1================================
2Source Level Debugging with LLVM
3================================
4
5.. contents::
6   :local:
7
8Introduction
9============
10
11This document is the central repository for all information pertaining to debug
12information in LLVM.  It describes the :ref:`actual format that the LLVM debug
13information takes <format>`, which is useful for those interested in creating
14front-ends or dealing directly with the information.  Further, this document
15provides specific examples of what debug information for C/C++ looks like.
16
17Philosophy behind LLVM debugging information
18--------------------------------------------
19
20The idea of the LLVM debugging information is to capture how the important
21pieces of the source-language's Abstract Syntax Tree map onto LLVM code.
22Several design aspects have shaped the solution that appears here.  The
23important ones are:
24
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.
28
29* LLVM optimizations should interact in :ref:`well-defined and easily described
30  ways <intro_debugopt>` with the debugging information.
31
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.
35
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.
39
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.
44
45The approach used by the LLVM implementation is to use a small set of
46:ref:`intrinsic functions <format_common_intrinsics>` to define a mapping
47between LLVM program objects and the source-level objects.  The description of
48the source-level program is maintained in LLVM metadata in an
49:ref:`implementation-defined format <ccxx_frontend>` (the C/C++ front-end
50currently uses working draft 7 of the `DWARF 3 standard
51<http://www.eagercon.com/dwarf/dwarf3std.htm>`_).
52
53When a program is being debugged, a debugger interacts with the user and turns
54the stored debug information into source-language specific information.  As
55such, a debugger must be aware of the source-language, and is thus tied to a
56specific language or family of languages.
57
58Debug information consumers
59---------------------------
60
61The role of debug information is to provide meta information normally stripped
62away during the compilation process.  This meta information provides an LLVM
63user a relationship between generated code and the original program source
64code.
65
66Currently, there are two backend consumers of debug info: DwarfDebug and
67CodeViewDebug. DwarfDebug produces DWARF suitable for use with GDB, LLDB, and
68other DWARF-based debuggers. :ref:`CodeViewDebug <codeview>` produces CodeView,
69the Microsoft debug info format, which is usable with Microsoft debuggers such
70as Visual Studio and WinDBG. LLVM's debug information format is mostly derived
71from and inspired by DWARF, but it is feasible to translate into other target
72debug info formats such as STABS.
73
74It would also be reasonable to use debug information to feed profiling tools
75for analysis of generated code, or, tools for reconstructing the original
76source from generated code.
77
78.. _intro_debugopt:
79
80Debug information and optimizations
81-----------------------------------
82
83An extremely high priority of LLVM debugging information is to make it interact
84well with optimizations and analysis.  In particular, the LLVM debug
85information provides the following guarantees:
86
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. :doc:`HowToUpdateDebugInfo` specifies how debug
90  info should be updated in various kinds of code transformations to avoid
91  breaking this guarantee, and how to preserve as much useful debug info as
92  possible.  Note that some optimizations may impact the ability to modify the
93  current state of the program with a debugger, such as setting program
94  variables, or calling functions that have been deleted.
95
96* As desired, LLVM optimizations can be upgraded to be aware of debugging
97  information, allowing them to update the debugging information as they
98  perform aggressive optimizations.  This means that, with effort, the LLVM
99  optimizers could optimize debug code just as well as non-debug code.
100
101* LLVM debug information does not prevent optimizations from
102  happening (for example inlining, basic block reordering/merging/cleanup,
103  tail duplication, etc).
104
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.
109
110Basically, the debug information allows you to compile a program with
111"``-O0 -g``" and get full debug information, allowing you to arbitrarily modify
112the program as it executes from a debugger.  Compiling a program with
113"``-O3 -g``" gives you full debug information that is always available and
114accurate for reading (e.g., you get accurate stack traces despite tail call
115elimination and inlining), but you might lose the ability to modify the program
116and call functions which were optimized out of the program, or inlined away
117completely.
118
119The :doc:`LLVM test-suite <TestSuiteMakefileGuide>` provides a framework to
120test the optimizer's handling of debugging information.  It can be run like
121this:
122
123.. code-block:: bash
124
125  % cd llvm/projects/test-suite/MultiSource/Benchmarks  # or some other level
126  % make TEST=dbgopt
127
128This will test impact of debugging information on optimization passes.  If
129debugging information influences optimization passes then it will be reported
130as a failure.  See :doc:`TestingGuide` for more information on LLVM test
131infrastructure and how to run various tests.
132
133.. _format:
134
135Debugging information format
136============================
137
138LLVM debugging information has been carefully designed to make it possible for
139the optimizer to optimize the program and debugging information without
140necessarily having to know anything about debugging information.  In
141particular, the use of metadata avoids duplicated debugging information from
142the beginning, and the global dead code elimination pass automatically deletes
143debugging information for a function if it decides to delete the function.
144
145To do this, most of the debugging information (descriptors for types,
146variables, functions, source files, etc) is inserted by the language front-end
147in the form of LLVM metadata.
148
149Debug information is designed to be agnostic about the target debugger and
150debugging information representation (e.g. DWARF/Stabs/etc).  It uses a generic
151pass to decode the information that represents variables, types, functions,
152namespaces, etc: this allows for arbitrary source-language semantics and
153type-systems to be used, as long as there is a module written for the target
154debugger to interpret the information.
155
156To provide basic functionality, the LLVM debugger does have to make some
157assumptions about the source-level language being debugged, though it keeps
158these to a minimum.  The only common features that the LLVM debugger assumes
159exist are `source files <LangRef.html#difile>`_, and `program objects
160<LangRef.html#diglobalvariable>`_.  These abstract objects are used by a
161debugger to form stack traces, show information about local variables, etc.
162
163This section of the documentation first describes the representation aspects
164common to any source-language.  :ref:`ccxx_frontend` describes the data layout
165conventions used by the C and C++ front-ends.
166
167Debug information descriptors are `specialized metadata nodes
168<LangRef.html#specialized-metadata>`_, first-class subclasses of ``Metadata``.
169
170.. _format_common_intrinsics:
171
172Debugger intrinsic functions
173----------------------------
174
175LLVM uses several intrinsic functions (name prefixed with "``llvm.dbg``") to
176track source local variables through optimization and code generation.
177
178``llvm.dbg.addr``
179^^^^^^^^^^^^^^^^^^^^
180
181.. code-block:: llvm
182
183  void @llvm.dbg.addr(metadata, metadata, metadata)
184
185This intrinsic provides information about a local element (e.g., variable).
186The first argument is metadata holding the address of variable, typically a
187static alloca in the function entry block.  The second argument is a
188`local variable <LangRef.html#dilocalvariable>`_ containing a description of
189the variable.  The third argument is a `complex expression
190<LangRef.html#diexpression>`_.  An `llvm.dbg.addr` intrinsic describes the
191*address* of a source variable.
192
193.. code-block:: text
194
195    %i.addr = alloca i32, align 4
196    call void @llvm.dbg.addr(metadata i32* %i.addr, metadata !1,
197                             metadata !DIExpression()), !dbg !2
198    !1 = !DILocalVariable(name: "i", ...) ; int i
199    !2 = !DILocation(...)
200    ...
201    %buffer = alloca [256 x i8], align 8
202    ; The address of i is buffer+64.
203    call void @llvm.dbg.addr(metadata [256 x i8]* %buffer, metadata !3,
204                             metadata !DIExpression(DW_OP_plus, 64)), !dbg !4
205    !3 = !DILocalVariable(name: "i", ...) ; int i
206    !4 = !DILocation(...)
207
208A frontend should generate exactly one call to ``llvm.dbg.addr`` at the point
209of declaration of a source variable. Optimization passes that fully promote the
210variable from memory to SSA values will replace this call with possibly
211multiple calls to `llvm.dbg.value`. Passes that delete stores are effectively
212partial promotion, and they will insert a mix of calls to ``llvm.dbg.value``
213and ``llvm.dbg.addr`` to track the source variable value when it is available.
214After optimization, there may be multiple calls to ``llvm.dbg.addr`` describing
215the program points where the variables lives in memory. All calls for the same
216concrete source variable must agree on the memory location.
217
218
219``llvm.dbg.declare``
220^^^^^^^^^^^^^^^^^^^^
221
222.. code-block:: llvm
223
224  void @llvm.dbg.declare(metadata, metadata, metadata)
225
226This intrinsic is identical to `llvm.dbg.addr`, except that there can only be
227one call to `llvm.dbg.declare` for a given concrete `local variable
228<LangRef.html#dilocalvariable>`_. It is not control-dependent, meaning that if
229a call to `llvm.dbg.declare` exists and has a valid location argument, that
230address is considered to be the true home of the variable across its entire
231lifetime. This makes it hard for optimizations to preserve accurate debug info
232in the presence of ``llvm.dbg.declare``, so we are transitioning away from it,
233and we plan to deprecate it in future LLVM releases.
234
235
236``llvm.dbg.value``
237^^^^^^^^^^^^^^^^^^
238
239.. code-block:: llvm
240
241  void @llvm.dbg.value(metadata, metadata, metadata)
242
243This intrinsic provides information when a user source variable is set to a new
244value.  The first argument is the new value (wrapped as metadata).  The second
245argument is a `local variable <LangRef.html#dilocalvariable>`_ containing a
246description of the variable.  The third argument is a `complex expression
247<LangRef.html#diexpression>`_.
248
249An `llvm.dbg.value` intrinsic describes the *value* of a source variable
250directly, not its address.  Note that the value operand of this intrinsic may
251be indirect (i.e, a pointer to the source variable), provided that interpreting
252the complex expression derives the direct value.
253
254Object lifetimes and scoping
255============================
256
257In many languages, the local variables in functions can have their lifetimes or
258scopes limited to a subset of a function.  In the C family of languages, for
259example, variables are only live (readable and writable) within the source
260block that they are defined in.  In functional languages, values are only
261readable after they have been defined.  Though this is a very obvious concept,
262it is non-trivial to model in LLVM, because it has no notion of scoping in this
263sense, and does not want to be tied to a language's scoping rules.
264
265In order to handle this, the LLVM debug format uses the metadata attached to
266llvm instructions to encode line number and scoping information.  Consider the
267following C fragment, for example:
268
269.. code-block:: c
270
271  1.  void foo() {
272  2.    int X = 21;
273  3.    int Y = 22;
274  4.    {
275  5.      int Z = 23;
276  6.      Z = X;
277  7.    }
278  8.    X = Y;
279  9.  }
280
281.. FIXME: Update the following example to use llvm.dbg.addr once that is the
282   default in clang.
