2 @setfilename stabs.info
9 * Stabs:: The "stabs" debugging information format.
15 This document describes the stabs debugging symbol tables.
17 Copyright 1992, 1993 Free Software Foundation, Inc.
18 Contributed by Cygnus Support. Written by Julia Menapace, Jim Kingdon,
21 Permission is granted to make and distribute verbatim copies of
22 this manual provided the copyright notice and this permission notice
23 are preserved on all copies.
26 Permission is granted to process this file through Tex and print the
27 results, provided the printed document carries copying permission
28 notice identical to this one except for the removal of this paragraph
29 (this paragraph not being relevant to the printed manual).
32 Permission is granted to copy or distribute modified versions of this
33 manual under the terms of the GPL (for which purpose this text may be
34 regarded as a program in the language TeX).
37 @setchapternewpage odd
40 @title The ``stabs'' debug format
41 @author Julia Menapace, Jim Kingdon, David MacKenzie
42 @author Cygnus Support
45 \def\$#1${{#1}} % Kluge: collect RCS revision info without $...$
46 \xdef\manvers{\$Revision$} % For use in headers, footers too
48 \hfill Cygnus Support\par
50 \hfill \TeX{}info \texinfoversion\par
54 @vskip 0pt plus 1filll
55 Copyright @copyright{} 1992, 1993 Free Software Foundation, Inc.
56 Contributed by Cygnus Support.
58 Permission is granted to make and distribute verbatim copies of
59 this manual provided the copyright notice and this permission notice
60 are preserved on all copies.
66 @top The "stabs" representation of debugging information
68 This document describes the stabs debugging format.
71 * Overview:: Overview of stabs
72 * Program Structure:: Encoding of the structure of the program
73 * Constants:: Constants
75 * Types:: Type definitions
76 * Symbol Tables:: Symbol information in symbol tables
77 * Cplusplus:: Appendixes:
78 * Stab Types:: Symbol types in a.out files
79 * Symbol Descriptors:: Table of symbol descriptors
80 * Type Descriptors:: Table of type descriptors
81 * Expanded Reference:: Reference information by stab type
82 * Questions:: Questions and anomolies
83 * XCOFF Differences:: Differences between GNU stabs in a.out
84 and GNU stabs in XCOFF
85 * Sun Differences:: Differences between GNU stabs and Sun
87 * Stabs In ELF:: Stabs in an ELF file.
88 * Symbol Types Index:: Index of symbolic stab symbol type names.
94 @chapter Overview of Stabs
96 @dfn{Stabs} refers to a format for information that describes a program
97 to a debugger. This format was apparently invented by
98 @c FIXME! <<name of inventor>> at
99 the University of California at Berkeley, for the @code{pdx} Pascal
100 debugger; the format has spread widely since then.
102 This document is one of the few published sources of documentation on
103 stabs. It is believed to be comprehensive for stabs used by C. The
104 lists of symbol descriptors (@pxref{Symbol Descriptors}) and type
105 descriptors (@pxref{Type Descriptors}) are believed to be completely
106 comprehensive. Stabs for COBOL-specific features and for variant
107 records (used by Pascal and Modula-2) are poorly documented here.
109 Other sources of information on stabs are @cite{Dbx and Dbxtool
110 Interfaces}, 2nd edition, by Sun, 1988, and @cite{AIX Version 3.2 Files
111 Reference}, Fourth Edition, September 1992, "dbx Stabstring Grammar" in
112 the a.out section, page 2-31. This document is believed to incorporate
113 the information from those two sources except where it explictly directs
114 you to them for more information.
117 * Flow:: Overview of debugging information flow
118 * Stabs Format:: Overview of stab format
119 * String Field:: The string field
120 * C Example:: A simple example in C source
121 * Assembly Code:: The simple example at the assembly level
125 @section Overview of Debugging Information Flow
127 The GNU C compiler compiles C source in a @file{.c} file into assembly
128 language in a @file{.s} file, which the assembler translates into
129 a @file{.o} file, which the linker combines with other @file{.o} files and
130 libraries to produce an executable file.
132 With the @samp{-g} option, GCC puts in the @file{.s} file additional
133 debugging information, which is slightly transformed by the assembler
134 and linker, and carried through into the final executable. This
135 debugging information describes features of the source file like line
136 numbers, the types and scopes of variables, and function names,
137 parameters, and scopes.
139 For some object file formats, the debugging information is encapsulated
140 in assembler directives known collectively as @dfn{stab} (symbol table)
141 directives, which are interspersed with the generated code. Stabs are
142 the native format for debugging information in the a.out and XCOFF
143 object file formats. The GNU tools can also emit stabs in the COFF and
144 ECOFF object file formats.
146 The assembler adds the information from stabs to the symbol information
147 it places by default in the symbol table and the string table of the
148 @file{.o} file it is building. The linker consolidates the @file{.o}
149 files into one executable file, with one symbol table and one string
150 table. Debuggers use the symbol and string tables in the executable as
151 a source of debugging information about the program.
154 @section Overview of Stab Format
156 There are three overall formats for stab assembler directives,
157 differentiated by the first word of the stab. The name of the directive
158 describes which combination of four possible data fields follows. It is
159 either @code{.stabs} (string), @code{.stabn} (number), or @code{.stabd}
160 (dot). IBM's XCOFF assembler uses @code{.stabx} (and some other
161 directives such as @code{.file} and @code{.bi}) instead of
162 @code{.stabs}, @code{.stabn} or @code{.stabd}.
164 The overall format of each class of stab is:
167 .stabs "@var{string}",@var{type},@var{other},@var{desc},@var{value}
168 .stabn @var{type},@var{other},@var{desc},@var{value}
169 .stabd @var{type},@var{other},@var{desc}
170 .stabx "@var{string}",@var{value},@var{type},@var{sdb-type}
173 @c what is the correct term for "current file location"? My AIX
174 @c assembler manual calls it "the value of the current location counter".
