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, there are two backend consumers of debug info: DwarfDebug and
67 CodeViewDebug. DwarfDebug produces DWARF suitable for use with GDB, LLDB, and
68 other DWARF-based debuggers. :ref:`CodeViewDebug <codeview>` produces CodeView,
69 the Microsoft debug info format, which is usable with Microsoft debuggers such
70 as Visual Studio and WinDBG. LLVM's debug information format is mostly derived
71 from and inspired by DWARF, but it is feasible to translate into other target
72 debug info formats such as STABS.
74 It would also be reasonable to use debug information to feed profiling tools
75 for analysis of generated code, or, tools for reconstructing the original
76 source from generated code.
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 debugging
96 information, allowing them to update the debugging information as they
97 perform aggressive optimizations. This means that, with effort, the LLVM
98 optimizers could optimize debug code just as well as non-debug code.
100 * LLVM debug information does not prevent optimizations from
101 happening (for example inlining, basic block reordering/merging/cleanup,
102 tail duplication, etc).
104 * LLVM debug information is automatically optimized along with the rest of
105 the program, using existing facilities. For example, duplicate
106 information is automatically merged by the linker, and unused information
107 is automatically removed.
109 Basically, the debug information allows you to compile a program with
110 "``-O0 -g``" and get full debug information, allowing you to arbitrarily modify
111 the program as it executes from a debugger. Compiling a program with
112 "``-O3 -g``" gives you full debug information that is always available and
113 accurate for reading (e.g., you get accurate stack traces despite tail call
114 elimination and inlining), but you might lose the ability to modify the program
115 and call functions which were optimized out of the program, or inlined away
118 The :ref:`LLVM test suite <test-suite-quickstart>` provides a framework to test
119 optimizer's handling of debugging information. It can be run like this:
123 % cd llvm/projects/test-suite/MultiSource/Benchmarks # or some other level
126 This will test impact of debugging information on optimization passes. If
127 debugging information influences optimization passes then it will be reported
128 as a failure. See :doc:`TestingGuide` for more information on LLVM test
129 infrastructure and how to run various tests.
133 Debugging information format
134 ============================
136 LLVM debugging information has been carefully designed to make it possible for
137 the optimizer to optimize the program and debugging information without
138 necessarily having to know anything about debugging information. In
139 particular, the use of metadata avoids duplicated debugging information from
140 the beginning, and the global dead code elimination pass automatically deletes
141 debugging information for a function if it decides to delete the function.
143 To do this, most of the debugging information (descriptors for types,
144 variables, functions, source files, etc) is inserted by the language front-end
145 in the form of LLVM metadata.
147 Debug information is designed to be agnostic about the target debugger and
148 debugging information representation (e.g. DWARF/Stabs/etc). It uses a generic
149 pass to decode the information that represents variables, types, functions,
150 namespaces, etc: this allows for arbitrary source-language semantics and
151 type-systems to be used, as long as there is a module written for the target
152 debugger to interpret the information.
154 To provide basic functionality, the LLVM debugger does have to make some
155 assumptions about the source-level language being debugged, though it keeps
156 these to a minimum. The only common features that the LLVM debugger assumes
157 exist are `source files <LangRef.html#difile>`_, and `program objects
158 <LangRef.html#diglobalvariable>`_. These abstract objects are used by a
159 debugger to form stack traces, show information about local variables, etc.
161 This section of the documentation first describes the representation aspects
162 common to any source-language. :ref:`ccxx_frontend` describes the data layout
163 conventions used by the C and C++ front-ends.
165 Debug information descriptors are `specialized metadata nodes
166 <LangRef.html#specialized-metadata>`_, first-class subclasses of ``Metadata``.
168 .. _format_common_intrinsics:
170 Debugger intrinsic functions
171 ----------------------------
173 LLVM uses several intrinsic functions (name prefixed with "``llvm.dbg``") to
174 provide debug information at various points in generated code.
181 void @llvm.dbg.declare(metadata, metadata, metadata)
183 This intrinsic provides information about a local element (e.g., variable).
184 The first argument is metadata holding the alloca for the variable. The second
185 argument is a `local variable <LangRef.html#dilocalvariable>`_ containing a
186 description of the variable. The third argument is a `complex expression
187 <LangRef.html#diexpression>`_.
194 void @llvm.dbg.value(metadata, i64, metadata, metadata)
196 This intrinsic provides information when a user source variable is set to a new
197 value. The first argument is the new value (wrapped as metadata). The second
198 argument is the offset in the user source variable where the new value is
199 written. The third argument is a `local variable
200 <LangRef.html#dilocalvariable>`_ containing a description of the variable. The
201 fourth argument is a `complex expression <LangRef.html#diexpression>`_.
203 Object lifetimes and scoping
204 ============================
206 In many languages, the local variables in functions can have their lifetimes or
207 scopes limited to a subset of a function. In the C family of languages, for
208 example, variables are only live (readable and writable) within the source
209 block that they are defined in. In functional languages, values are only
210 readable after they have been defined. Though this is a very obvious concept,
211 it is non-trivial to model in LLVM, because it has no notion of scoping in this
212 sense, and does not want to be tied to a language's scoping rules.