283
284Compiled to LLVM, this function would be represented like this:
285
286.. code-block:: text
287
288  ; Function Attrs: nounwind ssp uwtable
289  define void @foo() #0 !dbg !4 {
290  entry:
291    %X = alloca i32, align 4
292    %Y = alloca i32, align 4
293    %Z = alloca i32, align 4
294    call void @llvm.dbg.declare(metadata i32* %X, metadata !11, metadata !DIExpression()), !dbg !13
295    store i32 21, i32* %X, align 4, !dbg !13
296    call void @llvm.dbg.declare(metadata i32* %Y, metadata !14, metadata !DIExpression()), !dbg !15
297    store i32 22, i32* %Y, align 4, !dbg !15
298    call void @llvm.dbg.declare(metadata i32* %Z, metadata !16, metadata !DIExpression()), !dbg !18
299    store i32 23, i32* %Z, align 4, !dbg !18
300    %0 = load i32, i32* %X, align 4, !dbg !19
301    store i32 %0, i32* %Z, align 4, !dbg !20
302    %1 = load i32, i32* %Y, align 4, !dbg !21
303    store i32 %1, i32* %X, align 4, !dbg !22
304    ret void, !dbg !23
305  }
306
307  ; Function Attrs: nounwind readnone
308  declare void @llvm.dbg.declare(metadata, metadata, metadata) #1
309
310  attributes #0 = { nounwind ssp uwtable "less-precise-fpmad"="false" "frame-pointer"="all" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" }
311  attributes #1 = { nounwind readnone }
312
313  !llvm.dbg.cu = !{!0}
314  !llvm.module.flags = !{!7, !8, !9}
315  !llvm.ident = !{!10}
316
317  !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)
318  !1 = !DIFile(filename: "/dev/stdin", directory: "/Users/dexonsmith/data/llvm/debug-info")
319  !2 = !{}
320  !3 = !{!4}
321  !4 = distinct !DISubprogram(name: "foo", scope: !1, file: !1, line: 1, type: !5, isLocal: false, isDefinition: true, scopeLine: 1, isOptimized: false, variables: !2)
322  !5 = !DISubroutineType(types: !6)
323  !6 = !{null}
324  !7 = !{i32 2, !"Dwarf Version", i32 2}
325  !8 = !{i32 2, !"Debug Info Version", i32 3}
326  !9 = !{i32 1, !"PIC Level", i32 2}
327  !10 = !{!"clang version 3.7.0 (trunk 231150) (llvm/trunk 231154)"}
328  !11 = !DILocalVariable(name: "X", scope: !4, file: !1, line: 2, type: !12)
329  !12 = !DIBasicType(name: "int", size: 32, align: 32, encoding: DW_ATE_signed)
330  !13 = !DILocation(line: 2, column: 9, scope: !4)
331  !14 = !DILocalVariable(name: "Y", scope: !4, file: !1, line: 3, type: !12)
332  !15 = !DILocation(line: 3, column: 9, scope: !4)
333  !16 = !DILocalVariable(name: "Z", scope: !17, file: !1, line: 5, type: !12)
334  !17 = distinct !DILexicalBlock(scope: !4, file: !1, line: 4, column: 5)
335  !18 = !DILocation(line: 5, column: 11, scope: !17)
336  !19 = !DILocation(line: 6, column: 11, scope: !17)
337  !20 = !DILocation(line: 6, column: 9, scope: !17)
338  !21 = !DILocation(line: 8, column: 9, scope: !4)
339  !22 = !DILocation(line: 8, column: 7, scope: !4)
340  !23 = !DILocation(line: 9, column: 3, scope: !4)
341
342
343This example illustrates a few important details about LLVM debugging
344information.  In particular, it shows how the ``llvm.dbg.declare`` intrinsic and
345location information, which are attached to an instruction, are applied
346together to allow a debugger to analyze the relationship between statements,
347variable definitions, and the code used to implement the function.
348
349.. code-block:: llvm
350
351  call void @llvm.dbg.declare(metadata i32* %X, metadata !11, metadata !DIExpression()), !dbg !13
352    ; [debug line = 2:7] [debug variable = X]
353
354The first intrinsic ``%llvm.dbg.declare`` encodes debugging information for the
355variable ``X``.  The metadata ``!dbg !13`` attached to the intrinsic provides
356scope information for the variable ``X``.
357
358.. code-block:: text
359
360  !13 = !DILocation(line: 2, column: 9, scope: !4)
361  !4 = distinct !DISubprogram(name: "foo", scope: !1, file: !1, line: 1, type: !5,
362                              isLocal: false, isDefinition: true, scopeLine: 1,
363                              isOptimized: false, variables: !2)
364
365Here ``!13`` is metadata providing `location information
366<LangRef.html#dilocation>`_.  In this example, scope is encoded by ``!4``, a
367`subprogram descriptor <LangRef.html#disubprogram>`_.  This way the location
368information attached to the intrinsics indicates that the variable ``X`` is
369declared at line number 2 at a function level scope in function ``foo``.
370
371Now lets take another example.
372
373.. code-block:: llvm
374
375  call void @llvm.dbg.declare(metadata i32* %Z, metadata !16, metadata !DIExpression()), !dbg !18
376    ; [debug line = 5:9] [debug variable = Z]
377
378The third intrinsic ``%llvm.dbg.declare`` encodes debugging information for
379variable ``Z``.  The metadata ``!dbg !18`` attached to the intrinsic provides
380scope information for the variable ``Z``.
381
382.. code-block:: text
383
384  !17 = distinct !DILexicalBlock(scope: !4, file: !1, line: 4, column: 5)
385  !18 = !DILocation(line: 5, column: 11, scope: !17)
386
387Here ``!18`` indicates that ``Z`` is declared at line number 5 and column
388number 11 inside of lexical scope ``!17``.  The lexical scope itself resides
389inside of subprogram ``!4`` described above.
390
391The scope information attached with each instruction provides a straightforward
392way to find instructions covered by a scope.
393
394Object lifetime in optimized code
395=================================
396
397In the example above, every variable assignment uniquely corresponds to a
398memory store to the variable's position on the stack. However in heavily
399optimized code LLVM promotes most variables into SSA values, which can
400eventually be placed in physical registers or memory locations. To track SSA
401values through compilation, when objects are promoted to SSA values an
402``llvm.dbg.value`` intrinsic is created for each assignment, recording the
403variable's new location. Compared with the ``llvm.dbg.declare`` intrinsic:
404
405* A dbg.value terminates the effect of any preceding dbg.values for (any
406  overlapping fragments of) the specified variable.
407* The dbg.value's position in the IR defines where in the instruction stream
408  the variable's value changes.
409* Operands can be constants, indicating the variable is assigned a
410  constant value.
411
412Care must be taken to update ``llvm.dbg.value`` intrinsics when optimization
413passes alter or move instructions and blocks -- the developer could observe such
414changes reflected in the value of variables when debugging the program. For any
415execution of the optimized program, the set of variable values presented to the
416developer by the debugger should not show a state that would never have existed
417in the execution of the unoptimized program, given the same input. Doing so
418risks misleading the developer by reporting a state that does not exist,
419damaging their understanding of the optimized program and undermining their
420trust in the debugger.
421
422Sometimes perfectly preserving variable locations is not possible, often when a
423redundant calculation is optimized out. In such cases, a ``llvm.dbg.value``
424with operand ``undef`` should be used, to terminate earlier variable locations
425and let the debugger present ``optimized out`` to the developer. Withholding
426these potentially stale variable values from the developer diminishes the
427amount of available debug information, but increases the reliability of the
428remaining information.
429
430To illustrate some potential issues, consider the following example:
431
432.. code-block:: llvm
433
434  define i32 @foo(i32 %bar, i1 %cond) {
435  entry:
436    call @llvm.dbg.value(metadata i32 0, metadata !1, metadata !2)
437    br i1 %cond, label %truebr, label %falsebr
438  truebr:
439    %tval = add i32 %bar, 1
440    call @llvm.dbg.value(metadata i32 %tval, metadata !1, metadata !2)
441    %g1 = call i32 @gazonk()
442    br label %exit
443  falsebr:
444    %fval = add i32 %bar, 2
445    call @llvm.dbg.value(metadata i32 %fval, metadata !1, metadata !2)
446    %g2 = call i32 @gazonk()
447    br label %exit
448  exit:
449    %merge = phi [ %tval, %truebr ], [ %fval, %falsebr ]
450    %g = phi [ %g1, %truebr ], [ %g2, %falsebr ]
451    call @llvm.dbg.value(metadata i32 %merge, metadata !1, metadata !2)
452    call @llvm.dbg.value(metadata i32 %g, metadata !3, metadata !2)
453    %plusten = add i32 %merge, 10
454    %toret = add i32 %plusten, %g
455    call @llvm.dbg.value(metadata i32 %toret, metadata !1, metadata !2)
456    ret i32 %toret
457  }
458
459Containing two source-level variables in ``!1`` and ``!3``. The function could,
460perhaps, be optimized into the following code:
461
462.. code-block:: llvm
463
464  define i32 @foo(i32 %bar, i1 %cond) {
465  entry:
466    %g = call i32 @gazonk()
467    %addoper = select i1 %cond, i32 11, i32 12
468    %plusten = add i32 %bar, %addoper
469    %toret = add i32 %plusten, %g
470    ret i32 %toret
471  }
472
473What ``llvm.dbg.value`` intrinsics should be placed to represent the original variable
474locations in this code? Unfortunately the second, third and fourth
475dbg.values for ``!1`` in the source function have had their operands
476(%tval, %fval, %merge) optimized out. Assuming we cannot recover them, we
477might consider this placement of dbg.values:
478
479.. code-block:: llvm
480
481  define i32 @foo(i32 %bar, i1 %cond) {
482  entry:
483    call @llvm.dbg.value(metadata i32 0, metadata !1, metadata !2)
484    %g = call i32 @gazonk()
485    call @llvm.dbg.value(metadata i32 %g, metadata !3, metadata !2)
486    %addoper = select i1 %cond, i32 11, i32 12
487    %plusten = add i32 %bar, %addoper
488    %toret = add i32 %plusten, %g
489    call @llvm.dbg.value(metadata i32 %toret, metadata !1, metadata !2)
490    ret i32 %toret
491  }
492
493However, this will cause ``!3`` to have the return value of ``@gazonk()`` at
494the same time as ``!1`` has the constant value zero -- a pair of assignments
495that never occurred in the unoptimized program. To avoid this, we must terminate
496the range that ``!1`` has the constant value assignment by inserting an undef
497dbg.value before the dbg.value for ``!3``:
498
499.. code-block:: llvm
500
501  define i32 @foo(i32 %bar, i1 %cond) {
502  entry:
503    call @llvm.dbg.value(metadata i32 0, metadata !1, metadata !2)
504    %g = call i32 @gazonk()
505    call @llvm.dbg.value(metadata i32 undef, metadata !1, metadata !2)
506    call @llvm.dbg.value(metadata i32 %g, metadata !3, metadata !2)
507    %addoper = select i1 %cond, i32 11, i32 12
508    %plusten = add i32 %bar, %addoper
509    %toret = add i32 %plusten, %g
510    call @llvm.dbg.value(metadata i32 %toret, metadata !1, metadata !2)
511    ret i32 %toret
512  }
513
514In general, if any dbg.value has its operand optimized out and cannot be
515recovered, then an undef dbg.value is necessary to terminate earlier variable
516locations. Additional undef dbg.values may be necessary when the debugger can
517observe re-ordering of assignments.
