173 functions should emit both their full names and their basenames. For C or C++,
2174 the full name is the mangled name (if available) which is usually in the
2175 ``DW_AT_MIPS_linkage_name`` attribute, and the ``DW_AT_name`` contains the
2176 function basename. If global or static variables have a mangled name in a
2177 ``DW_AT_MIPS_linkage_name`` attribute, this should be emitted along with the
2178 simple name found in the ``DW_AT_name`` attribute.
2180 "``.apple_types``" sections should contain an entry for each DWARF DIE whose
2185 * DW_TAG_enumeration_type
2186 * DW_TAG_pointer_type
2187 * DW_TAG_reference_type
2188 * DW_TAG_string_type
2189 * DW_TAG_structure_type
2190 * DW_TAG_subroutine_type
2193 * DW_TAG_ptr_to_member_type
2195 * DW_TAG_subrange_type
2201 * DW_TAG_packed_type
2202 * DW_TAG_volatile_type
2203 * DW_TAG_restrict_type
2204 * DW_TAG_interface_type
2205 * DW_TAG_unspecified_type
2206 * DW_TAG_shared_type
2208 Only entries with a ``DW_AT_name`` attribute are included, and the entry must
2209 not be a forward declaration (``DW_AT_declaration`` attribute with a non-zero
2210 value). For example, using the following code:
2220 We get a few type DIEs:
2222 .. code-block:: none
2224 0x00000067: TAG_base_type [5]
2225 AT_encoding( DW_ATE_signed )
2227 AT_byte_size( 0x04 )
2229 0x0000006e: TAG_pointer_type [6]
2230 AT_type( {0x00000067} ( int ) )
2231 AT_byte_size( 0x08 )
2233 The DW_TAG_pointer_type is not included because it does not have a ``DW_AT_name``.
2235 "``.apple_namespaces``" section should contain all ``DW_TAG_namespace`` DIEs.
2236 If we run into a namespace that has no name this is an anonymous namespace, and
2237 the name should be output as "``(anonymous namespace)``" (without the quotes).
2238 Why? This matches the output of the ``abi::cxa_demangle()`` that is in the
2239 standard C++ library that demangles mangled names.
2242 Language Extensions and File Format Changes
2243 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
2245 Objective-C Extensions
2246 """"""""""""""""""""""
2248 "``.apple_objc``" section should contain all ``DW_TAG_subprogram`` DIEs for an
2249 Objective-C class. The name used in the hash table is the name of the
2250 Objective-C class itself. If the Objective-C class has a category, then an
2251 entry is made for both the class name without the category, and for the class
2252 name with the category. So if we have a DIE at offset 0x1234 with a name of
2253 method "``-[NSString(my_additions) stringWithSpecialString:]``", we would add
2254 an entry for "``NSString``" that points to DIE 0x1234, and an entry for
2255 "``NSString(my_additions)``" that points to 0x1234. This allows us to quickly
2256 track down all Objective-C methods for an Objective-C class when doing
2257 expressions. It is needed because of the dynamic nature of Objective-C where
2258 anyone can add methods to a class. The DWARF for Objective-C methods is also
2259 emitted differently from C++ classes where the methods are not usually
2260 contained in the class definition, they are scattered about across one or more
2261 compile units. Categories can also be defined in different shared libraries.
2262 So we need to be able to quickly find all of the methods and class functions
2263 given the Objective-C class name, or quickly find all methods and class
2264 functions for a class + category name. This table does not contain any
2265 selector names, it just maps Objective-C class names (or class names +
2266 category) to all of the methods and class functions. The selectors are added
2267 as function basenames in the "``.debug_names``" section.
2269 In the "``.apple_names``" section for Objective-C functions, the full name is
2270 the entire function name with the brackets ("``-[NSString
2271 stringWithCString:]``") and the basename is the selector only
2272 ("``stringWithCString:``").
2277 The sections names for the apple hash tables are for non mach-o files. For
2278 mach-o files, the sections should be contained in the ``__DWARF`` segment with
2281 * "``.apple_names``" -> "``__apple_names``"
2282 * "``.apple_types``" -> "``__apple_types``"
2283 * "``.apple_namespaces``" -> "``__apple_namespac``" (16 character limit)
2284 * "``.apple_objc``" -> "``__apple_objc``"