214 In order to handle this, the LLVM debug format uses the metadata attached to
215 llvm instructions to encode line number and scoping information. Consider the
216 following C fragment, for example:
230 Compiled to LLVM, this function would be represented like this:
234 ; Function Attrs: nounwind ssp uwtable
235 define void @foo() #0 !dbg !4 {
237 %X = alloca i32, align 4
238 %Y = alloca i32, align 4
239 %Z = alloca i32, align 4
240 call void @llvm.dbg.declare(metadata i32* %X, metadata !11, metadata !13), !dbg !14
241 store i32 21, i32* %X, align 4, !dbg !14
242 call void @llvm.dbg.declare(metadata i32* %Y, metadata !15, metadata !13), !dbg !16
243 store i32 22, i32* %Y, align 4, !dbg !16
244 call void @llvm.dbg.declare(metadata i32* %Z, metadata !17, metadata !13), !dbg !19
245 store i32 23, i32* %Z, align 4, !dbg !19
246 %0 = load i32, i32* %X, align 4, !dbg !20
247 store i32 %0, i32* %Z, align 4, !dbg !21
248 %1 = load i32, i32* %Y, align 4, !dbg !22
249 store i32 %1, i32* %X, align 4, !dbg !23
253 ; Function Attrs: nounwind readnone
254 declare void @llvm.dbg.declare(metadata, metadata, metadata) #1
256 attributes #0 = { nounwind ssp uwtable "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
257 attributes #1 = { nounwind readnone }
260 !llvm.module.flags = !{!7, !8, !9}
263 !0 = !DICompileUnit(language: DW_LANG_C99, file: !1, producer: "clang version 3.7.0 (trunk 231150) (llvm/trunk 231154)", isOptimized: false, runtimeVersion: 0, emissionKind: FullDebug, enums: !2, retainedTypes: !2, subprograms: !3, globals: !2, imports: !2)
264 !1 = !DIFile(filename: "/dev/stdin", directory: "/Users/dexonsmith/data/llvm/debug-info")
267 !4 = distinct !DISubprogram(name: "foo", scope: !1, file: !1, line: 1, type: !5, isLocal: false, isDefinition: true, scopeLine: 1, isOptimized: false, variables: !2)
268 !5 = !DISubroutineType(types: !6)
270 !7 = !{i32 2, !"Dwarf Version", i32 2}
271 !8 = !{i32 2, !"Debug Info Version", i32 3}
272 !9 = !{i32 1, !"PIC Level", i32 2}
273 !10 = !{!"clang version 3.7.0 (trunk 231150) (llvm/trunk 231154)"}
274 !11 = !DILocalVariable(name: "X", scope: !4, file: !1, line: 2, type: !12)
275 !12 = !DIBasicType(name: "int", size: 32, align: 32, encoding: DW_ATE_signed)
276 !13 = !DIExpression()
277 !14 = !DILocation(line: 2, column: 9, scope: !4)
278 !15 = !DILocalVariable(name: "Y", scope: !4, file: !1, line: 3, type: !12)
279 !16 = !DILocation(line: 3, column: 9, scope: !4)
280 !17 = !DILocalVariable(name: "Z", scope: !18, file: !1, line: 5, type: !12)
281 !18 = distinct !DILexicalBlock(scope: !4, file: !1, line: 4, column: 5)
282 !19 = !DILocation(line: 5, column: 11, scope: !18)
283 !20 = !DILocation(line: 6, column: 11, scope: !18)
284 !21 = !DILocation(line: 6, column: 9, scope: !18)
285 !22 = !DILocation(line: 8, column: 9, scope: !4)
286 !23 = !DILocation(line: 8, column: 7, scope: !4)
287 !24 = !DILocation(line: 9, column: 3, scope: !4)
290 This example illustrates a few important details about LLVM debugging
291 information. In particular, it shows how the ``llvm.dbg.declare`` intrinsic and
292 location information, which are attached to an instruction, are applied
293 together to allow a debugger to analyze the relationship between statements,
294 variable definitions, and the code used to implement the function.
298 call void @llvm.dbg.declare(metadata i32* %X, metadata !11, metadata !13), !dbg !14
299 ; [debug line = 2:7] [debug variable = X]
301 The first intrinsic ``%llvm.dbg.declare`` encodes debugging information for the
302 variable ``X``. The metadata ``!dbg !14`` attached to the intrinsic provides
303 scope information for the variable ``X``.
307 !14 = !DILocation(line: 2, column: 9, scope: !4)
308 !4 = distinct !DISubprogram(name: "foo", scope: !1, file: !1, line: 1, type: !5,
309 isLocal: false, isDefinition: true, scopeLine: 1,
310 isOptimized: false, variables: !2)
312 Here ``!14`` is metadata providing `location information
313 <LangRef.html#dilocation>`_. In this example, scope is encoded by ``!4``, a
314 `subprogram descriptor <LangRef.html#disubprogram>`_. This way the location
315 information attached to the intrinsics indicates that the variable ``X`` is
316 declared at line number 2 at a function level scope in function ``foo``.
318 Now lets take another example.
322 call void @llvm.dbg.declare(metadata i32* %Z, metadata !17, metadata !13), !dbg !19
323 ; [debug line = 5:9] [debug variable = Z]
325 The third intrinsic ``%llvm.dbg.declare`` encodes debugging information for
326 variable ``Z``. The metadata ``!dbg !19`` attached to the intrinsic provides
327 scope information for the variable ``Z``.
331 !18 = distinct !DILexicalBlock(scope: !4, file: !1, line: 4, column: 5)
332 !19 = !DILocation(line: 5, column: 11, scope: !18)
334 Here ``!19`` indicates that ``Z`` is declared at line number 5 and column
335 number 0 inside of lexical scope ``!18``. The lexical scope itself resides
336 inside of subprogram ``!4`` described above.
338 The scope information attached with each instruction provides a straightforward
339 way to find instructions covered by a scope.
343 C/C++ front-end specific debug information
344 ==========================================
346 The C and C++ front-ends represent information about the program in a format
347 that is effectively identical to `DWARF 3.0
348 <http://www.eagercon.com/dwarf/dwarf3std.htm>`_ in terms of information
349 content. This allows code generators to trivially support native debuggers by
350 generating standard dwarf information, and contains enough information for
351 non-dwarf targets to translate it as needed.
353 This section describes the forms used to represent C and C++ programs. Other
354 languages could pattern themselves after this (which itself is tuned to
355 representing programs in the same way that DWARF 3 does), or they could choose
356 to provide completely different forms if they don't fit into the DWARF model.
357 As support for debugging information gets added to the various LLVM
358 source-language front-ends, the information used should be documented here.
360 The following sections provide examples of a few C/C++ constructs and the debug
361 information that would best describe those constructs. The canonical
362 references are the ``DIDescriptor`` classes defined in
363 ``include/llvm/IR/DebugInfo.h`` and the implementations of the helper functions
364 in ``lib/IR/DIBuilder.cpp``.
366 C/C++ source file information
367 -----------------------------
369 ``llvm::Instruction`` provides easy access to metadata attached with an
370 instruction. One can extract line number information encoded in LLVM IR using
371 ``Instruction::getDebugLoc()`` and ``DILocation::getLine()``.
375 if (DILocation *Loc = I->getDebugLoc()) { // Here I is an LLVM instruction
376 unsigned Line = Loc->getLine();
377 StringRef File = Loc->getFilename();
378 StringRef Dir = Loc->getDirectory();
381 C/C++ global variable information
382 ---------------------------------
384 Given an integer global variable declared as follows:
388 _Alignas(8) int MyGlobal = 100;
390 a C/C++ front-end would generate the following descriptors:
395 ;; Define the global itself.