518
519How variable location metadata is transformed during CodeGen
520============================================================
521
522LLVM preserves debug information throughout mid-level and backend passes,
523ultimately producing a mapping between source-level information and
524instruction ranges. This
525is relatively straightforwards for line number information, as mapping
526instructions to line numbers is a simple association. For variable locations
527however the story is more complex. As each ``llvm.dbg.value`` intrinsic
528represents a source-level assignment of a value to a source variable, the
529variable location intrinsics effectively embed a small imperative program
530within the LLVM IR. By the end of CodeGen, this becomes a mapping from each
531variable to their machine locations over ranges of instructions.
532From IR to object emission, the major transformations which affect variable
533location fidelity are:
534
5351. Instruction Selection
5362. Register allocation
5373. Block layout
538
539each of which are discussed below. In addition, instruction scheduling can
540significantly change the ordering of the program, and occurs in a number of
541different passes.
542
543Some variable locations are not transformed during CodeGen. Stack locations
544specified by ``llvm.dbg.declare`` are valid and unchanging for the entire
545duration of the function, and are recorded in a simple MachineFunction table.
546Location changes in the prologue and epilogue of a function are also ignored:
547frame setup and destruction may take several instructions, require a
548disproportionate amount of debugging information in the output binary to
549describe, and should be stepped over by debuggers anyway.
550
551Variable locations in Instruction Selection and MIR
552---------------------------------------------------
553
554Instruction selection creates a MIR function from an IR function, and just as
555it transforms ``intermediate`` instructions into machine instructions, so must
556``intermediate`` variable locations become machine variable locations.
557Within IR, variable locations are always identified by a Value, but in MIR
558there can be different types of variable locations. In addition, some IR
559locations become unavailable, for example if the operation of multiple IR
560instructions are combined into one machine instruction (such as
561multiply-and-accumulate) then intermediate Values are lost. To track variable
562locations through instruction selection, they are first separated into
563locations that do not depend on code generation (constants, stack locations,
564allocated virtual registers) and those that do. For those that do, debug
565metadata is attached to SDNodes in SelectionDAGs. After instruction selection
566has occurred and a MIR function is created, if the SDNode associated with debug
567metadata is allocated a virtual register, that virtual register is used as the
568variable location. If the SDNode is folded into a machine instruction or
569otherwise transformed into a non-register, the variable location becomes
570unavailable.
571
572Locations that are unavailable are treated as if they have been optimized out:
573in IR the location would be assigned ``undef`` by a debug intrinsic, and in MIR
574the equivalent location is used.
575
576After MIR locations are assigned to each variable, machine pseudo-instructions
577corresponding to each ``llvm.dbg.value`` and ``llvm.dbg.addr`` intrinsic are
578inserted. There are two forms of this type of instruction.
579
580The first form, ``DBG_VALUE``, appears thus:
581
582.. code-block:: text
583
584  DBG_VALUE %1, $noreg, !123, !DIExpression()
585
586And has the following operands:
587 * The first operand can record the variable location as a register,
588   a frame index, an immediate, or the base address register if the original
589   debug intrinsic referred to memory. ``$noreg`` indicates the variable
590   location is undefined, equivalent to an ``undef`` dbg.value operand.
591 * The type of the second operand indicates whether the variable location is
592   directly referred to by the DBG_VALUE, or whether it is indirect. The
593   ``$noreg`` register signifies the former, an immediate operand (0) the
594   latter.
595 * Operand 3 is the Variable field of the original debug intrinsic.
596 * Operand 4 is the Expression field of the original debug intrinsic.
597
598The second form, ``DBG_VALUE_LIST``, appears thus:
599
600.. code-block:: text
601
602  DBG_VALUE_LIST !123, !DIExpression(DW_OP_LLVM_arg, 0, DW_OP_LLVM_arg, 1, DW_OP_plus), %1, %2
603
604And has the following operands:
605 * The first operand is the Variable field of the original debug intrinsic.
606 * The second operand is the Expression field of the original debug intrinsic.
607 * Any number of operands, from the 3rd onwards, record a sequence of variable
608   location operands, which may take any of the same values as the first
609   operand of the ``DBG_VALUE`` instruction above. These variable location
610   operands are inserted into the final DWARF Expression in positions indicated
611   by the DW_OP_LLVM_arg operator in the `DIExpression
612   <LangRef.html#diexpression>`.
613
614The position at which the DBG_VALUEs are inserted should correspond to the
615positions of their matching ``llvm.dbg.value`` intrinsics in the IR block.  As
616with optimization, LLVM aims to preserve the order in which variable
617assignments occurred in the source program. However SelectionDAG performs some
618instruction scheduling, which can reorder assignments (discussed below).
619Function parameter locations are moved to the beginning of the function if
620they're not already, to ensure they're immediately available on function entry.
621
622To demonstrate variable locations during instruction selection, consider
623the following example:
624
625.. code-block:: llvm
626
627  define i32 @foo(i32* %addr) {
628  entry:
629    call void @llvm.dbg.value(metadata i32 0, metadata !3, metadata !DIExpression()), !dbg !5
630    br label %bb1, !dbg !5
631
632  bb1:                                              ; preds = %bb1, %entry
633    %bar.0 = phi i32 [ 0, %entry ], [ %add, %bb1 ]
634    call void @llvm.dbg.value(metadata i32 %bar.0, metadata !3, metadata !DIExpression()), !dbg !5
635    %addr1 = getelementptr i32, i32 *%addr, i32 1, !dbg !5
636    call void @llvm.dbg.value(metadata i32 *%addr1, metadata !3, metadata !DIExpression()), !dbg !5
637    %loaded1 = load i32, i32* %addr1, !dbg !5
638    %addr2 = getelementptr i32, i32 *%addr, i32 %bar.0, !dbg !5
639    call void @llvm.dbg.value(metadata i32 *%addr2, metadata !3, metadata !DIExpression()), !dbg !5
640    %loaded2 = load i32, i32* %addr2, !dbg !5
641    %add = add i32 %bar.0, 1, !dbg !5
642    call void @llvm.dbg.value(metadata i32 %add, metadata !3, metadata !DIExpression()), !dbg !5
643    %added = add i32 %loaded1, %loaded2
644    %cond = icmp ult i32 %added, %bar.0, !dbg !5
645    br i1 %cond, label %bb1, label %bb2, !dbg !5
646
647  bb2:                                              ; preds = %bb1
648    ret i32 0, !dbg !5
649  }
650
651If one compiles this IR with ``llc -o - -start-after=codegen-prepare -stop-after=expand-isel-pseudos -mtriple=x86_64--``, the following MIR is produced:
652
653.. code-block:: text
654
655  bb.0.entry:
656    successors: %bb.1(0x80000000)
657    liveins: $rdi
658
659    %2:gr64 = COPY $rdi
660    %3:gr32 = MOV32r0 implicit-def dead $eflags
661    DBG_VALUE 0, $noreg, !3, !DIExpression(), debug-location !5
662
663  bb.1.bb1:
664    successors: %bb.1(0x7c000000), %bb.2(0x04000000)
665
666    %0:gr32 = PHI %3, %bb.0, %1, %bb.1
667    DBG_VALUE %0, $noreg, !3, !DIExpression(), debug-location !5
668    DBG_VALUE %2, $noreg, !3, !DIExpression(DW_OP_plus_uconst, 4, DW_OP_stack_value), debug-location !5
669    %4:gr32 = MOV32rm %2, 1, $noreg, 4, $noreg, debug-location !5 :: (load 4 from %ir.addr1)
670    %5:gr64_nosp = MOVSX64rr32 %0, debug-location !5
671    DBG_VALUE $noreg, $noreg, !3, !DIExpression(), debug-location !5
672    %1:gr32 = INC32r %0, implicit-def dead $eflags, debug-location !5
673    DBG_VALUE %1, $noreg, !3, !DIExpression(), debug-location !5
674    %6:gr32 = ADD32rm %4, %2, 4, killed %5, 0, $noreg, implicit-def dead $eflags :: (load 4 from %ir.addr2)
675    %7:gr32 = SUB32rr %6, %0, implicit-def $eflags, debug-location !5
676    JB_1 %bb.1, implicit $eflags, debug-location !5
677    JMP_1 %bb.2, debug-location !5
678
679  bb.2.bb2:
680    %8:gr32 = MOV32r0 implicit-def dead $eflags
681    $eax = COPY %8, debug-location !5
682    RET 0, $eax, debug-location !5
683
684Observe first that there is a DBG_VALUE instruction for every ``llvm.dbg.value``
685intrinsic in the source IR, ensuring no source level assignments go missing.
686Then consider the different ways in which variable locations have been recorded:
687
688* For the first dbg.value an immediate operand is used to record a zero value.
689* The dbg.value of the PHI instruction leads to a DBG_VALUE of virtual register
690  ``%0``.
691* The first GEP has its effect folded into the first load instruction
692  (as a 4-byte offset), but the variable location is salvaged by folding
693  the GEPs effect into the DIExpression.
694* The second GEP is also folded into the corresponding load. However, it is
695  insufficiently simple to be salvaged, and is emitted as a ``$noreg``
696  DBG_VALUE, indicating that the variable takes on an undefined location.
697* The final dbg.value has its Value placed in virtual register ``%1``.
698
699Instruction Scheduling
700----------------------
701
702A number of passes can reschedule instructions, notably instruction selection
703and the pre-and-post RA machine schedulers. Instruction scheduling can
704significantly change the nature of the program -- in the (very unlikely) worst
705case the instruction sequence could be completely reversed. In such
706circumstances LLVM follows the principle applied to optimizations, that it is
707better for the debugger not to display any state than a misleading state.
708Thus, whenever instructions are advanced in order of execution, any
709corresponding DBG_VALUE is kept in its original position, and if an instruction
710is delayed then the variable is given an undefined location for the duration
711of the delay. To illustrate, consider this pseudo-MIR:
712
713.. code-block:: text
714
715  %1:gr32 = MOV32rm %0, 1, $noreg, 4, $noreg, debug-location !5 :: (load 4 from %ir.addr1)
716  DBG_VALUE %1, $noreg, !1, !2
717  %4:gr32 = ADD32rr %3, %2, implicit-def dead $eflags
718  DBG_VALUE %4, $noreg, !3, !4
719  %7:gr32 = SUB32rr %6, %5, implicit-def dead $eflags
720  DBG_VALUE %7, $noreg, !5, !6
721
722Imagine that the SUB32rr were moved forward to give us the following MIR:
723
724.. code-block:: text
725
726  %7:gr32 = SUB32rr %6, %5, implicit-def dead $eflags
727  %1:gr32 = MOV32rm %0, 1, $noreg, 4, $noreg, debug-location !5 :: (load 4 from %ir.addr1)
728  DBG_VALUE %1, $noreg, !1, !2
729  %4:gr32 = ADD32rr %3, %2, implicit-def dead $eflags
730  DBG_VALUE %4, $noreg, !3, !4
731  DBG_VALUE %7, $noreg, !5, !6
732
733In this circumstance LLVM would leave the MIR as shown above. Were we to move
734the DBG_VALUE of virtual register %7 upwards with the SUB32rr, we would re-order
735assignments and introduce a new state of the program. Whereas with the solution
736above, the debugger will see one fewer combination of variable values, because
737``!3`` and ``!5`` will change value at the same time. This is preferred over
738misrepresenting the original program.