175 For @code{.stabn} and @code{.stabd}, there is no @var{string} (the
176 @code{n_strx} field is zero; see @ref{Symbol Tables}). For
177 @code{.stabd}, the @var{value} field is implicit and has the value of
178 the current file location. For @code{.stabx}, the @var{sdb-type} field
179 is unused for stabs and can always be set to zero. The @var{other}
180 field is almost always unused and can be set to zero.
182 The number in the @var{type} field gives some basic information about
183 which type of stab this is (or whether it @emph{is} a stab, as opposed
184 to an ordinary symbol). Each valid type number defines a different stab
185 type; further, the stab type defines the exact interpretation of, and
186 possible values for, any remaining @var{string}, @var{desc}, or
187 @var{value} fields present in the stab. @xref{Stab Types}, for a list
188 in numeric order of the valid @var{type} field values for stab directives.
191 @section The String Field
193 For most stabs the string field holds the meat of the
194 debugging information. The flexible nature of this field
195 is what makes stabs extensible. For some stab types the string field
196 contains only a name. For other stab types the contents can be a great
199 The overall format of the string field for most stab types is:
202 "@var{name}:@var{symbol-descriptor} @var{type-information}"
205 @var{name} is the name of the symbol represented by the stab.
206 @var{name} can be omitted, which means the stab represents an unnamed
207 object. For example, @samp{:t10=*2} defines type 10 as a pointer to
208 type 2, but does not give the type a name. Omitting the @var{name}
209 field is supported by AIX dbx and GDB after about version 4.8, but not
210 other debuggers. GCC sometimes uses a single space as the name instead
211 of omitting the name altogether; apparently that is supported by most
214 The @var{symbol-descriptor} following the @samp{:} is an alphabetic
215 character that tells more specifically what kind of symbol the stab
216 represents. If the @var{symbol-descriptor} is omitted, but type
217 information follows, then the stab represents a local variable. For a
218 list of symbol descriptors, see @ref{Symbol Descriptors}. The @samp{c}
219 symbol descriptor is an exception in that it is not followed by type
220 information. @xref{Constants}.
222 @var{type-information} is either a @var{type-number}, or
223 @samp{@var{type-number}=}. A @var{type-number} alone is a type
224 reference, referring directly to a type that has already been defined.
226 The @samp{@var{type-number}=} form is a type definition, where the
227 number represents a new type which is about to be defined. The type
228 definition may refer to other types by number, and those type numbers
229 may be followed by @samp{=} and nested definitions.
231 In a type definition, if the character that follows the equals sign is
232 non-numeric then it is a @var{type-descriptor}, and tells what kind of
233 type is about to be defined. Any other values following the
234 @var{type-descriptor} vary, depending on the @var{type-descriptor}.
235 @xref{Type Descriptors}, for a list of @var{type-descriptor} values. If
236 a number follows the @samp{=} then the number is a @var{type-reference}.
237 For a full description of types, @ref{Types}.
239 There is an AIX extension for type attributes. Following the @samp{=}
240 are any number of type attributes. Each one starts with @samp{@@} and
241 ends with @samp{;}. Debuggers, including AIX's dbx and GDB 4.10, skip
242 any type attributes they do not recognize. GDB 4.9 and other versions
243 of dbx may not do this. Because of a conflict with C++
244 (@pxref{Cplusplus}), new attributes should not be defined which begin
245 with a digit, @samp{(}, or @samp{-}; GDB may be unable to distinguish
246 those from the C++ type descriptor @samp{@@}. The attributes are:
249 @item a@var{boundary}
250 @var{boundary} is an integer specifying the alignment. I assume it
251 applies to all variables of this type.
254 Size in bits of a variable of this type.
257 Pointer class (for checking). Not sure what this means, or how
258 @var{integer} is interpreted.
261 Indicate this is a packed type, meaning that structure fields or array
262 elements are placed more closely in memory, to save memory at the
266 All of this can make the string field quite long. All
267 versions of GDB, and some versions of dbx, can handle arbitrarily long
268 strings. But many versions of dbx cretinously limit the strings to
269 about 80 characters, so compilers which must work with such dbx's need
270 to split the @code{.stabs} directive into several @code{.stabs}
271 directives. Each stab duplicates exactly all but the
272 string field. The string field of
273 every stab except the last is marked as continued with a
274 double-backslash at the end. Removing the backslashes and concatenating
275 the string fields of each stab produces the original,
279 @section A Simple Example in C Source
281 To get the flavor of how stabs describe source information for a C
282 program, let's look at the simple program:
287 printf("Hello world");
291 When compiled with @samp{-g}, the program above yields the following
292 @file{.s} file. Line numbers have been added to make it easier to refer
293 to parts of the @file{.s} file in the description of the stabs that
297 @section The Simple Example at the Assembly Level
299 This simple ``hello world'' example demonstrates several of the stab
300 types used to describe C language source files.
304 2 .stabs "/cygint/s1/users/jcm/play/",100,0,0,Ltext0
305 3 .stabs "hello.c",100,0,0,Ltext0
308 6 .stabs "int:t1=r1;-2147483648;2147483647;",128,0,0,0
309 7 .stabs "char:t2=r2;0;127;",128,0,0,0
310 8 .stabs "long int:t3=r1;-2147483648;2147483647;",128,0,0,0
311 9 .stabs "unsigned int:t4=r1;0;-1;",128,0,0,0
312 10 .stabs "long unsigned int:t5=r1;0;-1;",128,0,0,0
313 11 .stabs "short int:t6=r1;-32768;32767;",128,0,0,0
314 12 .stabs "long long int:t7=r1;0;-1;",128,0,0,0
315 13 .stabs "short unsigned int:t8=r1;0;65535;",128,0,0,0
316 14 .stabs "long long unsigned int:t9=r1;0;-1;",128,0,0,0
317 15 .stabs "signed char:t10=r1;-128;127;",128,0,0,0
318 16 .stabs "unsigned char:t11=r1;0;255;",128,0,0,0
319 17 .stabs "float:t12=r1;4;0;",128,0,0,0
320 18 .stabs "double:t13=r1;8;0;",128,0,0,0
321 19 .stabs "long double:t14=r1;8;0;",128,0,0,0
322 20 .stabs "void:t15=15",128,0,0,0
325 23 .ascii "Hello, world!\12\0"
340 38 sethi %hi(LC0),%o1
341 39 or %o1,%lo(LC0),%o0
352 50 .stabs "main:F1",36,0,0,_main
353 51 .stabn 192,0,0,LBB2
354 52 .stabn 224,0,0,LBE2
357 @node Program Structure
358 @chapter Encoding the Structure of the Program
360 The elements of the program structure that stabs encode include the name
361 of the main function, the names of the source and include files, the
362 line numbers, procedure names and types, and the beginnings and ends of
366 * Main Program:: Indicate what the main program is
367 * Source Files:: The path and name of the source file
368 * Include Files:: Names of include files
371 * Nested Procedures::
376 @section Main Program
379 Most languages allow the main program to have any name. The
380 @code{N_MAIN} stab type tells the debugger the name that is used in this
381 program. Only the string field is significant; it is the name of
382 a function which is the main program. Most C compilers do not use this
383 stab (they expect the debugger to assume that the name is @code{main}),
384 but some C compilers emit an @code{N_MAIN} stab for the @code{main}
388 @section Paths and Names of the Source Files
391 Before any other stabs occur, there must be a stab specifying the source
392 file. This information is contained in a symbol of stab type
393 @code{N_SO}; the string field contains the name of the file. The
394 value of the symbol is the start address of the portion of the
395 text section corresponding to that file.