397 @MyGlobal = global i32 100, align 8, !dbg !0
400 ;; List of debug info of globals
404 ;; Some unrelated metadata.
405 !llvm.module.flags = !{!6, !7}
408 ;; Define the global variable itself
409 !0 = distinct !DIGlobalVariable(name: "MyGlobal", scope: !1, file: !2, line: 1, type: !5, isLocal: false, isDefinition: true, align: 64)
411 ;; Define the compile unit.
412 !1 = distinct !DICompileUnit(language: DW_LANG_C99, file: !2,
413 producer: "clang version 4.0.0 (http://llvm.org/git/clang.git ae4deadbea242e8ea517eef662c30443f75bd086) (http://llvm.org/git/llvm.git 818b4c1539df3e51dc7e62c89ead4abfd348827d)",
414 isOptimized: false, runtimeVersion: 0, emissionKind: FullDebug,
415 enums: !3, globals: !4)
420 !2 = !DIFile(filename: "/dev/stdin",
421 directory: "/Users/dexonsmith/data/llvm/debug-info")
426 ;; The Array of Global Variables
432 !5 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed)
434 ;; Dwarf version to output.
435 !6 = !{i32 2, !"Dwarf Version", i32 4}
437 ;; Debug info schema version.
438 !7 = !{i32 2, !"Debug Info Version", i32 3}
440 ;; Compiler identification
441 !8 = !{!"clang version 4.0.0 (http://llvm.org/git/clang.git ae4deadbea242e8ea517eef662c30443f75bd086) (http://llvm.org/git/llvm.git 818b4c1539df3e51dc7e62c89ead4abfd348827d)"}
444 The align value in DIGlobalVariable description specifies variable alignment in
445 case it was forced by C11 _Alignas(), C++11 alignas() keywords or compiler
446 attribute __attribute__((aligned ())). In other case (when this field is missing)
447 alignment is considered default. This is used when producing DWARF output
448 for DW_AT_alignment value.
450 C/C++ function information
451 --------------------------
453 Given a function declared as follows:
457 int main(int argc, char *argv[]) {
461 a C/C++ front-end would generate the following descriptors:
466 ;; Define the anchor for subprograms.
468 !4 = !DISubprogram(name: "main", scope: !1, file: !1, line: 1, type: !5,
469 isLocal: false, isDefinition: true, scopeLine: 1,
470 flags: DIFlagPrototyped, isOptimized: false,
474 ;; Define the subprogram itself.
476 define i32 @main(i32 %argc, i8** %argv) !dbg !4 {
480 Debugging information format
481 ============================
483 Debugging Information Extension for Objective C Properties
484 ----------------------------------------------------------
489 Objective C provides a simpler way to declare and define accessor methods using
490 declared properties. The language provides features to declare a property and
491 to let compiler synthesize accessor methods.
493 The debugger lets developer inspect Objective C interfaces and their instance
494 variables and class variables. However, the debugger does not know anything
495 about the properties defined in Objective C interfaces. The debugger consumes
496 information generated by compiler in DWARF format. The format does not support
497 encoding of Objective C properties. This proposal describes DWARF extensions to
498 encode Objective C properties, which the debugger can use to let developers
499 inspect Objective C properties.
504 Objective C properties exist separately from class members. A property can be
505 defined only by "setter" and "getter" selectors, and be calculated anew on each
506 access. Or a property can just be a direct access to some declared ivar.
507 Finally it can have an ivar "automatically synthesized" for it by the compiler,
508 in which case the property can be referred to in user code directly using the
509 standard C dereference syntax as well as through the property "dot" syntax, but
510 there is no entry in the ``@interface`` declaration corresponding to this ivar.
512 To facilitate debugging, these properties we will add a new DWARF TAG into the
513 ``DW_TAG_structure_type`` definition for the class to hold the description of a
514 given property, and a set of DWARF attributes that provide said description.
515 The property tag will also contain the name and declared type of the property.
517 If there is a related ivar, there will also be a DWARF property attribute placed
518 in the ``DW_TAG_member`` DIE for that ivar referring back to the property TAG
519 for that property. And in the case where the compiler synthesizes the ivar
520 directly, the compiler is expected to generate a ``DW_TAG_member`` for that
521 ivar (with the ``DW_AT_artificial`` set to 1), whose name will be the name used
522 to access this ivar directly in code, and with the property attribute pointing
523 back to the property it is backing.
525 The following examples will serve as illustration for our discussion:
542 This produces the following DWARF (this is a "pseudo dwarfdump" output):
546 0x00000100: TAG_structure_type [7] *
547 AT_APPLE_runtime_class( 0x10 )
549 AT_decl_file( "Objc_Property.m" )
552 0x00000110 TAG_APPLE_property
554 AT_type ( {0x00000150} ( int ) )
556 0x00000120: TAG_APPLE_property
558 AT_type ( {0x00000150} ( int ) )
560 0x00000130: TAG_member [8]
562 AT_APPLE_property ( {0x00000110} "p1" )
563 AT_type( {0x00000150} ( int ) )
564 AT_artificial ( 0x1 )
566 0x00000140: TAG_member [8]
568 AT_APPLE_property ( {0x00000120} "p2" )
569 AT_type( {0x00000150} ( int ) )
571 0x00000150: AT_type( ( int ) )
573 Note, the current convention is that the name of the ivar for an
574 auto-synthesized property is the name of the property from which it derives
575 with an underscore prepended, as is shown in the example. But we actually
576 don't need to know this convention, since we are given the name of the ivar
579 Also, it is common practice in ObjC to have different property declarations in
580 the @interface and @implementation - e.g. to provide a read-only property in
581 the interface,and a read-write interface in the implementation. In that case,
582 the compiler should emit whichever property declaration will be in force in the
583 current translation unit.
585 Developers can decorate a property with attributes which are encoded using
586 ``DW_AT_APPLE_property_attribute``.
590 @property (readonly, nonatomic) int pr;
594 TAG_APPLE_property [8]
596 AT_type ( {0x00000147} (int) )
597 AT_APPLE_property_attribute (DW_APPLE_PROPERTY_readonly, DW_APPLE_PROPERTY_nonatomic)
599 The setter and getter method names are attached to the property using
600 ``DW_AT_APPLE_property_setter`` and ``DW_AT_APPLE_property_getter`` attributes.