739
740In comparison, if one sunk the MOV32rm, LLVM would produce the following:
741
742.. code-block:: text
743
744  DBG_VALUE $noreg, $noreg, !1, !2
745  %4:gr32 = ADD32rr %3, %2, implicit-def dead $eflags
746  DBG_VALUE %4, $noreg, !3, !4
747  %7:gr32 = SUB32rr %6, %5, implicit-def dead $eflags
748  DBG_VALUE %7, $noreg, !5, !6
749  %1:gr32 = MOV32rm %0, 1, $noreg, 4, $noreg, debug-location !5 :: (load 4 from %ir.addr1)
750  DBG_VALUE %1, $noreg, !1, !2
751
752Here, to avoid presenting a state in which the first assignment to ``!1``
753disappears, the DBG_VALUE at the top of the block assigns the variable the
754undefined location, until its value is available at the end of the block where
755an additional DBG_VALUE is added. Were any other DBG_VALUE for ``!1`` to occur
756in the instructions that the MOV32rm was sunk past, the DBG_VALUE for ``%1``
757would be dropped and the debugger would never observe it in the variable. This
758accurately reflects that the value is not available during the corresponding
759portion of the original program.
760
761Variable locations during Register Allocation
762---------------------------------------------
763
764To avoid debug instructions interfering with the register allocator, the
765LiveDebugVariables pass extracts variable locations from a MIR function and
766deletes the corresponding DBG_VALUE instructions. Some localized copy
767propagation is performed within blocks. After register allocation, the
768VirtRegRewriter pass re-inserts DBG_VALUE instructions in their original
769positions, translating virtual register references into their physical
770machine locations. To avoid encoding incorrect variable locations, in this
771pass any DBG_VALUE of a virtual register that is not live, is replaced by
772the undefined location.
773
774LiveDebugValues expansion of variable locations
775-----------------------------------------------
776
777After all optimizations have run and shortly before emission, the
778LiveDebugValues pass runs to achieve two aims:
779
780* To propagate the location of variables through copies and register spills,
781* For every block, to record every valid variable location in that block.
782
783After this pass the DBG_VALUE instruction changes meaning: rather than
784corresponding to a source-level assignment where the variable may change value,
785it asserts the location of a variable in a block, and loses effect outside the
786block. Propagating variable locations through copies and spills is
787straightforwards: determining the variable location in every basic block
788requires the consideration of control flow. Consider the following IR, which
789presents several difficulties:
790
791.. code-block:: text
792
793  define dso_local i32 @foo(i1 %cond, i32 %input) !dbg !12 {
794  entry:
795    br i1 %cond, label %truebr, label %falsebr
796
797  bb1:
798    %value = phi i32 [ %value1, %truebr ], [ %value2, %falsebr ]
799    br label %exit, !dbg !26
800
801  truebr:
802    call void @llvm.dbg.value(metadata i32 %input, metadata !30, metadata !DIExpression()), !dbg !23
803    call void @llvm.dbg.value(metadata i32 1, metadata !22, metadata !DIExpression()), !dbg !23
804    %value1 = add i32 %input, 1
805    br label %bb1
806
807  falsebr:
808    call void @llvm.dbg.value(metadata i32 %input, metadata !30, metadata !DIExpression()), !dbg !23
809    call void @llvm.dbg.value(metadata i32 2, metadata !22, metadata !DIExpression()), !dbg !23
810    %value = add i32 %input, 2
811    br label %bb1
812
813  exit:
814    ret i32 %value, !dbg !30
815  }
816
817Here the difficulties are:
818
819* The control flow is roughly the opposite of basic block order
820* The value of the ``!22`` variable merges into ``%bb1``, but there is no PHI
821  node
822
823As mentioned above, the ``llvm.dbg.value`` intrinsics essentially form an
824imperative program embedded in the IR, with each intrinsic defining a variable
825location. This *could* be converted to an SSA form by mem2reg, in the same way
826that it uses use-def chains to identify control flow merges and insert phi
827nodes for IR Values. However, because debug variable locations are defined for
828every machine instruction, in effect every IR instruction uses every variable
829location, which would lead to a large number of debugging intrinsics being
830generated.
831
832Examining the example above, variable ``!30`` is assigned ``%input`` on both
833conditional paths through the function, while ``!22`` is assigned differing
834constant values on either path. Where control flow merges in ``%bb1`` we would
835want ``!30`` to keep its location (``%input``), but ``!22`` to become undefined
836as we cannot determine at runtime what value it should have in %bb1 without
837inserting a PHI node. mem2reg does not insert the PHI node to avoid changing
838codegen when debugging is enabled, and does not insert the other dbg.values
839to avoid adding very large numbers of intrinsics.
840
841Instead, LiveDebugValues determines variable locations when control
842flow merges. A dataflow analysis is used to propagate locations between blocks:
843when control flow merges, if a variable has the same location in all
844predecessors then that location is propagated into the successor. If the
845predecessor locations disagree, the location becomes undefined.
846
847Once LiveDebugValues has run, every block should have all valid variable
848locations described by DBG_VALUE instructions within the block. Very little
849effort is then required by supporting classes (such as
850DbgEntityHistoryCalculator) to build a map of each instruction to every
851valid variable location, without the need to consider control flow. From
852the example above, it is otherwise difficult to determine that the location
853of variable ``!30`` should flow "up" into block ``%bb1``, but that the location
854of variable ``!22`` should not flow "down" into the ``%exit`` block.
855
856.. _ccxx_frontend:
857
858C/C++ front-end specific debug information
859==========================================
860
861The C and C++ front-ends represent information about the program in a
862format that is effectively identical to `DWARF <http://www.dwarfstd.org/>`_
863in terms of information content.  This allows code generators to
864trivially support native debuggers by generating standard dwarf
865information, and contains enough information for non-dwarf targets to
866translate it as needed.
867
868This section describes the forms used to represent C and C++ programs.  Other
869languages could pattern themselves after this (which itself is tuned to
870representing programs in the same way that DWARF does), or they could choose
871to provide completely different forms if they don't fit into the DWARF model.
872As support for debugging information gets added to the various LLVM
873source-language front-ends, the information used should be documented here.
874
875The following sections provide examples of a few C/C++ constructs and
876the debug information that would best describe those constructs.  The
877canonical references are the ``DINode`` classes defined in
878``include/llvm/IR/DebugInfoMetadata.h`` and the implementations of the
879helper functions in ``lib/IR/DIBuilder.cpp``.
880
881C/C++ source file information
882-----------------------------
883
884``llvm::Instruction`` provides easy access to metadata attached with an
885instruction.  One can extract line number information encoded in LLVM IR using
886``Instruction::getDebugLoc()`` and ``DILocation::getLine()``.
887
888.. code-block:: c++
889
890  if (DILocation *Loc = I->getDebugLoc()) { // Here I is an LLVM instruction
891    unsigned Line = Loc->getLine();
892    StringRef File = Loc->getFilename();
893    StringRef Dir = Loc->getDirectory();
894    bool ImplicitCode = Loc->isImplicitCode();
895  }
896
897When the flag ImplicitCode is true then it means that the Instruction has been
898added by the front-end but doesn't correspond to source code written by the user. For example
899
900.. code-block:: c++
901
902  if (MyBoolean) {
903    MyObject MO;
904    ...
905  }
906
907At the end of the scope the MyObject's destructor is called but it isn't written
908explicitly. This information is useful to avoid to have counters on brackets when
909making code coverage.
910
911C/C++ global variable information
912---------------------------------
913
914Given an integer global variable declared as follows:
915
916.. code-block:: c
917
918  _Alignas(8) int MyGlobal = 100;
919
920a C/C++ front-end would generate the following descriptors:
921
922.. code-block:: text
923
924  ;;
925  ;; Define the global itself.
926  ;;
927  @MyGlobal = global i32 100, align 8, !dbg !0
928
929  ;;
930  ;; List of debug info of globals
931  ;;
932  !llvm.dbg.cu = !{!1}
933
934  ;; Some unrelated metadata.
935  !llvm.module.flags = !{!6, !7}
936  !llvm.ident = !{!8}
937
938  ;; Define the global variable itself
939  !0 = distinct !DIGlobalVariable(name: "MyGlobal", scope: !1, file: !2, line: 1, type: !5, isLocal: false, isDefinition: true, align: 64)
940
941  ;; Define the compile unit.
942  !1 = distinct !DICompileUnit(language: DW_LANG_C99, file: !2,
943                               producer: "clang version 4.0.0",
944                               isOptimized: false, runtimeVersion: 0, emissionKind: FullDebug,
945                               enums: !3, globals: !4)
946
947  ;;
948  ;; Define the file
949  ;;
950  !2 = !DIFile(filename: "/dev/stdin",
951               directory: "/Users/dexonsmith/data/llvm/debug-info")
952
953  ;; An empty array.
954  !3 = !{}
955
956  ;; The Array of Global Variables
957  !4 = !{!0}
958
959  ;;
960  ;; Define the type
961  ;;
962  !5 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed)
963
964  ;; Dwarf version to output.
965  !6 = !{i32 2, !"Dwarf Version", i32 4}
966
967  ;; Debug info schema version.
968  !7 = !{i32 2, !"Debug Info Version", i32 3}
969
970  ;; Compiler identification
971  !8 = !{!"clang version 4.0.0"}
972
973
974The align value in DIGlobalVariable description specifies variable alignment in
975case it was forced by C11 _Alignas(), C++11 alignas() keywords or compiler
976attribute __attribute__((aligned ())). In other case (when this field is missing)
977alignment is considered default. This is used when producing DWARF output
978for DW_AT_alignment value.
979
980C/C++ function information
981--------------------------
982
983Given a function declared as follows:
984
985.. code-block:: c
986
987  int main(int argc, char *argv[]) {
988    return 0;
989  }
990
991a C/C++ front-end would generate the following descriptors:
992
993.. code-block:: text
994
995  ;;
996  ;; Define the anchor for subprograms.
997  ;;
998  !4 = !DISubprogram(name: "main", scope: !1, file: !1, line: 1, type: !5,
999                     isLocal: false, isDefinition: true, scopeLine: 1,
1000                     flags: DIFlagPrototyped, isOptimized: false,
1001                     variables: !2)
1002
1003  ;;
1004  ;; Define the subprogram itself.
1005  ;;
1006  define i32 @main(i32 %argc, i8** %argv) !dbg !4 {
1007  ...
1008  }
1009
1010C++ specific debug information
1011==============================
1012
1013C++ special member functions information
1014----------------------------------------
1015
1016DWARF v5 introduces attributes defined to enhance debugging information of C++ programs. LLVM can generate (or omit) these appropriate DWARF attributes. In C++ a special member function Ctors, Dtors, Copy/Move Ctors, assignment operators can be declared with C++11 keyword deleted. This is represented in LLVM using spFlags value DISPFlagDeleted.