397 With the Sun Solaris2 compiler, the desc field contains a
398 source-language code.
399 @c Do the debuggers use it? What are the codes? -djm
401 Some compilers (for example, GCC2 and SunOS4 @file{/bin/cc}) also
402 include the directory in which the source was compiled, in a second
403 @code{N_SO} symbol preceding the one containing the file name. This
404 symbol can be distinguished by the fact that it ends in a slash. Code
405 from the @code{cfront} C++ compiler can have additional @code{N_SO} symbols for
406 nonexistent source files after the @code{N_SO} for the real source file;
407 these are believed to contain no useful information.
412 .stabs "/cygint/s1/users/jcm/play/",100,0,0,Ltext0 # @r{100 is N_SO}
413 .stabs "hello.c",100,0,0,Ltext0
418 Instead of @code{N_SO} symbols, XCOFF uses a @code{.file} assembler
419 directive which assembles to a standard COFF @code{.file} symbol;
420 explaining this in detail is outside the scope of this document.
423 @section Names of Include Files
425 There are several schemes for dealing with include files: the
426 traditional @code{N_SOL} approach, Sun's @code{N_BINCL} approach, and the
427 XCOFF @code{C_BINCL} approach (which despite the similar name has little in
428 common with @code{N_BINCL}).
431 An @code{N_SOL} symbol specifies which include file subsequent symbols
432 refer to. The string field is the name of the file and the value is the
433 text address corresponding to the end of the previous include file and
434 the start of this one. To specify the main source file again, use an
435 @code{N_SOL} symbol with the name of the main source file.
440 The @code{N_BINCL} approach works as follows. An @code{N_BINCL} symbol
441 specifies the start of an include file. In an object file, only the
442 string is significant; the Sun linker puts data into some of the
443 other fields. The end of the include file is marked by an
444 @code{N_EINCL} symbol (which has no string field). In an object
445 file, there is no significant data in the @code{N_EINCL} symbol; the Sun
446 linker puts data into some of the fields. @code{N_BINCL} and
447 @code{N_EINCL} can be nested.
449 If the linker detects that two source files have identical stabs between
450 an @code{N_BINCL} and @code{N_EINCL} pair (as will generally be the case
451 for a header file), then it only puts out the stabs once. Each
452 additional occurance is replaced by an @code{N_EXCL} symbol. I believe
453 the Sun (SunOS4, not sure about Solaris) linker is the only one which
454 supports this feature.
455 @c What do the fields of N_EXCL contain? -djm
459 For the start of an include file in XCOFF, use the @file{.bi} assembler
460 directive, which generates a @code{C_BINCL} symbol. A @file{.ei}
461 directive, which generates a @code{C_EINCL} symbol, denotes the end of
462 the include file. Both directives are followed by the name of the
463 source file in quotes, which becomes the string for the symbol.
464 The value of each symbol, produced automatically by the assembler
465 and linker, is the offset into the executable of the beginning
466 (inclusive, as you'd expect) or end (inclusive, as you would not expect)
467 of the portion of the COFF line table that corresponds to this include
468 file. @code{C_BINCL} and @code{C_EINCL} do not nest.
471 @section Line Numbers
474 An @code{N_SLINE} symbol represents the start of a source line. The
475 desc field contains the line number and the value
476 contains the code address for the start of that source line. On most
477 machines the address is absolute; for Sun's stabs-in-ELF, it is relative
478 to the function in which the @code{N_SLINE} symbol occurs.
482 GNU documents @code{N_DSLINE} and @code{N_BSLINE} symbols for line
483 numbers in the data or bss segments, respectively. They are identical
484 to @code{N_SLINE} but are relocated differently by the linker. They
485 were intended to be used to describe the source location of a variable
486 declaration, but I believe that GCC2 actually puts the line number in
487 the desc field of the stab for the variable itself. GDB has been
488 ignoring these symbols (unless they contain a string field) since
491 For single source lines that generate discontiguous code, such as flow
492 of control statements, there may be more than one line number entry for
493 the same source line. In this case there is a line number entry at the
494 start of each code range, each with the same line number.