605 @property (setter=myOwnP3Setter:) int p3;
606 -(void)myOwnP3Setter:(int)a;
611 -(void)myOwnP3Setter:(int)a{ }
614 The DWARF for this would be:
618 0x000003bd: TAG_structure_type [7] *
619 AT_APPLE_runtime_class( 0x10 )
621 AT_decl_file( "Objc_Property.m" )
624 0x000003cd TAG_APPLE_property
626 AT_APPLE_property_setter ( "myOwnP3Setter:" )
627 AT_type( {0x00000147} ( int ) )
629 0x000003f3: TAG_member [8]
631 AT_type ( {0x00000147} ( int ) )
632 AT_APPLE_property ( {0x000003cd} )
633 AT_artificial ( 0x1 )
638 +-----------------------+--------+
640 +=======================+========+
641 | DW_TAG_APPLE_property | 0x4200 |
642 +-----------------------+--------+
647 +--------------------------------+--------+-----------+
648 | Attribute | Value | Classes |
649 +================================+========+===========+
650 | DW_AT_APPLE_property | 0x3fed | Reference |
651 +--------------------------------+--------+-----------+
652 | DW_AT_APPLE_property_getter | 0x3fe9 | String |
653 +--------------------------------+--------+-----------+
654 | DW_AT_APPLE_property_setter | 0x3fea | String |
655 +--------------------------------+--------+-----------+
656 | DW_AT_APPLE_property_attribute | 0x3feb | Constant |
657 +--------------------------------+--------+-----------+
662 +--------------------------------------+-------+
664 +======================================+=======+
665 | DW_APPLE_PROPERTY_readonly | 0x01 |
666 +--------------------------------------+-------+
667 | DW_APPLE_PROPERTY_getter | 0x02 |
668 +--------------------------------------+-------+
669 | DW_APPLE_PROPERTY_assign | 0x04 |
670 +--------------------------------------+-------+
671 | DW_APPLE_PROPERTY_readwrite | 0x08 |
672 +--------------------------------------+-------+
673 | DW_APPLE_PROPERTY_retain | 0x10 |
674 +--------------------------------------+-------+
675 | DW_APPLE_PROPERTY_copy | 0x20 |
676 +--------------------------------------+-------+
677 | DW_APPLE_PROPERTY_nonatomic | 0x40 |
678 +--------------------------------------+-------+
679 | DW_APPLE_PROPERTY_setter | 0x80 |
680 +--------------------------------------+-------+
681 | DW_APPLE_PROPERTY_atomic | 0x100 |
682 +--------------------------------------+-------+
683 | DW_APPLE_PROPERTY_weak | 0x200 |
684 +--------------------------------------+-------+
685 | DW_APPLE_PROPERTY_strong | 0x400 |
686 +--------------------------------------+-------+
687 | DW_APPLE_PROPERTY_unsafe_unretained | 0x800 |
688 +--------------------------------------+-------+
689 | DW_APPLE_PROPERTY_nullability | 0x1000|
690 +--------------------------------------+-------+
691 | DW_APPLE_PROPERTY_null_resettable | 0x2000|
692 +--------------------------------------+-------+
693 | DW_APPLE_PROPERTY_class | 0x4000|
694 +--------------------------------------+-------+
696 Name Accelerator Tables
697 -----------------------
702 The "``.debug_pubnames``" and "``.debug_pubtypes``" formats are not what a
703 debugger needs. The "``pub``" in the section name indicates that the entries
704 in the table are publicly visible names only. This means no static or hidden
705 functions show up in the "``.debug_pubnames``". No static variables or private
706 class variables are in the "``.debug_pubtypes``". Many compilers add different
707 things to these tables, so we can't rely upon the contents between gcc, icc, or
710 The typical query given by users tends not to match up with the contents of
711 these tables. For example, the DWARF spec states that "In the case of the name
712 of a function member or static data member of a C++ structure, class or union,
713 the name presented in the "``.debug_pubnames``" section is not the simple name
714 given by the ``DW_AT_name attribute`` of the referenced debugging information
715 entry, but rather the fully qualified name of the data or function member."
716 So the only names in these tables for complex C++ entries is a fully
717 qualified name. Debugger users tend not to enter their search strings as
718 "``a::b::c(int,const Foo&) const``", but rather as "``c``", "``b::c``" , or
719 "``a::b::c``". So the name entered in the name table must be demangled in
720 order to chop it up appropriately and additional names must be manually entered
721 into the table to make it effective as a name lookup table for debuggers to
724 All debuggers currently ignore the "``.debug_pubnames``" table as a result of
725 its inconsistent and useless public-only name content making it a waste of
726 space in the object file. These tables, when they are written to disk, are not
727 sorted in any way, leaving every debugger to do its own parsing and sorting.
728 These tables also include an inlined copy of the string values in the table
729 itself making the tables much larger than they need to be on disk, especially
730 for large C++ programs.
732 Can't we just fix the sections by adding all of the names we need to this
733 table? No, because that is not what the tables are defined to contain and we
734 won't know the difference between the old bad tables and the new good tables.
735 At best we could make our own renamed sections that contain all of the data we
738 These tables are also insufficient for what a debugger like LLDB needs. LLDB
739 uses clang for its expression parsing where LLDB acts as a PCH. LLDB is then
740 often asked to look for type "``foo``" or namespace "``bar``", or list items in
741 namespace "``baz``". Namespaces are not included in the pubnames or pubtypes
742 tables. Since clang asks a lot of questions when it is parsing an expression,
743 we need to be very fast when looking up names, as it happens a lot. Having new
744 accelerator tables that are optimized for very quick lookups will benefit this
745 type of debugging experience greatly.
747 We would like to generate name lookup tables that can be mapped into memory
748 from disk, and used as is, with little or no up-front parsing. We would also
749 be able to control the exact content of these different tables so they contain
750 exactly what we need. The Name Accelerator Tables were designed to fix these
751 issues. In order to solve these issues we need to:
753 * Have a format that can be mapped into memory from disk and used as is
754 * Lookups should be very fast
755 * Extensible table format so these tables can be made by many producers
756 * Contain all of the names needed for typical lookups out of the box
757 * Strict rules for the contents of tables
759 Table size is important and the accelerator table format should allow the reuse
760 of strings from common string tables so the strings for the names are not
761 duplicated. We also want to make sure the table is ready to be used as-is by
762 simply mapping the table into memory with minimal header parsing.