1017
1018Given a class declaration with copy constructor declared as deleted:
1019
1020.. code-block:: c
1021
1022  class foo {
1023   public:
1024     foo(const foo&) = deleted;
1025  };
1026
1027A C++ frontend would generate following:
1028
1029.. code-block:: text
1030
1031  !17 = !DISubprogram(name: "foo", scope: !11, file: !1, line: 5, type: !18, scopeLine: 5, flags: DIFlagPublic | DIFlagPrototyped, spFlags: DISPFlagDeleted)
1032
1033and this will produce an additional DWARF attribute as:
1034
1035.. code-block:: text
1036
1037  DW_TAG_subprogram [7] *
1038    DW_AT_name [DW_FORM_strx1]    (indexed (00000006) string = "foo")
1039    DW_AT_decl_line [DW_FORM_data1]       (5)
1040    ...
1041    DW_AT_deleted [DW_FORM_flag_present]  (true)
1042
1043Fortran specific debug information
1044==================================
1045
1046Fortran function information
1047----------------------------
1048
1049There are a few DWARF attributes defined to support client debugging of Fortran programs.  LLVM can generate (or omit) the appropriate DWARF attributes for the prefix-specs of ELEMENTAL, PURE, IMPURE, RECURSIVE, and NON_RECURSIVE.  This is done by using the spFlags values: DISPFlagElemental, DISPFlagPure, and DISPFlagRecursive.
1050
1051.. code-block:: fortran
1052
1053  elemental function elem_func(a)
1054
1055a Fortran front-end would generate the following descriptors:
1056
1057.. code-block:: text
1058
1059  !11 = distinct !DISubprogram(name: "subroutine2", scope: !1, file: !1,
1060          line: 5, type: !8, scopeLine: 6,
1061          spFlags: DISPFlagDefinition | DISPFlagElemental, unit: !0,
1062          retainedNodes: !2)
1063
1064and this will materialize an additional DWARF attribute as:
1065
1066.. code-block:: text
1067
1068  DW_TAG_subprogram [3]
1069     DW_AT_low_pc [DW_FORM_addr]     (0x0000000000000010 ".text")
1070     DW_AT_high_pc [DW_FORM_data4]   (0x00000001)
1071     ...
1072     DW_AT_elemental [DW_FORM_flag_present]  (true)
1073
1074There are a few DWARF tags defined to represent Fortran specific constructs i.e DW_TAG_string_type for representing Fortran character(n). In LLVM this is represented as DIStringType.
1075
1076.. code-block:: fortran
1077
1078  character(len=*), intent(in) :: string
1079
1080a Fortran front-end would generate the following descriptors:
1081
1082.. code-block:: text
1083
1084  !DILocalVariable(name: "string", arg: 1, scope: !10, file: !3, line: 4, type: !15)
1085  !DIStringType(name: "character(*)!2", stringLength: !16, stringLengthExpression: !DIExpression(), size: 32)
1086
1087and this will materialize in DWARF tags as:
1088
1089.. code-block:: text
1090
1091   DW_TAG_string_type
1092                DW_AT_name      ("character(*)!2")
1093                DW_AT_string_length     (0x00000064)
1094   0x00000064:    DW_TAG_variable
1095                  DW_AT_location      (DW_OP_fbreg +16)
1096                  DW_AT_type  (0x00000083 "integer*8")
1097                  ...
1098                  DW_AT_artificial    (true)
1099
1100Debugging information format
1101============================
1102
1103Debugging Information Extension for Objective C Properties
1104----------------------------------------------------------
1105
1106Introduction
1107^^^^^^^^^^^^
1108
1109Objective C provides a simpler way to declare and define accessor methods using
1110declared properties.  The language provides features to declare a property and
1111to let compiler synthesize accessor methods.
1112
1113The debugger lets developer inspect Objective C interfaces and their instance
1114variables and class variables.  However, the debugger does not know anything
1115about the properties defined in Objective C interfaces.  The debugger consumes
1116information generated by compiler in DWARF format.  The format does not support
1117encoding of Objective C properties.  This proposal describes DWARF extensions to
1118encode Objective C properties, which the debugger can use to let developers
1119inspect Objective C properties.
1120
1121Proposal
1122^^^^^^^^
1123
1124Objective C properties exist separately from class members.  A property can be
1125defined only by "setter" and "getter" selectors, and be calculated anew on each
1126access.  Or a property can just be a direct access to some declared ivar.
1127Finally it can have an ivar "automatically synthesized" for it by the compiler,
1128in which case the property can be referred to in user code directly using the
1129standard C dereference syntax as well as through the property "dot" syntax, but
1130there is no entry in the ``@interface`` declaration corresponding to this ivar.
1131
1132To facilitate debugging, these properties we will add a new DWARF TAG into the
1133``DW_TAG_structure_type`` definition for the class to hold the description of a
1134given property, and a set of DWARF attributes that provide said description.
1135The property tag will also contain the name and declared type of the property.
1136
1137If there is a related ivar, there will also be a DWARF property attribute placed
1138in the ``DW_TAG_member`` DIE for that ivar referring back to the property TAG
1139for that property.  And in the case where the compiler synthesizes the ivar
1140directly, the compiler is expected to generate a ``DW_TAG_member`` for that
1141ivar (with the ``DW_AT_artificial`` set to 1), whose name will be the name used
1142to access this ivar directly in code, and with the property attribute pointing
1143back to the property it is backing.
1144
1145The following examples will serve as illustration for our discussion:
1146
1147.. code-block:: objc
1148
1149  @interface I1 {
1150    int n2;
1151  }
1152
1153  @property int p1;
1154  @property int p2;
1155  @end
1156
1157  @implementation I1
1158  @synthesize p1;
1159  @synthesize p2 = n2;
1160  @end
1161
1162This produces the following DWARF (this is a "pseudo dwarfdump" output):
1163
1164.. code-block:: none
1165
1166  0x00000100:  TAG_structure_type [7] *
1167                 AT_APPLE_runtime_class( 0x10 )
1168                 AT_name( "I1" )
1169                 AT_decl_file( "Objc_Property.m" )
1170                 AT_decl_line( 3 )
1171
1172  0x00000110    TAG_APPLE_property
1173                  AT_name ( "p1" )
1174                  AT_type ( {0x00000150} ( int ) )
1175
1176  0x00000120:   TAG_APPLE_property
1177                  AT_name ( "p2" )
1178                  AT_type ( {0x00000150} ( int ) )
1179
1180  0x00000130:   TAG_member [8]
1181                  AT_name( "_p1" )
1182                  AT_APPLE_property ( {0x00000110} "p1" )
1183                  AT_type( {0x00000150} ( int ) )
1184                  AT_artificial ( 0x1 )
1185
1186  0x00000140:    TAG_member [8]
1187                   AT_name( "n2" )
1188                   AT_APPLE_property ( {0x00000120} "p2" )
1189                   AT_type( {0x00000150} ( int ) )
1190
1191  0x00000150:  AT_type( ( int ) )
1192
1193Note, the current convention is that the name of the ivar for an
1194auto-synthesized property is the name of the property from which it derives
1195with an underscore prepended, as is shown in the example.  But we actually
1196don't need to know this convention, since we are given the name of the ivar
1197directly.
1198
1199Also, it is common practice in ObjC to have different property declarations in
1200the @interface and @implementation - e.g. to provide a read-only property in
1201the interface, and a read-write interface in the implementation.  In that case,
1202the compiler should emit whichever property declaration will be in force in the
1203current translation unit.
1204
1205Developers can decorate a property with attributes which are encoded using
1206``DW_AT_APPLE_property_attribute``.
1207
1208.. code-block:: objc
1209
1210  @property (readonly, nonatomic) int pr;
1211
1212.. code-block:: none
1213
1214  TAG_APPLE_property [8]
1215    AT_name( "pr" )
1216    AT_type ( {0x00000147} (int) )
1217    AT_APPLE_property_attribute (DW_APPLE_PROPERTY_readonly, DW_APPLE_PROPERTY_nonatomic)
1218
1219The setter and getter method names are attached to the property using
1220``DW_AT_APPLE_property_setter`` and ``DW_AT_APPLE_property_getter`` attributes.
1221
1222.. code-block:: objc
1223
1224  @interface I1
1225  @property (setter=myOwnP3Setter:) int p3;
1226  -(void)myOwnP3Setter:(int)a;
1227  @end
1228
1229  @implementation I1
1230  @synthesize p3;
1231  -(void)myOwnP3Setter:(int)a{ }
1232  @end
1233
1234The DWARF for this would be:
1235
1236.. code-block:: none
1237
1238  0x000003bd: TAG_structure_type [7] *
1239                AT_APPLE_runtime_class( 0x10 )
1240                AT_name( "I1" )
1241                AT_decl_file( "Objc_Property.m" )
1242                AT_decl_line( 3 )
1243
1244  0x000003cd      TAG_APPLE_property
1245                    AT_name ( "p3" )
1246                    AT_APPLE_property_setter ( "myOwnP3Setter:" )
1247                    AT_type( {0x00000147} ( int ) )
1248
1249  0x000003f3:     TAG_member [8]
1250                    AT_name( "_p3" )
1251                    AT_type ( {0x00000147} ( int ) )
1252                    AT_APPLE_property ( {0x000003cd} )
1253                    AT_artificial ( 0x1 )
1254
1255New DWARF Tags
1256^^^^^^^^^^^^^^
1257
1258+-----------------------+--------+
1259| TAG                   | Value  |
1260+=======================+========+
1261| DW_TAG_APPLE_property | 0x4200 |
1262+-----------------------+--------+
1263
1264New DWARF Attributes
1265^^^^^^^^^^^^^^^^^^^^
1266
1267+--------------------------------+--------+-----------+
1268| Attribute                      | Value  | Classes   |
1269+================================+========+===========+
1270| DW_AT_APPLE_property           | 0x3fed | Reference |
1271+--------------------------------+--------+-----------+
1272| DW_AT_APPLE_property_getter    | 0x3fe9 | String    |
1273+--------------------------------+--------+-----------+
1274| DW_AT_APPLE_property_setter    | 0x3fea | String    |
1275+--------------------------------+--------+-----------+
1276| DW_AT_APPLE_property_attribute | 0x3feb | Constant  |
1277+--------------------------------+--------+-----------+
1278
1279New DWARF Constants
1280^^^^^^^^^^^^^^^^^^^
1281
1282+--------------------------------------+-------+
1283| Name                                 | Value |
1284+======================================+=======+
1285| DW_APPLE_PROPERTY_readonly           | 0x01  |
1286+--------------------------------------+-------+
1287| DW_APPLE_PROPERTY_getter             | 0x02  |
1288+--------------------------------------+-------+
1289| DW_APPLE_PROPERTY_assign             | 0x04  |
1290+--------------------------------------+-------+
1291| DW_APPLE_PROPERTY_readwrite          | 0x08  |
1292+--------------------------------------+-------+
1293| DW_APPLE_PROPERTY_retain             | 0x10  |
1294+--------------------------------------+-------+
1295| DW_APPLE_PROPERTY_copy               | 0x20  |
1296+--------------------------------------+-------+
1297| DW_APPLE_PROPERTY_nonatomic          | 0x40  |
1298+--------------------------------------+-------+
1299| DW_APPLE_PROPERTY_setter             | 0x80  |
1300+--------------------------------------+-------+
1301| DW_APPLE_PROPERTY_atomic             | 0x100 |
1302+--------------------------------------+-------+
1303| DW_APPLE_PROPERTY_weak               | 0x200 |
1304+--------------------------------------+-------+
1305| DW_APPLE_PROPERTY_strong             | 0x400 |
1306+--------------------------------------+-------+
1307| DW_APPLE_PROPERTY_unsafe_unretained  | 0x800 |
1308+--------------------------------------+-------+
1309| DW_APPLE_PROPERTY_nullability        | 0x1000|
1310+--------------------------------------+-------+
1311| DW_APPLE_PROPERTY_null_resettable    | 0x2000|
1312+--------------------------------------+-------+
1313| DW_APPLE_PROPERTY_class              | 0x4000|
1314+--------------------------------------+-------+
1315
1316Name Accelerator Tables
1317-----------------------
1318
1319Introduction
1320^^^^^^^^^^^^
1321
1322The "``.debug_pubnames``" and "``.debug_pubtypes``" formats are not what a
1323debugger needs.  The "``pub``" in the section name indicates that the entries
1324in the table are publicly visible names only.  This means no static or hidden
1325functions show up in the "``.debug_pubnames``".  No static variables or private
1326class variables are in the "``.debug_pubtypes``".  Many compilers add different
1327things to these tables, so we can't rely upon the contents between gcc, icc, or
1328clang.