496 XCOFF does not use stabs for line numbers. Instead, it uses COFF line
497 numbers (which are outside the scope of this document). Standard COFF
498 line numbers cannot deal with include files, but in XCOFF this is fixed
499 with the @code{C_BINCL} method of marking include files (@pxref{Include
505 @findex N_FUN, for functions
507 @findex N_STSYM, for functions (Sun acc)
508 @findex N_GSYM, for functions (Sun acc)
509 All of the following stabs normally use the @code{N_FUN} symbol type.
510 However, Sun's @code{acc} compiler on SunOS4 uses @code{N_GSYM} and
511 @code{N_STSYM}, which means that the value of the stab for the function
512 is useless and the debugger must get the address of the function from
513 the non-stab symbols instead. BSD Fortran is said to use @code{N_FNAME}
514 with the same restriction; the value of the symbol is not useful (I'm
515 not sure it really does use this, because GDB doesn't handle this and no
518 A function is represented by an @samp{F} symbol descriptor for a global
519 (extern) function, and @samp{f} for a static (local) function. The
520 value is the address of the start of the function. For @code{a.out}, it
521 is already relocated. For stabs in ELF, the SunPRO compiler version
522 2.0.1 and GCC put out an address which gets relocated by the linker. In
523 a future release SunPRO is planning to put out zero, in which case the
524 address can be found from the ELF (non-stab) symbol. Because looking
525 things up in the ELF symbols would probably be slow, I'm not sure how to
526 find which symbol of that name is the right one, and this doesn't
527 provide any way to deal with nested functions, it would probably be
528 better to make the value of the stab an address relative to the start of
529 the file. See @ref{Stabs In ELF} for more information on linker
530 relocation of stabs in ELF files.
532 The type information of the stab represents the return type of the
533 function; thus @samp{foo:f5} means that foo is a function returning type
534 5. There is no need to try to get the line number of the start of the
535 function from the stab for the function; it is in the next
536 @code{N_SLINE} symbol.
538 @c FIXME: verify whether the "I suspect" below is true or not.
539 Some compilers (such as Sun's Solaris compiler) support an extension for
540 specifying the types of the arguments. I suspect this extension is not
541 used for old (non-prototyped) function definitions in C. If the
542 extension is in use, the type information of the stab for the function
543 is followed by type information for each argument, with each argument
544 preceded by @samp{;}. An argument type of 0 means that additional
545 arguments are being passed, whose types and number may vary (@samp{...}
546 in ANSI C). GDB has tolerated this extension (parsed the syntax, if not
547 necessarily used the information) since at least version 4.8; I don't
548 know whether all versions of dbx tolerate it. The argument types given
549 here are not redundant with the symbols for the formal parameters
550 (@pxref{Parameters}); they are the types of the arguments as they are
551 passed, before any conversions might take place. For example, if a C
552 function which is declared without a prototype takes a @code{float}
553 argument, the value is passed as a @code{double} but then converted to a
554 @code{float}. Debuggers need to use the types given in the arguments
555 when printing values, but when calling the function they need to use the
556 types given in the symbol defining the function.
558 If the return type and types of arguments of a function which is defined
559 in another source file are specified (i.e., a function prototype in ANSI
560 C), traditionally compilers emit no stab; the only way for the debugger
561 to find the information is if the source file where the function is
562 defined was also compiled with debugging symbols. As an extension the
563 Solaris compiler uses symbol descriptor @samp{P} followed by the return
564 type of the function, followed by the arguments, each preceded by
565 @samp{;}, as in a stab with symbol descriptor @samp{f} or @samp{F}.
566 This use of symbol descriptor @samp{P} can be distinguished from its use
567 for register parameters (@pxref{Register Parameters}) by the fact that it has
568 symbol type @code{N_FUN}.
570 The AIX documentation also defines symbol descriptor @samp{J} as an
571 internal function. I assume this means a function nested within another
572 function. It also says symbol descriptor @samp{m} is a module in
573 Modula-2 or extended Pascal.
575 Procedures (functions which do not return values) are represented as
576 functions returning the @code{void} type in C. I don't see why this couldn't
577 be used for all languages (inventing a @code{void} type for this purpose if
578 necessary), but the AIX documentation defines @samp{I}, @samp{P}, and
579 @samp{Q} for internal, global, and static procedures, respectively.
580 These symbol descriptors are unusual in that they are not followed by
583 The following example shows a stab for a function @code{main} which
584 returns type number @code{1}. The @code{_main} specified for the value
585 is a reference to an assembler label which is used to fill in the start
586 address of the function.
589 .stabs "main:F1",36,0,0,_main # @r{36 is N_FUN}
592 The stab representing a procedure is located immediately following the
593 code of the procedure. This stab is in turn directly followed by a
594 group of other stabs describing elements of the procedure. These other
595 stabs describe the procedure's parameters, its block local variables, and
598 @node Nested Procedures
599 @section Nested Procedures
601 For any of the symbol descriptors representing procedures, after the
602 symbol descriptor and the type information is optionally a scope
603 specifier. This consists of a comma, the name of the procedure, another
604 comma, and the name of the enclosing procedure. The first name is local
605 to the scope specified, and seems to be redundant with the name of the
606 symbol (before the @samp{:}). This feature is used by GCC, and
607 presumably Pascal, Modula-2, etc., compilers, for nested functions.
609 If procedures are nested more than one level deep, only the immediately
610 containing scope is specified. For example, this code:
622 return baz (x + 2 * y);
624 return x + bar (3 * x);
632 .stabs "baz:f1,baz,bar",36,0,0,_baz.15 # @r{36 is N_FUN}
633 .stabs "bar:f1,bar,foo",36,0,0,_bar.12
634 .stabs "foo:F1",36,0,0,_foo
637 @node Block Structure
638 @section Block Structure
642 The program's block structure is represented by the @code{N_LBRAC} (left
643 brace) and the @code{N_RBRAC} (right brace) stab types. The variables
644 defined inside a block precede the @code{N_LBRAC} symbol for most
645 compilers, including GCC. Other compilers, such as the Convex, Acorn
646 RISC machine, and Sun @code{acc} compilers, put the variables after the
647 @code{N_LBRAC} symbol. The values of the @code{N_LBRAC} and
648 @code{N_RBRAC} symbols are the start and end addresses of the code of
649 the block, respectively. For most machines, they are relative to the
650 starting address of this source file. For the Gould NP1, they are
651 absolute. For Sun's stabs-in-ELF, they are relative to the function in
654 The @code{N_LBRAC} and @code{N_RBRAC} stabs that describe the block
655 scope of a procedure are located after the @code{N_FUN} stab that
656 represents the procedure itself.
658 Sun documents the desc field of @code{N_LBRAC} and
659 @code{N_RBRAC} symbols as containing the nesting level of the block.