764 The name lookups need to be fast and optimized for the kinds of lookups that
765 debuggers tend to do. Optimally we would like to touch as few parts of the
766 mapped table as possible when doing a name lookup and be able to quickly find
767 the name entry we are looking for, or discover there are no matches. In the
768 case of debuggers we optimized for lookups that fail most of the time.
770 Each table that is defined should have strict rules on exactly what is in the
771 accelerator tables and documented so clients can rely on the content.
779 Typical hash tables have a header, buckets, and each bucket points to the
792 The BUCKETS are an array of offsets to DATA for each hash:
797 | 0x00001000 | BUCKETS[0]
798 | 0x00002000 | BUCKETS[1]
799 | 0x00002200 | BUCKETS[2]
800 | 0x000034f0 | BUCKETS[3]
802 | 0xXXXXXXXX | BUCKETS[n_buckets]
805 So for ``bucket[3]`` in the example above, we have an offset into the table
806 0x000034f0 which points to a chain of entries for the bucket. Each bucket must
807 contain a next pointer, full 32 bit hash value, the string itself, and the data
808 for the current string value.
813 0x000034f0: | 0x00003500 | next pointer
814 | 0x12345678 | 32 bit hash
815 | "erase" | string value
816 | data[n] | HashData for this bucket
818 0x00003500: | 0x00003550 | next pointer
819 | 0x29273623 | 32 bit hash
820 | "dump" | string value
821 | data[n] | HashData for this bucket
823 0x00003550: | 0x00000000 | next pointer
824 | 0x82638293 | 32 bit hash
825 | "main" | string value
826 | data[n] | HashData for this bucket
829 The problem with this layout for debuggers is that we need to optimize for the
830 negative lookup case where the symbol we're searching for is not present. So
831 if we were to lookup "``printf``" in the table above, we would make a 32-bit
832 hash for "``printf``", it might match ``bucket[3]``. We would need to go to
833 the offset 0x000034f0 and start looking to see if our 32 bit hash matches. To
834 do so, we need to read the next pointer, then read the hash, compare it, and
835 skip to the next bucket. Each time we are skipping many bytes in memory and
836 touching new pages just to do the compare on the full 32 bit hash. All of
837 these accesses then tell us that we didn't have a match.
842 To solve the issues mentioned above we have structured the hash tables a bit
843 differently: a header, buckets, an array of all unique 32 bit hash values,
844 followed by an array of hash value data offsets, one for each hash value, then
845 the data for all hash values:
861 The ``BUCKETS`` in the name tables are an index into the ``HASHES`` array. By
862 making all of the full 32 bit hash values contiguous in memory, we allow
863 ourselves to efficiently check for a match while touching as little memory as
864 possible. Most often checking the 32 bit hash values is as far as the lookup
865 goes. If it does match, it usually is a match with no collisions. So for a
866 table with "``n_buckets``" buckets, and "``n_hashes``" unique 32 bit hash
867 values, we can clarify the contents of the ``BUCKETS``, ``HASHES`` and
872 .-------------------------.
873 | HEADER.magic | uint32_t
874 | HEADER.version | uint16_t
875 | HEADER.hash_function | uint16_t
876 | HEADER.bucket_count | uint32_t
877 | HEADER.hashes_count | uint32_t
878 | HEADER.header_data_len | uint32_t
879 | HEADER_DATA | HeaderData
880 |-------------------------|
881 | BUCKETS | uint32_t[n_buckets] // 32 bit hash indexes
882 |-------------------------|
883 | HASHES | uint32_t[n_hashes] // 32 bit hash values
884 |-------------------------|
885 | OFFSETS | uint32_t[n_hashes] // 32 bit offsets to hash value data
886 |-------------------------|
888 `-------------------------'
890 So taking the exact same data from the standard hash example above we end up
903 | ... | BUCKETS[n_buckets]
905 | 0x........ | HASHES[0]
906 | 0x........ | HASHES[1]
907 | 0x........ | HASHES[2]
908 | 0x........ | HASHES[3]
909 | 0x........ | HASHES[4]
910 | 0x........ | HASHES[5]
911 | 0x12345678 | HASHES[6] hash for BUCKETS[3]
912 | 0x29273623 | HASHES[7] hash for BUCKETS[3]
913 | 0x82638293 | HASHES[8] hash for BUCKETS[3]
914 | 0x........ | HASHES[9]
915 | 0x........ | HASHES[10]
916 | 0x........ | HASHES[11]
917 | 0x........ | HASHES[12]
918 | 0x........ | HASHES[13]
919 | 0x........ | HASHES[n_hashes]
921 | 0x........ | OFFSETS[0]
922 | 0x........ | OFFSETS[1]
923 | 0x........ | OFFSETS[2]
924 | 0x........ | OFFSETS[3]
925 | 0x........ | OFFSETS[4]
926 | 0x........ | OFFSETS[5]
927 | 0x000034f0 | OFFSETS[6] offset for BUCKETS[3]
928 | 0x00003500 | OFFSETS[7] offset for BUCKETS[3]
929 | 0x00003550 | OFFSETS[8] offset for BUCKETS[3]
930 | 0x........ | OFFSETS[9]
931 | 0x........ | OFFSETS[10]
932 | 0x........ | OFFSETS[11]
933 | 0x........ | OFFSETS[12]
934 | 0x........ | OFFSETS[13]
935 | 0x........ | OFFSETS[n_hashes]
943 0x000034f0: | 0x00001203 | .debug_str ("erase")
944 | 0x00000004 | A 32 bit array count - number of HashData with name "erase"
945 | 0x........ | HashData[0]
946 | 0x........ | HashData[1]
947 | 0x........ | HashData[2]
948 | 0x........ | HashData[3]
949 | 0x00000000 | String offset into .debug_str (terminate data for hash)
951 0x00003500: | 0x00001203 | String offset into .debug_str ("collision")
952 | 0x00000002 | A 32 bit array count - number of HashData with name "collision"
953 | 0x........ | HashData[0]
954 | 0x........ | HashData[1]
955 | 0x00001203 | String offset into .debug_str ("dump")
956 | 0x00000003 | A 32 bit array count - number of HashData with name "dump"
957 | 0x........ | HashData[0]
958 | 0x........ | HashData[1]
959 | 0x........ | HashData[2]
960 | 0x00000000 | String offset into .debug_str (terminate data for hash)
962 0x00003550: | 0x00001203 | String offset into .debug_str ("main")
963 | 0x00000009 | A 32 bit array count - number of HashData with name "main"
964 | 0x........ | HashData[0]
965 | 0x........ | HashData[1]
966 | 0x........ | HashData[2]
967 | 0x........ | HashData[3]
968 | 0x........ | HashData[4]
969 | 0x........ | HashData[5]
970 | 0x........ | HashData[6]
971 | 0x........ | HashData[7]
972 | 0x........ | HashData[8]
973 | 0x00000000 | String offset into .debug_str (terminate data for hash)
976 So we still have all of the same data, we just organize it more efficiently for
977 debugger lookup. If we repeat the same "``printf``" lookup from above, we
978 would hash "``printf``" and find it matches ``BUCKETS[3]`` by taking the 32 bit
979 hash value and modulo it by ``n_buckets``. ``BUCKETS[3]`` contains "6" which
980 is the index into the ``HASHES`` table. We would then compare any consecutive
981 32 bit hashes values in the ``HASHES`` array as long as the hashes would be in
982 ``BUCKETS[3]``. We do this by verifying that each subsequent hash value modulo
983 ``n_buckets`` is still 3. In the case of a failed lookup we would access the
984 memory for ``BUCKETS[3]``, and then compare a few consecutive 32 bit hashes
985 before we know that we have no match. We don't end up marching through
986 multiple words of memory and we really keep the number of processor data cache
987 lines being accessed as small as possible.