1329
1330The typical query given by users tends not to match up with the contents of
1331these tables.  For example, the DWARF spec states that "In the case of the name
1332of a function member or static data member of a C++ structure, class or union,
1333the name presented in the "``.debug_pubnames``" section is not the simple name
1334given by the ``DW_AT_name attribute`` of the referenced debugging information
1335entry, but rather the fully qualified name of the data or function member."
1336So the only names in these tables for complex C++ entries is a fully
1337qualified name.  Debugger users tend not to enter their search strings as
1338"``a::b::c(int,const Foo&) const``", but rather as "``c``", "``b::c``" , or
1339"``a::b::c``".  So the name entered in the name table must be demangled in
1340order to chop it up appropriately and additional names must be manually entered
1341into the table to make it effective as a name lookup table for debuggers to
1342use.
1343
1344All debuggers currently ignore the "``.debug_pubnames``" table as a result of
1345its inconsistent and useless public-only name content making it a waste of
1346space in the object file.  These tables, when they are written to disk, are not
1347sorted in any way, leaving every debugger to do its own parsing and sorting.
1348These tables also include an inlined copy of the string values in the table
1349itself making the tables much larger than they need to be on disk, especially
1350for large C++ programs.
1351
1352Can't we just fix the sections by adding all of the names we need to this
1353table? No, because that is not what the tables are defined to contain and we
1354won't know the difference between the old bad tables and the new good tables.
1355At best we could make our own renamed sections that contain all of the data we
1356need.
1357
1358These tables are also insufficient for what a debugger like LLDB needs.  LLDB
1359uses clang for its expression parsing where LLDB acts as a PCH.  LLDB is then
1360often asked to look for type "``foo``" or namespace "``bar``", or list items in
1361namespace "``baz``".  Namespaces are not included in the pubnames or pubtypes
1362tables.  Since clang asks a lot of questions when it is parsing an expression,
1363we need to be very fast when looking up names, as it happens a lot.  Having new
1364accelerator tables that are optimized for very quick lookups will benefit this
1365type of debugging experience greatly.
1366
1367We would like to generate name lookup tables that can be mapped into memory
1368from disk, and used as is, with little or no up-front parsing.  We would also
1369be able to control the exact content of these different tables so they contain
1370exactly what we need.  The Name Accelerator Tables were designed to fix these
1371issues.  In order to solve these issues we need to:
1372
1373* Have a format that can be mapped into memory from disk and used as is
1374* Lookups should be very fast
1375* Extensible table format so these tables can be made by many producers
1376* Contain all of the names needed for typical lookups out of the box
1377* Strict rules for the contents of tables
1378
1379Table size is important and the accelerator table format should allow the reuse
1380of strings from common string tables so the strings for the names are not
1381duplicated.  We also want to make sure the table is ready to be used as-is by
1382simply mapping the table into memory with minimal header parsing.
1383
1384The name lookups need to be fast and optimized for the kinds of lookups that
1385debuggers tend to do.  Optimally we would like to touch as few parts of the
1386mapped table as possible when doing a name lookup and be able to quickly find
1387the name entry we are looking for, or discover there are no matches.  In the
1388case of debuggers we optimized for lookups that fail most of the time.
1389
1390Each table that is defined should have strict rules on exactly what is in the
1391accelerator tables and documented so clients can rely on the content.
1392
1393Hash Tables
1394^^^^^^^^^^^
1395
1396Standard Hash Tables
1397""""""""""""""""""""
1398
1399Typical hash tables have a header, buckets, and each bucket points to the
1400bucket contents:
1401
1402.. code-block:: none
1403
1404  .------------.
1405  |  HEADER    |
1406  |------------|
1407  |  BUCKETS   |
1408  |------------|
1409  |  DATA      |
1410  `------------'
1411
1412The BUCKETS are an array of offsets to DATA for each hash:
1413
1414.. code-block:: none
1415
1416  .------------.
1417  | 0x00001000 | BUCKETS[0]
1418  | 0x00002000 | BUCKETS[1]
1419  | 0x00002200 | BUCKETS[2]
1420  | 0x000034f0 | BUCKETS[3]
1421  |            | ...
1422  | 0xXXXXXXXX | BUCKETS[n_buckets]
1423  '------------'
1424
1425So for ``bucket[3]`` in the example above, we have an offset into the table
14260x000034f0 which points to a chain of entries for the bucket.  Each bucket must
1427contain a next pointer, full 32 bit hash value, the string itself, and the data
1428for the current string value.
1429
1430.. code-block:: none
1431
1432              .------------.
1433  0x000034f0: | 0x00003500 | next pointer
1434              | 0x12345678 | 32 bit hash
1435              | "erase"    | string value
1436              | data[n]    | HashData for this bucket
1437              |------------|
1438  0x00003500: | 0x00003550 | next pointer
1439              | 0x29273623 | 32 bit hash
1440              | "dump"     | string value
1441              | data[n]    | HashData for this bucket
1442              |------------|
1443  0x00003550: | 0x00000000 | next pointer
1444              | 0x82638293 | 32 bit hash
1445              | "main"     | string value
1446              | data[n]    | HashData for this bucket
1447              `------------'
1448
1449The problem with this layout for debuggers is that we need to optimize for the
1450negative lookup case where the symbol we're searching for is not present.  So
1451if we were to lookup "``printf``" in the table above, we would make a 32-bit
1452hash for "``printf``", it might match ``bucket[3]``.  We would need to go to
1453the offset 0x000034f0 and start looking to see if our 32 bit hash matches.  To
1454do so, we need to read the next pointer, then read the hash, compare it, and
1455skip to the next bucket.  Each time we are skipping many bytes in memory and
1456touching new pages just to do the compare on the full 32 bit hash.  All of
1457these accesses then tell us that we didn't have a match.
1458
1459Name Hash Tables
1460""""""""""""""""
1461
1462To solve the issues mentioned above we have structured the hash tables a bit
1463differently: a header, buckets, an array of all unique 32 bit hash values,
1464followed by an array of hash value data offsets, one for each hash value, then
1465the data for all hash values:
1466
1467.. code-block:: none
1468
1469  .-------------.
1470  |  HEADER     |
1471  |-------------|
1472  |  BUCKETS    |
1473  |-------------|
1474  |  HASHES     |
1475  |-------------|
1476  |  OFFSETS    |
1477  |-------------|
1478  |  DATA       |
1479  `-------------'
1480
1481The ``BUCKETS`` in the name tables are an index into the ``HASHES`` array.  By
1482making all of the full 32 bit hash values contiguous in memory, we allow
1483ourselves to efficiently check for a match while touching as little memory as
1484possible.  Most often checking the 32 bit hash values is as far as the lookup
1485goes.  If it does match, it usually is a match with no collisions.  So for a
1486table with "``n_buckets``" buckets, and "``n_hashes``" unique 32 bit hash
1487values, we can clarify the contents of the ``BUCKETS``, ``HASHES`` and
1488``OFFSETS`` as:
1489
1490.. code-block:: none
1491
1492  .-------------------------.
1493  |  HEADER.magic           | uint32_t
1494  |  HEADER.version         | uint16_t
1495  |  HEADER.hash_function   | uint16_t
1496  |  HEADER.bucket_count    | uint32_t
1497  |  HEADER.hashes_count    | uint32_t
1498  |  HEADER.header_data_len | uint32_t
1499  |  HEADER_DATA            | HeaderData
1500  |-------------------------|
1501  |  BUCKETS                | uint32_t[n_buckets] // 32 bit hash indexes
1502  |-------------------------|
1503  |  HASHES                 | uint32_t[n_hashes] // 32 bit hash values
1504  |-------------------------|
1505  |  OFFSETS                | uint32_t[n_hashes] // 32 bit offsets to hash value data
1506  |-------------------------|
1507  |  ALL HASH DATA          |
1508  `-------------------------'
1509
1510So taking the exact same data from the standard hash example above we end up
1511with:
1512
1513.. code-block:: none
1514
1515              .------------.
1516              | HEADER     |
1517              |------------|
1518              |          0 | BUCKETS[0]
1519              |          2 | BUCKETS[1]
1520              |          5 | BUCKETS[2]
1521              |          6 | BUCKETS[3]
1522              |            | ...