660 However, dbx seems to not care, and GCC always sets desc to
666 The @samp{c} symbol descriptor indicates that this stab represents a
667 constant. This symbol descriptor is an exception to the general rule
668 that symbol descriptors are followed by type information. Instead, it
669 is followed by @samp{=} and one of the following:
673 Boolean constant. @var{value} is a numeric value; I assume it is 0 for
677 Character constant. @var{value} is the numeric value of the constant.
679 @item e @var{type-information} , @var{value}
680 Constant whose value can be represented as integral.
681 @var{type-information} is the type of the constant, as it would appear
682 after a symbol descriptor (@pxref{String Field}). @var{value} is the
683 numeric value of the constant. GDB 4.9 does not actually get the right
684 value if @var{value} does not fit in a host @code{int}, but it does not
685 do anything violent, and future debuggers could be extended to accept
686 integers of any size (whether unsigned or not). This constant type is
687 usually documented as being only for enumeration constants, but GDB has
688 never imposed that restriction; I don't know about other debuggers.
691 Integer constant. @var{value} is the numeric value. The type is some
692 sort of generic integer type (for GDB, a host @code{int}); to specify
693 the type explicitly, use @samp{e} instead.
696 Real constant. @var{value} is the real value, which can be @samp{INF}
697 (optionally preceded by a sign) for infinity, @samp{QNAN} for a quiet
698 NaN (not-a-number), or @samp{SNAN} for a signalling NaN. If it is a
699 normal number the format is that accepted by the C library function
703 String constant. @var{string} is a string enclosed in either @samp{'}
704 (in which case @samp{'} characters within the string are represented as
705 @samp{\'} or @samp{"} (in which case @samp{"} characters within the
706 string are represented as @samp{\"}).
708 @item S @var{type-information} , @var{elements} , @var{bits} , @var{pattern}
709 Set constant. @var{type-information} is the type of the constant, as it
710 would appear after a symbol descriptor (@pxref{String Field}).
711 @var{elements} is the number of elements in the set (does this means
712 how many bits of @var{pattern} are actually used, which would be
713 redundant with the type, or perhaps the number of bits set in
714 @var{pattern}? I don't get it), @var{bits} is the number of bits in the
715 constant (meaning it specifies the length of @var{pattern}, I think),
716 and @var{pattern} is a hexadecimal representation of the set. AIX
717 documentation refers to a limit of 32 bytes, but I see no reason why
718 this limit should exist. This form could probably be used for arbitrary
719 constants, not just sets; the only catch is that @var{pattern} should be
720 understood to be target, not host, byte order and format.
723 The boolean, character, string, and set constants are not supported by
724 GDB 4.9, but it ignores them. GDB 4.8 and earlier gave an error
725 message and refused to read symbols from the file containing the
728 The above information is followed by @samp{;}.
733 Different types of stabs describe the various ways that variables can be
734 allocated: on the stack, globally, in registers, in common blocks,
735 statically, or as arguments to a function.
738 * Stack Variables:: Variables allocated on the stack.
739 * Global Variables:: Variables used by more than one source file.
740 * Register Variables:: Variables in registers.
741 * Common Blocks:: Variables statically allocated together.
742 * Statics:: Variables local to one source file.
743 * Based Variables:: Fortran pointer based variables.
744 * Parameters:: Variables for arguments to functions.
747 @node Stack Variables
748 @section Automatic Variables Allocated on the Stack
750 If a variable's scope is local to a function and its lifetime is only as
751 long as that function executes (C calls such variables
752 @dfn{automatic}), it can be allocated in a register (@pxref{Register
753 Variables}) or on the stack.
756 Each variable allocated on the stack has a stab with the symbol
757 descriptor omitted. Since type information should begin with a digit,
758 @samp{-}, or @samp{(}, only those characters precluded from being used
759 for symbol descriptors. However, the Acorn RISC machine (ARM) is said
760 to get this wrong: it puts out a mere type definition here, without the
761 preceding @samp{@var{type-number}=}. This is a bad idea; there is no
762 guarantee that type descriptors are distinct from symbol descriptors.
763 Stabs for stack variables use the @code{N_LSYM} stab type.
765 The value of the stab is the offset of the variable within the
766 local variables. On most machines this is an offset from the frame
767 pointer and is negative. The location of the stab specifies which block
768 it is defined in; see @ref{Block Structure}.
770 For example, the following C code:
780 produces the following stabs:
783 .stabs "main:F1",36,0,0,_main # @r{36 is N_FUN}
784 .stabs "x:1",128,0,0,-12 # @r{128 is N_LSYM}
785 .stabn 192,0,0,LBB2 # @r{192 is N_LBRAC}
786 .stabn 224,0,0,LBE2 # @r{224 is N_RBRAC}
789 @xref{Procedures} for more information on the @code{N_FUN} stab, and
790 @ref{Block Structure} for more information on the @code{N_LBRAC} and
791 @code{N_RBRAC} stabs.
793 @node Global Variables
794 @section Global Variables
797 A variable whose scope is not specific to just one source file is
798 represented by the @samp{G} symbol descriptor. These stabs use the
799 @code{N_GSYM} stab type. The type information for the stab
800 (@pxref{String Field}) gives the type of the variable.
802 For example, the following source code:
809 yields the following assembly code:
812 .stabs "g_foo:G2",32,0,0,0 # @r{32 is N_GSYM}
819 The address of the variable represented by the @code{N_GSYM} is not
820 contained in the @code{N_GSYM} stab. The debugger gets this information
821 from the external symbol for the global variable. In the example above,
822 the @code{.global _g_foo} and @code{_g_foo:} lines tell the assembler to
823 produce an external symbol.
825 @node Register Variables
826 @section Register Variables
829 @c According to an old version of this manual, AIX uses C_RPSYM instead
830 @c of C_RSYM. I am skeptical; this should be verified.
831 Register variables have their own stab type, @code{N_RSYM}, and their
832 own symbol descriptor, @samp{r}. The stab's value is the
833 number of the register where the variable data will be stored.