989 The string hash that is used for these lookup tables is the Daniel J.
990 Bernstein hash which is also used in the ELF ``GNU_HASH`` sections. It is a
991 very good hash for all kinds of names in programs with very few hash
994 Empty buckets are designated by using an invalid hash index of ``UINT32_MAX``.
999 These name hash tables are designed to be generic where specializations of the
1000 table get to define additional data that goes into the header ("``HeaderData``"),
1001 how the string value is stored ("``KeyType``") and the content of the data for each
1007 The header has a fixed part, and the specialized part. The exact format of the
1014 uint32_t magic; // 'HASH' magic value to allow endian detection
1015 uint16_t version; // Version number
1016 uint16_t hash_function; // The hash function enumeration that was used
1017 uint32_t bucket_count; // The number of buckets in this hash table
1018 uint32_t hashes_count; // The total number of unique hash values and hash data offsets in this table
1019 uint32_t header_data_len; // The bytes to skip to get to the hash indexes (buckets) for correct alignment
1020 // Specifically the length of the following HeaderData field - this does not
1021 // include the size of the preceding fields
1022 HeaderData header_data; // Implementation specific header data
1025 The header starts with a 32 bit "``magic``" value which must be ``'HASH'``
1026 encoded as an ASCII integer. This allows the detection of the start of the
1027 hash table and also allows the table's byte order to be determined so the table
1028 can be correctly extracted. The "``magic``" value is followed by a 16 bit
1029 ``version`` number which allows the table to be revised and modified in the
1030 future. The current version number is 1. ``hash_function`` is a ``uint16_t``
1031 enumeration that specifies which hash function was used to produce this table.
1032 The current values for the hash function enumerations include:
1036 enum HashFunctionType
1038 eHashFunctionDJB = 0u, // Daniel J Bernstein hash function
1041 ``bucket_count`` is a 32 bit unsigned integer that represents how many buckets
1042 are in the ``BUCKETS`` array. ``hashes_count`` is the number of unique 32 bit
1043 hash values that are in the ``HASHES`` array, and is the same number of offsets
1044 are contained in the ``OFFSETS`` array. ``header_data_len`` specifies the size
1045 in bytes of the ``HeaderData`` that is filled in by specialized versions of
1051 The header is followed by the buckets, hashes, offsets, and hash value data.
1057 uint32_t buckets[Header.bucket_count]; // An array of hash indexes into the "hashes[]" array below
1058 uint32_t hashes [Header.hashes_count]; // Every unique 32 bit hash for the entire table is in this table
1059 uint32_t offsets[Header.hashes_count]; // An offset that corresponds to each item in the "hashes[]" array above
1062 ``buckets`` is an array of 32 bit indexes into the ``hashes`` array. The
1063 ``hashes`` array contains all of the 32 bit hash values for all names in the
1064 hash table. Each hash in the ``hashes`` table has an offset in the ``offsets``
1065 array that points to the data for the hash value.
1067 This table setup makes it very easy to repurpose these tables to contain
1068 different data, while keeping the lookup mechanism the same for all tables.
1069 This layout also makes it possible to save the table to disk and map it in
1070 later and do very efficient name lookups with little or no parsing.
1072 DWARF lookup tables can be implemented in a variety of ways and can store a lot
1073 of information for each name. We want to make the DWARF tables extensible and
1074 able to store the data efficiently so we have used some of the DWARF features
1075 that enable efficient data storage to define exactly what kind of data we store
1078 The ``HeaderData`` contains a definition of the contents of each HashData chunk.
1079 We might want to store an offset to all of the debug information entries (DIEs)
1080 for each name. To keep things extensible, we create a list of items, or
1081 Atoms, that are contained in the data for each name. First comes the type of
1082 the data in each atom:
1089 eAtomTypeDIEOffset = 1u, // DIE offset, check form for encoding
1090 eAtomTypeCUOffset = 2u, // DIE offset of the compiler unit header that contains the item in question
1091 eAtomTypeTag = 3u, // DW_TAG_xxx value, should be encoded as DW_FORM_data1 (if no tags exceed 255) or DW_FORM_data2
1092 eAtomTypeNameFlags = 4u, // Flags from enum NameFlags
1093 eAtomTypeTypeFlags = 5u, // Flags from enum TypeFlags
1096 The enumeration values and their meanings are:
1098 .. code-block:: none
1100 eAtomTypeNULL - a termination atom that specifies the end of the atom list
1101 eAtomTypeDIEOffset - an offset into the .debug_info section for the DWARF DIE for this name
1102 eAtomTypeCUOffset - an offset into the .debug_info section for the CU that contains the DIE
1103 eAtomTypeDIETag - The DW_TAG_XXX enumeration value so you don't have to parse the DWARF to see what it is
1104 eAtomTypeNameFlags - Flags for functions and global variables (isFunction, isInlined, isExternal...)