1523              |        ... | BUCKETS[n_buckets]
1524              |------------|
1525              | 0x........ | HASHES[0]
1526              | 0x........ | HASHES[1]
1527              | 0x........ | HASHES[2]
1528              | 0x........ | HASHES[3]
1529              | 0x........ | HASHES[4]
1530              | 0x........ | HASHES[5]
1531              | 0x12345678 | HASHES[6]    hash for BUCKETS[3]
1532              | 0x29273623 | HASHES[7]    hash for BUCKETS[3]
1533              | 0x82638293 | HASHES[8]    hash for BUCKETS[3]
1534              | 0x........ | HASHES[9]
1535              | 0x........ | HASHES[10]
1536              | 0x........ | HASHES[11]
1537              | 0x........ | HASHES[12]
1538              | 0x........ | HASHES[13]
1539              | 0x........ | HASHES[n_hashes]
1540              |------------|
1541              | 0x........ | OFFSETS[0]
1542              | 0x........ | OFFSETS[1]
1543              | 0x........ | OFFSETS[2]
1544              | 0x........ | OFFSETS[3]
1545              | 0x........ | OFFSETS[4]
1546              | 0x........ | OFFSETS[5]
1547              | 0x000034f0 | OFFSETS[6]   offset for BUCKETS[3]
1548              | 0x00003500 | OFFSETS[7]   offset for BUCKETS[3]
1549              | 0x00003550 | OFFSETS[8]   offset for BUCKETS[3]
1550              | 0x........ | OFFSETS[9]
1551              | 0x........ | OFFSETS[10]
1552              | 0x........ | OFFSETS[11]
1553              | 0x........ | OFFSETS[12]
1554              | 0x........ | OFFSETS[13]
1555              | 0x........ | OFFSETS[n_hashes]
1556              |------------|
1557              |            |
1558              |            |
1559              |            |
1560              |            |
1561              |            |
1562              |------------|
1563  0x000034f0: | 0x00001203 | .debug_str ("erase")
1564              | 0x00000004 | A 32 bit array count - number of HashData with name "erase"
1565              | 0x........ | HashData[0]
1566              | 0x........ | HashData[1]
1567              | 0x........ | HashData[2]
1568              | 0x........ | HashData[3]
1569              | 0x00000000 | String offset into .debug_str (terminate data for hash)
1570              |------------|
1571  0x00003500: | 0x00001203 | String offset into .debug_str ("collision")
1572              | 0x00000002 | A 32 bit array count - number of HashData with name "collision"
1573              | 0x........ | HashData[0]
1574              | 0x........ | HashData[1]
1575              | 0x00001203 | String offset into .debug_str ("dump")
1576              | 0x00000003 | A 32 bit array count - number of HashData with name "dump"
1577              | 0x........ | HashData[0]
1578              | 0x........ | HashData[1]
1579              | 0x........ | HashData[2]
1580              | 0x00000000 | String offset into .debug_str (terminate data for hash)
1581              |------------|
1582  0x00003550: | 0x00001203 | String offset into .debug_str ("main")
1583              | 0x00000009 | A 32 bit array count - number of HashData with name "main"
1584              | 0x........ | HashData[0]
1585              | 0x........ | HashData[1]
1586              | 0x........ | HashData[2]
1587              | 0x........ | HashData[3]
1588              | 0x........ | HashData[4]
1589              | 0x........ | HashData[5]
1590              | 0x........ | HashData[6]
1591              | 0x........ | HashData[7]
1592              | 0x........ | HashData[8]
1593              | 0x00000000 | String offset into .debug_str (terminate data for hash)
1594              `------------'
1595
1596So we still have all of the same data, we just organize it more efficiently for
1597debugger lookup.  If we repeat the same "``printf``" lookup from above, we
1598would hash "``printf``" and find it matches ``BUCKETS[3]`` by taking the 32 bit
1599hash value and modulo it by ``n_buckets``.  ``BUCKETS[3]`` contains "6" which
1600is the index into the ``HASHES`` table.  We would then compare any consecutive
160132 bit hashes values in the ``HASHES`` array as long as the hashes would be in
1602``BUCKETS[3]``.  We do this by verifying that each subsequent hash value modulo
1603``n_buckets`` is still 3.  In the case of a failed lookup we would access the
1604memory for ``BUCKETS[3]``, and then compare a few consecutive 32 bit hashes
1605before we know that we have no match.  We don't end up marching through
1606multiple words of memory and we really keep the number of processor data cache
1607lines being accessed as small as possible.
1608
1609The string hash that is used for these lookup tables is the Daniel J.
1610Bernstein hash which is also used in the ELF ``GNU_HASH`` sections.  It is a
1611very good hash for all kinds of names in programs with very few hash
1612collisions.
1613
1614Empty buckets are designated by using an invalid hash index of ``UINT32_MAX``.
1615
1616Details
1617^^^^^^^
1618
1619These name hash tables are designed to be generic where specializations of the
1620table get to define additional data that goes into the header ("``HeaderData``"),
1621how the string value is stored ("``KeyType``") and the content of the data for each
1622hash value.
1623
1624Header Layout
1625"""""""""""""
1626
1627The header has a fixed part, and the specialized part.  The exact format of the
1628header is:
1629
1630.. code-block:: c
1631
1632  struct Header
1633  {
1634    uint32_t   magic;           // 'HASH' magic value to allow endian detection
1635    uint16_t   version;         // Version number
1636    uint16_t   hash_function;   // The hash function enumeration that was used
1637    uint32_t   bucket_count;    // The number of buckets in this hash table
1638    uint32_t   hashes_count;    // The total number of unique hash values and hash data offsets in this table
1639    uint32_t   header_data_len; // The bytes to skip to get to the hash indexes (buckets) for correct alignment
1640                                // Specifically the length of the following HeaderData field - this does not
1641                                // include the size of the preceding fields
1642    HeaderData header_data;     // Implementation specific header data
1643  };
1644
1645The header starts with a 32 bit "``magic``" value which must be ``'HASH'``
1646encoded as an ASCII integer.  This allows the detection of the start of the
1647hash table and also allows the table's byte order to be determined so the table
1648can be correctly extracted.  The "``magic``" value is followed by a 16 bit
1649``version`` number which allows the table to be revised and modified in the
1650future.  The current version number is 1. ``hash_function`` is a ``uint16_t``
1651enumeration that specifies which hash function was used to produce this table.
1652The current values for the hash function enumerations include:
1653
1654.. code-block:: c
1655
1656  enum HashFunctionType
1657  {
1658    eHashFunctionDJB = 0u, // Daniel J Bernstein hash function
1659  };
1660
1661``bucket_count`` is a 32 bit unsigned integer that represents how many buckets
1662are in the ``BUCKETS`` array.  ``hashes_count`` is the number of unique 32 bit
1663hash values that are in the ``HASHES`` array, and is the same number of offsets
1664are contained in the ``OFFSETS`` array.  ``header_data_len`` specifies the size
1665in bytes of the ``HeaderData`` that is filled in by specialized versions of
1666this table.
1667
1668Fixed Lookup
1669""""""""""""
1670
1671The header is followed by the buckets, hashes, offsets, and hash value data.
1672
1673.. code-block:: c
1674
1675  struct FixedTable
1676  {
1677    uint32_t buckets[Header.bucket_count];  // An array of hash indexes into the "hashes[]" array below
1678    uint32_t hashes [Header.hashes_count];  // Every unique 32 bit hash for the entire table is in this table
1679    uint32_t offsets[Header.hashes_count];  // An offset that corresponds to each item in the "hashes[]" array above
1680  };
1681
1682``buckets`` is an array of 32 bit indexes into the ``hashes`` array.  The
1683``hashes`` array contains all of the 32 bit hash values for all names in the
1684hash table.  Each hash in the ``hashes`` table has an offset in the ``offsets``
1685array that points to the data for the hash value.
1686
1687This table setup makes it very easy to repurpose these tables to contain
1688different data, while keeping the lookup mechanism the same for all tables.
1689This layout also makes it possible to save the table to disk and map it in
1690later and do very efficient name lookups with little or no parsing.
1691
1692DWARF lookup tables can be implemented in a variety of ways and can store a lot
1693of information for each name.  We want to make the DWARF tables extensible and
1694able to store the data efficiently so we have used some of the DWARF features
1695that enable efficient data storage to define exactly what kind of data we store
1696for each name.
1697
1698The ``HeaderData`` contains a definition of the contents of each HashData chunk.
1699We might want to store an offset to all of the debug information entries (DIEs)
1700for each name.  To keep things extensible, we create a list of items, or
1701Atoms, that are contained in the data for each name.  First comes the type of
1702the data in each atom:
1703
1704.. code-block:: c
1705
1706  enum AtomType
1707  {
1708    eAtomTypeNULL       = 0u,
1709    eAtomTypeDIEOffset  = 1u,   // DIE offset, check form for encoding
1710    eAtomTypeCUOffset   = 2u,   // DIE offset of the compiler unit header that contains the item in question
1711    eAtomTypeTag        = 3u,   // DW_TAG_xxx value, should be encoded as DW_FORM_data1 (if no tags exceed 255) or DW_FORM_data2
1712    eAtomTypeNameFlags  = 4u,   // Flags from enum NameFlags
1713    eAtomTypeTypeFlags  = 5u,   // Flags from enum TypeFlags
1714  };
1715
1716The enumeration values and their meanings are:
1717
1718.. code-block:: none
1719
1720  eAtomTypeNULL       - a termination atom that specifies the end of the atom list
1721  eAtomTypeDIEOffset  - an offset into the .debug_info section for the DWARF DIE for this name
1722  eAtomTypeCUOffset   - an offset into the .debug_info section for the CU that contains the DIE
1723  eAtomTypeDIETag     - The DW_TAG_XXX enumeration value so you don't have to parse the DWARF to see what it is
1724  eAtomTypeNameFlags  - Flags for functions and global variables (isFunction, isInlined, isExternal...)
1725  eAtomTypeTypeFlags  - Flags for types (isCXXClass, isObjCClass, ...)
1726
1727Then we allow each atom type to define the atom type and how the data for each
1728atom type data is encoded:
1729
1730.. code-block:: c
1731
1732  struct Atom
1733  {
1734    uint16_t type;  // AtomType enum value
1735    uint16_t form;  // DWARF DW_FORM_XXX defines
1736  };
1737
1738The ``form`` type above is from the DWARF specification and defines the exact
1739encoding of the data for the Atom type.  See the DWARF specification for the
1740``DW_FORM_`` definitions.
1741
1742.. code-block:: c
1743
1744  struct HeaderData
1745  {
1746    uint32_t die_offset_base;
1747    uint32_t atom_count;
1748    Atoms    atoms[atom_count0];
1749  };
1750
1751``HeaderData`` defines the base DIE offset that should be added to any atoms
1752that are encoded using the ``DW_FORM_ref1``, ``DW_FORM_ref2``,
1753``DW_FORM_ref4``, ``DW_FORM_ref8`` or ``DW_FORM_ref_udata``.  It also defines
1754what is contained in each ``HashData`` object -- ``Atom.form`` tells us how large
1755each field will be in the ``HashData`` and the ``Atom.type`` tells us how this data
1756should be interpreted.
1757
1758For the current implementations of the "``.apple_names``" (all functions +
1759globals), the "``.apple_types``" (names of all types that are defined), and
1760the "``.apple_namespaces``" (all namespaces), we currently set the ``Atom``
1761array to be:
1762
1763.. code-block:: c
1764
1765  HeaderData.atom_count = 1;
1766  HeaderData.atoms[0].type = eAtomTypeDIEOffset;
1767  HeaderData.atoms[0].form = DW_FORM_data4;
1768
1769This defines the contents to be the DIE offset (eAtomTypeDIEOffset) that is
1770encoded as a 32 bit value (DW_FORM_data4).  This allows a single name to have
1771multiple matching DIEs in a single file, which could come up with an inlined
1772function for instance.  Future tables could include more information about the
1773DIE such as flags indicating if the DIE is a function, method, block,
1774or inlined.
1775
1776The KeyType for the DWARF table is a 32 bit string table offset into the
1777".debug_str" table.  The ".debug_str" is the string table for the DWARF which
1778may already contain copies of all of the strings.  This helps make sure, with
1779help from the compiler, that we reuse the strings between all of the DWARF
1780sections and keeps the hash table size down.  Another benefit to having the
1781compiler generate all strings as DW_FORM_strp in the debug info, is that
1782DWARF parsing can be made much faster.