834 @c .stabs "name:type",N_RSYM,0,RegSize,RegNumber (Sun doc)
836 AIX defines a separate symbol descriptor @samp{d} for floating point
837 registers. This seems unnecessary; why not just just give floating
838 point registers different register numbers? I have not verified whether
839 the compiler actually uses @samp{d}.
841 If the register is explicitly allocated to a global variable, but not
845 register int g_bar asm ("%g5");
849 then the stab may be emitted at the end of the object file, with
850 the other bss symbols.
853 @section Common Blocks
855 A common block is a statically allocated section of memory which can be
856 referred to by several source files. It may contain several variables.
857 I believe Fortran is the only language with this feature.
863 A @code{N_BCOMM} stab begins a common block and an @code{N_ECOMM} stab
864 ends it. The only field that is significant in these two stabs is the
865 string, which names a normal (non-debugging) symbol that gives the
866 address of the common block. According to IBM documentation, only the
867 @code{N_BCOMM} has the name of the common block (even though their
868 compiler actually puts it both places).
872 The stabs for the members of the common block are between the
873 @code{N_BCOMM} and the @code{N_ECOMM}; the value of each stab is the
874 offset within the common block of that variable. IBM uses the
875 @code{C_ECOML} stab type, and there is a corresponding @code{N_ECOML}
876 stab type, but Sun's Fortran compiler uses @code{N_GSYM} instead. The
877 variables within a common block use the @samp{V} symbol descriptor (I
878 believe this is true of all Fortran variables). Other stabs (at least
879 type declarations using @code{C_DECL}) can also be between the
880 @code{N_BCOMM} and the @code{N_ECOMM}.
883 @section Static Variables
885 Initialized static variables are represented by the @samp{S} and
886 @samp{V} symbol descriptors. @samp{S} means file scope static, and
887 @samp{V} means procedure scope static.
889 @c This is probably not worth mentioning; it is only true on the sparc
890 @c for `double' variables which although declared const are actually in
891 @c the data segment (the text segment can't guarantee 8 byte alignment).
893 @c 2.4.5 has a bug in that it uses @code{N_FUN}, so neither dbx nor GDB can
894 @c find the variables)
897 @findex N_FUN, for variables
899 In a.out files, @code{N_STSYM} means the data section, @code{N_FUN}
900 means the text section, and @code{N_LCSYM} means the bss section. For
901 those systems with a read-only data section separate from the text
902 section (Solaris), @code{N_ROSYM} means the read-only data section.
904 For example, the source lines:
907 static const int var_const = 5;
908 static int var_init = 2;
909 static int var_noinit;
913 yield the following stabs:
916 .stabs "var_const:S1",36,0,0,_var_const # @r{36 is N_FUN}
918 .stabs "var_init:S1",38,0,0,_var_init # @r{38 is N_STSYM}
920 .stabs "var_noinit:S1",40,0,0,_var_noinit # @r{40 is N_LCSYM}
923 In XCOFF files, each symbol has a section number, so the stab type
924 need not indicate the section.
926 In ECOFF files, the storage class is used to specify the section, so the
927 stab type need not indicate the section.
929 In ELF files, for the SunPRO compiler version 2.0.1, symbol descriptor
930 @samp{S} means that the address is absolute (the linker relocates it)
931 and symbol descriptor @samp{V} means that the address is relative to the
932 start of the relevant section for that compilation unit. SunPRO has
933 plans to have the linker stop relocating stabs; I suspect that their the
934 debugger gets the address from the corresponding ELF (not stab) symbol.
935 I'm not sure how to find which symbol of that name is the right one.
936 The clean way to do all this would be to have a the value of a symbol
937 descriptor @samp{S} symbol be an offset relative to the start of the
938 file, just like everything else, but that introduces obvious
939 compatibility problems. For more information on linker stab relocation,
942 @node Based Variables
943 @section Fortran Based Variables
945 Fortran (at least, the Sun and SGI dialects of FORTRAN-77) has a feature
946 which allows allocating arrays with @code{malloc}, but which avoids
947 blurring the line between arrays and pointers the way that C does. In
948 stabs such a variable uses the @samp{b} symbol descriptor.
950 For example, the Fortran declarations
953 real foo, foo10(10), foo10_5(10,5)
955 pointer (foo10p, foo10)
956 pointer (foo105p, foo10_5)
964 foo10_5:bar3;1;5;ar3;1;10;6
967 In this example, @code{real} is type 6 and type 3 is an integral type
968 which is the type of the subscripts of the array (probably
971 The @samp{b} symbol descriptor is like @samp{V} in that it denotes a
972 statically allocated symbol whose scope is local to a function; see
973 @xref{Statics}. The value of the symbol, instead of being the address
974 of the variable itself, is the address of a pointer to that variable.
975 So in the above example, the value of the @code{foo} stab is the address
976 of a pointer to a real, the value of the @code{foo10} stab is the
977 address of a pointer to a 10-element array of reals, and the value of
978 the @code{foo10_5} stab is the address of a pointer to a 5-element array
979 of 10-element arrays of reals.
984 Formal parameters to a function are represented by a stab (or sometimes
985 two; see below) for each parameter. The stabs are in the order in which
986 the debugger should print the parameters (i.e., the order in which the
987 parameters are declared in the source file). The exact form of the stab
988 depends on how the parameter is being passed.
991 Parameters passed on the stack use the symbol descriptor @samp{p} and
992 the @code{N_PSYM} symbol type. The value of the symbol is an offset
993 used to locate the parameter on the stack; its exact meaning is
994 machine-dependent, but on most machines it is an offset from the frame
997 As a simple example, the code:
1008 .stabs "main:F1",36,0,0,_main # @r{36 is N_FUN}
1009 .stabs "argc:p1",160,0,0,68 # @r{160 is N_PSYM}
1010 .stabs "argv:p20=*21=*2",160,0,0,72
1013 The type definition of @code{argv} is interesting because it contains
1014 several type definitions. Type 21 is pointer to type 2 (char) and
1015 @code{argv} (type 20) is pointer to type 21.
1017 @c FIXME: figure out what these mean and describe them coherently.
1018 The following symbol descriptors are also said to go with @code{N_PSYM}.
1019 The value of the symbol is said to be an offset from the argument
1020 pointer (I'm not sure whether this is true or not).