1105 eAtomTypeTypeFlags - Flags for types (isCXXClass, isObjCClass, ...)
1107 Then we allow each atom type to define the atom type and how the data for each
1108 atom type data is encoded:
1114 uint16_t type; // AtomType enum value
1115 uint16_t form; // DWARF DW_FORM_XXX defines
1118 The ``form`` type above is from the DWARF specification and defines the exact
1119 encoding of the data for the Atom type. See the DWARF specification for the
1120 ``DW_FORM_`` definitions.
1126 uint32_t die_offset_base;
1127 uint32_t atom_count;
1128 Atoms atoms[atom_count0];
1131 ``HeaderData`` defines the base DIE offset that should be added to any atoms
1132 that are encoded using the ``DW_FORM_ref1``, ``DW_FORM_ref2``,
1133 ``DW_FORM_ref4``, ``DW_FORM_ref8`` or ``DW_FORM_ref_udata``. It also defines
1134 what is contained in each ``HashData`` object -- ``Atom.form`` tells us how large
1135 each field will be in the ``HashData`` and the ``Atom.type`` tells us how this data
1136 should be interpreted.
1138 For the current implementations of the "``.apple_names``" (all functions +
1139 globals), the "``.apple_types``" (names of all types that are defined), and
1140 the "``.apple_namespaces``" (all namespaces), we currently set the ``Atom``
1145 HeaderData.atom_count = 1;
1146 HeaderData.atoms[0].type = eAtomTypeDIEOffset;
1147 HeaderData.atoms[0].form = DW_FORM_data4;
1149 This defines the contents to be the DIE offset (eAtomTypeDIEOffset) that is
1150 encoded as a 32 bit value (DW_FORM_data4). This allows a single name to have
1151 multiple matching DIEs in a single file, which could come up with an inlined
1152 function for instance. Future tables could include more information about the
1153 DIE such as flags indicating if the DIE is a function, method, block,
1156 The KeyType for the DWARF table is a 32 bit string table offset into the
1157 ".debug_str" table. The ".debug_str" is the string table for the DWARF which
1158 may already contain copies of all of the strings. This helps make sure, with
1159 help from the compiler, that we reuse the strings between all of the DWARF
1160 sections and keeps the hash table size down. Another benefit to having the
1161 compiler generate all strings as DW_FORM_strp in the debug info, is that
1162 DWARF parsing can be made much faster.
1164 After a lookup is made, we get an offset into the hash data. The hash data
1165 needs to be able to deal with 32 bit hash collisions, so the chunk of data
1166 at the offset in the hash data consists of a triple:
1171 uint32_t hash_data_count
1172 HashData[hash_data_count]
1174 If "str_offset" is zero, then the bucket contents are done. 99.9% of the
1175 hash data chunks contain a single item (no 32 bit hash collision):
1177 .. code-block:: none
1180 | 0x00001023 | uint32_t KeyType (.debug_str[0x0001023] => "main")
1181 | 0x00000004 | uint32_t HashData count
1182 | 0x........ | uint32_t HashData[0] DIE offset
1183 | 0x........ | uint32_t HashData[1] DIE offset
1184 | 0x........ | uint32_t HashData[2] DIE offset
1185 | 0x........ | uint32_t HashData[3] DIE offset
1186 | 0x00000000 | uint32_t KeyType (end of hash chain)
1189 If there are collisions, you will have multiple valid string offsets:
1191 .. code-block:: none
1194 | 0x00001023 | uint32_t KeyType (.debug_str[0x0001023] => "main")
1195 | 0x00000004 | uint32_t HashData count
1196 | 0x........ | uint32_t HashData[0] DIE offset
1197 | 0x........ | uint32_t HashData[1] DIE offset
1198 | 0x........ | uint32_t HashData[2] DIE offset
1199 | 0x........ | uint32_t HashData[3] DIE offset
1200 | 0x00002023 | uint32_t KeyType (.debug_str[0x0002023] => "print")
1201 | 0x00000002 | uint32_t HashData count
1202 | 0x........ | uint32_t HashData[0] DIE offset
1203 | 0x........ | uint32_t HashData[1] DIE offset
1204 | 0x00000000 | uint32_t KeyType (end of hash chain)
1207 Current testing with real world C++ binaries has shown that there is around 1
1208 32 bit hash collision per 100,000 name entries.
1213 As we said, we want to strictly define exactly what is included in the
1214 different tables. For DWARF, we have 3 tables: "``.apple_names``",
1215 "``.apple_types``", and "``.apple_namespaces``".
1217 "``.apple_names``" sections should contain an entry for each DWARF DIE whose
1218 ``DW_TAG`` is a ``DW_TAG_label``, ``DW_TAG_inlined_subroutine``, or
1219 ``DW_TAG_subprogram`` that has address attributes: ``DW_AT_low_pc``,
1220 ``DW_AT_high_pc``, ``DW_AT_ranges`` or ``DW_AT_entry_pc``. It also contains
1221 ``DW_TAG_variable`` DIEs that have a ``DW_OP_addr`` in the location (global and
1222 static variables). All global and static variables should be included,
1223 including those scoped within functions and classes. For example using the
1235 Both of the static ``var`` variables would be included in the table. All
1236 functions should emit both their full names and their basenames. For C or C++,
1237 the full name is the mangled name (if available) which is usually in the
1238 ``DW_AT_MIPS_linkage_name`` attribute, and the ``DW_AT_name`` contains the
1239 function basename. If global or static variables have a mangled name in a
1240 ``DW_AT_MIPS_linkage_name`` attribute, this should be emitted along with the
1241 simple name found in the ``DW_AT_name`` attribute.