1783
1784After a lookup is made, we get an offset into the hash data.  The hash data
1785needs to be able to deal with 32 bit hash collisions, so the chunk of data
1786at the offset in the hash data consists of a triple:
1787
1788.. code-block:: c
1789
1790  uint32_t str_offset
1791  uint32_t hash_data_count
1792  HashData[hash_data_count]
1793
1794If "str_offset" is zero, then the bucket contents are done. 99.9% of the
1795hash data chunks contain a single item (no 32 bit hash collision):
1796
1797.. code-block:: none
1798
1799  .------------.
1800  | 0x00001023 | uint32_t KeyType (.debug_str[0x0001023] => "main")
1801  | 0x00000004 | uint32_t HashData count
1802  | 0x........ | uint32_t HashData[0] DIE offset
1803  | 0x........ | uint32_t HashData[1] DIE offset
1804  | 0x........ | uint32_t HashData[2] DIE offset
1805  | 0x........ | uint32_t HashData[3] DIE offset
1806  | 0x00000000 | uint32_t KeyType (end of hash chain)
1807  `------------'
1808
1809If there are collisions, you will have multiple valid string offsets:
1810
1811.. code-block:: none
1812
1813  .------------.
1814  | 0x00001023 | uint32_t KeyType (.debug_str[0x0001023] => "main")
1815  | 0x00000004 | uint32_t HashData count
1816  | 0x........ | uint32_t HashData[0] DIE offset
1817  | 0x........ | uint32_t HashData[1] DIE offset
1818  | 0x........ | uint32_t HashData[2] DIE offset
1819  | 0x........ | uint32_t HashData[3] DIE offset
1820  | 0x00002023 | uint32_t KeyType (.debug_str[0x0002023] => "print")
1821  | 0x00000002 | uint32_t HashData count
1822  | 0x........ | uint32_t HashData[0] DIE offset
1823  | 0x........ | uint32_t HashData[1] DIE offset
1824  | 0x00000000 | uint32_t KeyType (end of hash chain)
1825  `------------'
1826
1827Current testing with real world C++ binaries has shown that there is around 1
182832 bit hash collision per 100,000 name entries.
1829
1830Contents
1831^^^^^^^^
1832
1833As we said, we want to strictly define exactly what is included in the
1834different tables.  For DWARF, we have 3 tables: "``.apple_names``",
1835"``.apple_types``", and "``.apple_namespaces``".
1836
1837"``.apple_names``" sections should contain an entry for each DWARF DIE whose
1838``DW_TAG`` is a ``DW_TAG_label``, ``DW_TAG_inlined_subroutine``, or
1839``DW_TAG_subprogram`` that has address attributes: ``DW_AT_low_pc``,
1840``DW_AT_high_pc``, ``DW_AT_ranges`` or ``DW_AT_entry_pc``.  It also contains
1841``DW_TAG_variable`` DIEs that have a ``DW_OP_addr`` in the location (global and
1842static variables).  All global and static variables should be included,
1843including those scoped within functions and classes.  For example using the
1844following code:
1845
1846.. code-block:: c
1847
1848  static int var = 0;
1849
1850  void f ()
1851  {
1852    static int var = 0;
1853  }
1854
1855Both of the static ``var`` variables would be included in the table.  All
1856functions should emit both their full names and their basenames.  For C or C++,
1857the full name is the mangled name (if available) which is usually in the
1858``DW_AT_MIPS_linkage_name`` attribute, and the ``DW_AT_name`` contains the
1859function basename.  If global or static variables have a mangled name in a
1860``DW_AT_MIPS_linkage_name`` attribute, this should be emitted along with the
1861simple name found in the ``DW_AT_name`` attribute.
1862
1863"``.apple_types``" sections should contain an entry for each DWARF DIE whose
1864tag is one of:
1865
1866* DW_TAG_array_type
1867* DW_TAG_class_type
1868* DW_TAG_enumeration_type
1869* DW_TAG_pointer_type
1870* DW_TAG_reference_type
1871* DW_TAG_string_type
1872* DW_TAG_structure_type
1873* DW_TAG_subroutine_type
1874* DW_TAG_typedef
1875* DW_TAG_union_type
1876* DW_TAG_ptr_to_member_type
1877* DW_TAG_set_type
1878* DW_TAG_subrange_type
1879* DW_TAG_base_type
1880* DW_TAG_const_type
1881* DW_TAG_file_type
1882* DW_TAG_namelist
1883* DW_TAG_packed_type
1884* DW_TAG_volatile_type
1885* DW_TAG_restrict_type
1886* DW_TAG_atomic_type
1887* DW_TAG_interface_type
1888* DW_TAG_unspecified_type
1889* DW_TAG_shared_type
1890
1891Only entries with a ``DW_AT_name`` attribute are included, and the entry must
1892not be a forward declaration (``DW_AT_declaration`` attribute with a non-zero
1893value).  For example, using the following code:
1894
1895.. code-block:: c
1896
1897  int main ()
1898  {
1899    int *b = 0;
1900    return *b;
1901  }
1902
1903We get a few type DIEs:
1904
1905.. code-block:: none
1906
1907  0x00000067:     TAG_base_type [5]
1908                  AT_encoding( DW_ATE_signed )
1909                  AT_name( "int" )
1910                  AT_byte_size( 0x04 )
1911
1912  0x0000006e:     TAG_pointer_type [6]
1913                  AT_type( {0x00000067} ( int ) )
1914                  AT_byte_size( 0x08 )
1915
1916The DW_TAG_pointer_type is not included because it does not have a ``DW_AT_name``.
1917
1918"``.apple_namespaces``" section should contain all ``DW_TAG_namespace`` DIEs.
1919If we run into a namespace that has no name this is an anonymous namespace, and
1920the name should be output as "``(anonymous namespace)``" (without the quotes).
1921Why?  This matches the output of the ``abi::cxa_demangle()`` that is in the
1922standard C++ library that demangles mangled names.
1923
1924
1925Language Extensions and File Format Changes
1926^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
1927
1928Objective-C Extensions
1929""""""""""""""""""""""
1930
1931"``.apple_objc``" section should contain all ``DW_TAG_subprogram`` DIEs for an
1932Objective-C class.  The name used in the hash table is the name of the
1933Objective-C class itself.  If the Objective-C class has a category, then an
1934entry is made for both the class name without the category, and for the class
1935name with the category.  So if we have a DIE at offset 0x1234 with a name of
1936method "``-[NSString(my_additions) stringWithSpecialString:]``", we would add
1937an entry for "``NSString``" that points to DIE 0x1234, and an entry for
1938"``NSString(my_additions)``" that points to 0x1234.  This allows us to quickly
1939track down all Objective-C methods for an Objective-C class when doing
1940expressions.  It is needed because of the dynamic nature of Objective-C where
1941anyone can add methods to a class.  The DWARF for Objective-C methods is also
1942emitted differently from C++ classes where the methods are not usually
1943contained in the class definition, they are scattered about across one or more
1944compile units.  Categories can also be defined in different shared libraries.
1945So we need to be able to quickly find all of the methods and class functions
1946given the Objective-C class name, or quickly find all methods and class
1947functions for a class + category name.  This table does not contain any
1948selector names, it just maps Objective-C class names (or class names +
1949category) to all of the methods and class functions.  The selectors are added
1950as function basenames in the "``.debug_names``" section.
1951
1952In the "``.apple_names``" section for Objective-C functions, the full name is
1953the entire function name with the brackets ("``-[NSString
1954stringWithCString:]``") and the basename is the selector only
1955("``stringWithCString:``").
1956
1957Mach-O Changes
1958""""""""""""""
1959
1960The sections names for the apple hash tables are for non-mach-o files.  For
1961mach-o files, the sections should be contained in the ``__DWARF`` segment with
1962names as follows:
1963
1964* "``.apple_names``" -> "``__apple_names``"
1965* "``.apple_types``" -> "``__apple_types``"
1966* "``.apple_namespaces``" -> "``__apple_namespac``" (16 character limit)
1967* "``.apple_objc``" -> "``__apple_objc``"
1968
1969.. _codeview:
1970
1971CodeView Debug Info Format
1972==========================
1973
1974LLVM supports emitting CodeView, the Microsoft debug info format, and this
1975section describes the design and implementation of that support.
1976
1977Format Background
1978-----------------
1979
1980CodeView as a format is clearly oriented around C++ debugging, and in C++, the
1981majority of debug information tends to be type information. Therefore, the
1982overriding design constraint of CodeView is the separation of type information
1983from other "symbol" information so that type information can be efficiently
1984merged across translation units. Both type information and symbol information is
1985generally stored as a sequence of records, where each record begins with a
198616-bit record size and a 16-bit record kind.
1987
1988Type information is usually stored in the ``.debug$T`` section of the object
1989file.  All other debug info, such as line info, string table, symbol info, and
1990inlinee info, is stored in one or more ``.debug$S`` sections. There may only be
1991one ``.debug$T`` section per object file, since all other debug info refers to
1992it. If a PDB (enabled by the ``/Zi`` MSVC option) was used during compilation,
1993the ``.debug$T`` section will contain only an ``LF_TYPESERVER2`` record pointing
1994to the PDB. When using PDBs, symbol information appears to remain in the object
1995file ``.debug$S`` sections.
1996
1997Type records are referred to by their index, which is the number of records in
1998the stream before a given record plus ``0x1000``. Many common basic types, such
1999as the basic integral types and unqualified pointers to them, are represented
2000using type indices less than ``0x1000``. Such basic types are built in to
2001CodeView consumers and do not require type records.
2002
2003Each type record may only contain type indices that are less than its own type
2004index. This ensures that the graph of type stream references is acyclic. While
2005the source-level type graph may contain cycles through pointer types (consider a
2006linked list struct), these cycles are removed from the type stream by always
2007referring to the forward declaration record of user-defined record types. Only
2008"symbol" records in the ``.debug$S`` streams may refer to complete,
2009non-forward-declaration type records.
2010
2011Working with CodeView
2012---------------------
2013
2014These are instructions for some common tasks for developers working to improve
2015LLVM's CodeView support. Most of them revolve around using the CodeView dumper
2016embedded in ``llvm-readobj``.
2017
2018* Testing MSVC's output::
2019
2020    $ cl -c -Z7 foo.cpp # Use /Z7 to keep types in the object file
2021    $ llvm-readobj --codeview foo.obj
2022
2023* Getting LLVM IR debug info out of Clang::
2024
2025    $ clang -g -gcodeview --target=x86_64-windows-msvc foo.cpp -S -emit-llvm
2026
2027  Use this to generate LLVM IR for LLVM test cases.
2028
2029* Generate and dump CodeView from LLVM IR metadata::
2030
2031    $ llc foo.ll -filetype=obj -o foo.obj
2032    $ llvm-readobj --codeview foo.obj > foo.txt
2033
2034  Use this pattern in lit test cases and FileCheck the output of llvm-readobj
2035
2036Improving LLVM's CodeView support is a process of finding interesting type
2037records, constructing a C++ test case that makes MSVC emit those records,
2038dumping the records, understanding them, and then generating equivalent records
2039in LLVM's backend.
2040