1024 pF Fortran function parameter
1025 X (function result variable)
1029 * Register Parameters::
1030 * Local Variable Parameters::
1031 * Reference Parameters::
1032 * Conformant Arrays::
1035 @node Register Parameters
1036 @subsection Passing Parameters in Registers
1038 If the parameter is passed in a register, then traditionally there are
1039 two symbols for each argument:
1042 .stabs "arg:p1" . . . ; N_PSYM
1043 .stabs "arg:r1" . . . ; N_RSYM
1046 Debuggers use the second one to find the value, and the first one to
1047 know that it is an argument.
1050 @findex N_RSYM, for parameters
1051 Because that approach is kind of ugly, some compilers use symbol
1052 descriptor @samp{P} or @samp{R} to indicate an argument which is in a
1053 register. Symbol type @code{C_RPSYM} is used with @samp{R} and
1054 @code{N_RSYM} is used with @samp{P}. The symbol's value is
1055 the register number. @samp{P} and @samp{R} mean the same thing; the
1056 difference is that @samp{P} is a GNU invention and @samp{R} is an IBM
1057 (XCOFF) invention. As of version 4.9, GDB should handle either one.
1059 There is at least one case where GCC uses a @samp{p} and @samp{r} pair
1060 rather than @samp{P}; this is where the argument is passed in the
1061 argument list and then loaded into a register.
1063 According to the AIX documentation, symbol descriptor @samp{D} is for a
1064 parameter passed in a floating point register. This seems
1065 unnecessary---why not just use @samp{R} with a register number which
1066 indicates that it's a floating point register? I haven't verified
1067 whether the system actually does what the documentation indicates.
1069 @c FIXME: On the hppa this is for any type > 8 bytes, I think, and not
1070 @c for small structures (investigate).
1071 On the sparc and hppa, for a @samp{P} symbol whose type is a structure
1072 or union, the register contains the address of the structure. On the
1073 sparc, this is also true of a @samp{p} and @samp{r} pair (using Sun
1074 @code{cc}) or a @samp{p} symbol. However, if a (small) structure is
1075 really in a register, @samp{r} is used. And, to top it all off, on the
1076 hppa it might be a structure which was passed on the stack and loaded
1077 into a register and for which there is a @samp{p} and @samp{r} pair! I
1078 believe that symbol descriptor @samp{i} is supposed to deal with this
1079 case (it is said to mean "value parameter by reference, indirect
1080 access"; I don't know the source for this information), but I don't know
1081 details or what compilers or debuggers use it, if any (not GDB or GCC).
1082 It is not clear to me whether this case needs to be dealt with
1083 differently than parameters passed by reference (@pxref{Reference Parameters}).
1085 @node Local Variable Parameters
1086 @subsection Storing Parameters as Local Variables
1088 There is a case similar to an argument in a register, which is an
1089 argument that is actually stored as a local variable. Sometimes this
1090 happens when the argument was passed in a register and then the compiler
1091 stores it as a local variable. If possible, the compiler should claim
1092 that it's in a register, but this isn't always done.
1094 If a parameter is passed as one type and converted to a smaller type by
1095 the prologue (for example, the parameter is declared as a @code{float},
1096 but the calling conventions specify that it is passed as a
1097 @code{double}), then GCC2 (sometimes) uses a pair of symbols. The first
1098 symbol uses symbol descriptor @samp{p} and the type which is passed.
1099 The second symbol has the type and location which the parameter actually
1100 has after the prologue. For example, suppose the following C code
1101 appears with no prototypes involved:
1110 if @code{f} is passed as a double at stack offset 8, and the prologue
1111 converts it to a float in register number 0, then the stabs look like:
1114 .stabs "f:p13",160,0,3,8 # @r{160 is @code{N_PSYM}, here 13 is @code{double}}
1115 .stabs "f:r12",64,0,3,0 # @r{64 is @code{N_RSYM}, here 12 is @code{float}}
1118 In both stabs 3 is the line number where @code{f} is declared
1119 (@pxref{Line Numbers}).
1121 @findex N_LSYM, for parameter
1122 GCC, at least on the 960, has another solution to the same problem. It
1123 uses a single @samp{p} symbol descriptor for an argument which is stored
1124 as a local variable but uses @code{N_LSYM} instead of @code{N_PSYM}. In
1125 this case, the value of the symbol is an offset relative to the local
1126 variables for that function, not relative to the arguments; on some
1127 machines those are the same thing, but not on all.
1129 @c This is mostly just background info; the part that logically belongs
1130 @c here is the last sentence.
1131 On the VAX or on other machines in which the calling convention includes
1132 the number of words of arguments actually passed, the debugger (GDB at
1133 least) uses the parameter symbols to keep track of whether it needs to
1134 print nameless arguments in addition to the formal parameters which it
1135 has printed because each one has a stab. For example, in
1138 extern int fprintf (FILE *stream, char *format, @dots{});
1140 fprintf (stdout, "%d\n", x);
1143 there are stabs for @code{stream} and @code{format}. On most machines,
1144 the debugger can only print those two arguments (because it has no way
1145 of knowing that additional arguments were passed), but on the VAX or
1146 other machines with a calling convention which indicates the number of
1147 words of arguments, the debugger can print all three arguments. To do
1148 so, the parameter symbol (symbol descriptor @samp{p}) (not necessarily
1149 @samp{r} or symbol descriptor omitted symbols) needs to contain the
1150 actual type as passed (for example, @code{double} not @code{float} if it
1151 is passed as a double and converted to a float).
1153 @node Reference Parameters
1154 @subsection Passing Parameters by Reference
1156 If the parameter is passed by reference (e.g., Pascal @code{VAR}
1157 parameters), then the symbol descriptor is @samp{v} if it is in the
1158 argument list, or @samp{a} if it in a register. Other than the fact
1159 that these contain the address of the parameter rather than the
1160 parameter itself, they are identical to @samp{p} and @samp{R},
1161 respectively. I believe @samp{a} is an AIX invention; @samp{v} is
1162 supported by all stabs-using systems as far as I know.