1243 "``.apple_types``" sections should contain an entry for each DWARF DIE whose
1248 * DW_TAG_enumeration_type
1249 * DW_TAG_pointer_type
1250 * DW_TAG_reference_type
1251 * DW_TAG_string_type
1252 * DW_TAG_structure_type
1253 * DW_TAG_subroutine_type
1256 * DW_TAG_ptr_to_member_type
1258 * DW_TAG_subrange_type
1263 * DW_TAG_packed_type
1264 * DW_TAG_volatile_type
1265 * DW_TAG_restrict_type
1266 * DW_TAG_atomic_type
1267 * DW_TAG_interface_type
1268 * DW_TAG_unspecified_type
1269 * DW_TAG_shared_type
1271 Only entries with a ``DW_AT_name`` attribute are included, and the entry must
1272 not be a forward declaration (``DW_AT_declaration`` attribute with a non-zero
1273 value). For example, using the following code:
1283 We get a few type DIEs:
1285 .. code-block:: none
1287 0x00000067: TAG_base_type [5]
1288 AT_encoding( DW_ATE_signed )
1290 AT_byte_size( 0x04 )
1292 0x0000006e: TAG_pointer_type [6]
1293 AT_type( {0x00000067} ( int ) )
1294 AT_byte_size( 0x08 )
1296 The DW_TAG_pointer_type is not included because it does not have a ``DW_AT_name``.
1298 "``.apple_namespaces``" section should contain all ``DW_TAG_namespace`` DIEs.
1299 If we run into a namespace that has no name this is an anonymous namespace, and
1300 the name should be output as "``(anonymous namespace)``" (without the quotes).
1301 Why? This matches the output of the ``abi::cxa_demangle()`` that is in the
1302 standard C++ library that demangles mangled names.
1305 Language Extensions and File Format Changes
1306 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1308 Objective-C Extensions
1309 """"""""""""""""""""""
1311 "``.apple_objc``" section should contain all ``DW_TAG_subprogram`` DIEs for an
1312 Objective-C class. The name used in the hash table is the name of the
1313 Objective-C class itself. If the Objective-C class has a category, then an
1314 entry is made for both the class name without the category, and for the class
1315 name with the category. So if we have a DIE at offset 0x1234 with a name of
1316 method "``-[NSString(my_additions) stringWithSpecialString:]``", we would add
1317 an entry for "``NSString``" that points to DIE 0x1234, and an entry for
1318 "``NSString(my_additions)``" that points to 0x1234. This allows us to quickly
1319 track down all Objective-C methods for an Objective-C class when doing
1320 expressions. It is needed because of the dynamic nature of Objective-C where
1321 anyone can add methods to a class. The DWARF for Objective-C methods is also
1322 emitted differently from C++ classes where the methods are not usually
1323 contained in the class definition, they are scattered about across one or more
1324 compile units. Categories can also be defined in different shared libraries.
1325 So we need to be able to quickly find all of the methods and class functions
1326 given the Objective-C class name, or quickly find all methods and class
1327 functions for a class + category name. This table does not contain any
1328 selector names, it just maps Objective-C class names (or class names +
1329 category) to all of the methods and class functions. The selectors are added
1330 as function basenames in the "``.debug_names``" section.
1332 In the "``.apple_names``" section for Objective-C functions, the full name is
1333 the entire function name with the brackets ("``-[NSString
1334 stringWithCString:]``") and the basename is the selector only
1335 ("``stringWithCString:``").
1340 The sections names for the apple hash tables are for non-mach-o files. For
1341 mach-o files, the sections should be contained in the ``__DWARF`` segment with
1344 * "``.apple_names``" -> "``__apple_names``"
1345 * "``.apple_types``" -> "``__apple_types``"
1346 * "``.apple_namespaces``" -> "``__apple_namespac``" (16 character limit)
1347 * "``.apple_objc``" -> "``__apple_objc``"
1351 CodeView Debug Info Format
1352 ==========================
1354 LLVM supports emitting CodeView, the Microsoft debug info format, and this
1355 section describes the design and implementation of that support.
1360 CodeView as a format is clearly oriented around C++ debugging, and in C++, the
1361 majority of debug information tends to be type information. Therefore, the
1362 overriding design constraint of CodeView is the separation of type information
1363 from other "symbol" information so that type information can be efficiently
1364 merged across translation units. Both type information and symbol information is
1365 generally stored as a sequence of records, where each record begins with a
1366 16-bit record size and a 16-bit record kind.
1368 Type information is usually stored in the ``.debug$T`` section of the object
1369 file. All other debug info, such as line info, string table, symbol info, and
1370 inlinee info, is stored in one or more ``.debug$S`` sections. There may only be
1371 one ``.debug$T`` section per object file, since all other debug info refers to
1372 it. If a PDB (enabled by the ``/Zi`` MSVC option) was used during compilation,
1373 the ``.debug$T`` section will contain only an ``LF_TYPESERVER2`` record pointing
1374 to the PDB. When using PDBs, symbol information appears to remain in the object
1375 file ``.debug$S`` sections.
1377 Type records are referred to by their index, which is the number of records in
1378 the stream before a given record plus ``0x1000``. Many common basic types, such
1379 as the basic integral types and unqualified pointers to them, are represented
1380 using type indices less than ``0x1000``. Such basic types are built in to
1381 CodeView consumers and do not require type records.
1383 Each type record may only contain type indices that are less than its own type
1384 index. This ensures that the graph of type stream references is acyclic. While
1385 the source-level type graph may contain cycles through pointer types (consider a
1386 linked list struct), these cycles are removed from the type stream by always
1387 referring to the forward declaration record of user-defined record types. Only
1388 "symbol" records in the ``.debug$S`` streams may refer to complete,
1389 non-forward-declaration type records.
1391 Working with CodeView
1392 ---------------------
1394 These are instructions for some common tasks for developers working to improve
1395 LLVM's CodeView support. Most of them revolve around using the CodeView dumper
1396 embedded in ``llvm-readobj``.
1398 * Testing MSVC's output::
1400 $ cl -c -Z7 foo.cpp # Use /Z7 to keep types in the object file
1401 $ llvm-readobj -codeview foo.obj
1403 * Getting LLVM IR debug info out of Clang::
1405 $ clang -g -gcodeview --target=x86_64-windows-msvc foo.cpp -S -emit-llvm
1407 Use this to generate LLVM IR for LLVM test cases.
1409 * Generate and dump CodeView from LLVM IR metadata::
1411 $ llc foo.ll -filetype=obj -o foo.obj
1412 $ llvm-readobj -codeview foo.obj > foo.txt
1414 Use this pattern in lit test cases and FileCheck the output of llvm-readobj
1416 Improving LLVM's CodeView support is a process of finding interesting type
1417 records, constructing a C++ test case that makes MSVC emit those records,
1418 dumping the records, understanding them, and then generating equivalent records