1164 @node Conformant Arrays
1165 @subsection Passing Conformant Array Parameters
1167 @c Is this paragraph correct? It is based on piecing together patchy
1168 @c information and some guesswork
1169 Conformant arrays are a feature of Modula-2, and perhaps other
1170 languages, in which the size of an array parameter is not known to the
1171 called function until run-time. Such parameters have two stabs: a
1172 @samp{x} for the array itself, and a @samp{C}, which represents the size
1173 of the array. The value of the @samp{x} stab is the offset in the
1174 argument list where the address of the array is stored (it this right?
1175 it is a guess); the value of the @samp{C} stab is the offset in the
1176 argument list where the size of the array (in elements? in bytes?) is
1180 @chapter Defining Types
1182 The examples so far have described types as references to previously
1183 defined types, or defined in terms of subranges of or pointers to
1184 previously defined types. This chapter describes the other type
1185 descriptors that may follow the @samp{=} in a type definition.
1188 * Builtin Types:: Integers, floating point, void, etc.
1189 * Miscellaneous Types:: Pointers, sets, files, etc.
1190 * Cross-References:: Referring to a type not yet defined.
1191 * Subranges:: A type with a specific range.
1192 * Arrays:: An aggregate type of same-typed elements.
1193 * Strings:: Like an array but also has a length.
1194 * Enumerations:: Like an integer but the values have names.
1195 * Structures:: An aggregate type of different-typed elements.
1196 * Typedefs:: Giving a type a name.
1197 * Unions:: Different types sharing storage.
1202 @section Builtin Types
1204 Certain types are built in (@code{int}, @code{short}, @code{void},
1205 @code{float}, etc.); the debugger recognizes these types and knows how
1206 to handle them. Thus, don't be surprised if some of the following ways
1207 of specifying builtin types do not specify everything that a debugger
1208 would need to know about the type---in some cases they merely specify
1209 enough information to distinguish the type from other types.
1211 The traditional way to define builtin types is convolunted, so new ways
1212 have been invented to describe them. Sun's @code{acc} uses special
1213 builtin type descriptors (@samp{b} and @samp{R}), and IBM uses negative
1214 type numbers. GDB accepts all three ways, as of version 4.8; dbx just
1215 accepts the traditional builtin types and perhaps one of the other two
1216 formats. The following sections describe each of these formats.
1219 * Traditional Builtin Types:: Put on your seatbelts and prepare for kludgery
1220 * Builtin Type Descriptors:: Builtin types with special type descriptors
1221 * Negative Type Numbers:: Builtin types using negative type numbers
1224 @node Traditional Builtin Types
1225 @subsection Traditional Builtin Types
1227 This is the traditional, convoluted method for defining builtin types.
1228 There are several classes of such type definitions: integer, floating
1229 point, and @code{void}.
1232 * Traditional Integer Types::
1233 * Traditional Other Types::
1236 @node Traditional Integer Types
1237 @subsubsection Traditional Integer Types
1239 Often types are defined as subranges of themselves. If the bounding values
1240 fit within an @code{int}, then they are given normally. For example:
1243 .stabs "int:t1=r1;-2147483648;2147483647;",128,0,0,0 # @r{128 is N_LSYM}
1244 .stabs "char:t2=r2;0;127;",128,0,0,0
1247 Builtin types can also be described as subranges of @code{int}:
1250 .stabs "unsigned short:t6=r1;0;65535;",128,0,0,0
1253 If the lower bound of a subrange is 0 and the upper bound is -1,
1254 the type is an unsigned integral type whose bounds are too
1255 big to describe in an @code{int}. Traditionally this is only used for
1256 @code{unsigned int} and @code{unsigned long}:
1259 .stabs "unsigned int:t4=r1;0;-1;",128,0,0,0
1262 For larger types, GCC 2.4.5 puts out bounds in octal, with one or more
1263 leading zeroes. In this case a negative bound consists of a number
1264 which is a 1 bit (for the sign bit) followed by a 0 bit for each bit in
1265 the number (except the sign bit), and a positive bound is one which is a
1266 1 bit for each bit in the number (except possibly the sign bit). All
1267 known versions of dbx and GDB version 4 accept this (at least in the
1268 sense of not refusing to process the file), but GDB 3.5 refuses to read
1269 the whole file containing such symbols. So GCC 2.3.3 did not output the
1270 proper size for these types. As an example of octal bounds, the string
1271 fields of the stabs for 64 bit integer types look like:
1273 @c .stabs directives, etc., omitted to make it fit on the page.
1275 long int:t3=r1;001000000000000000000000;000777777777777777777777;
1276 long unsigned int:t5=r1;000000000000000000000000;001777777777777777777777;
1279 If the lower bound of a subrange is 0 and the upper bound is negative,
1280 the type is an unsigned integral type whose size in bytes is the
1281 absolute value of the upper bound. I believe this is a Convex
1282 convention for @code{unsigned long long}.
1284 If the lower bound of a subrange is negative and the upper bound is 0,
1285 the type is a signed integral type whose size in bytes is
1286 the absolute value of the lower bound. I believe this is a Convex
1287 convention for @code{long long}. To distinguish this from a legitimate
1288 subrange, the type should be a subrange of itself. I'm not sure whether
1289 this is the case for Convex.
1291 @node Traditional Other Types
1292 @subsubsection Traditional Other Types
1294 If the upper bound of a subrange is 0 and the lower bound is positive,
1295 the type is a floating point type, and the lower bound of the subrange
1296 indicates the number of bytes in the type:
1299 .stabs "float:t12=r1;4;0;",128,0,0,0
1300 .stabs "double:t13=r1;8;0;",128,0,0,0
1303 However, GCC writes @code{long double} the same way it writes
1304 @code{double}, so there is no way to distinguish.
1307 .stabs "long double:t14=r1;8;0;",128,0,0,0
1310 Complex types are defined the same way as floating-point types; there is
1311 no way to distinguish a single-precision complex from a double-precision
1312 floating-point type.
1314 The C @code{void} type is defined as itself:
1317 .stabs "void:t15=15",128,0,0,0
1320 I'm not sure how a boolean type is represented.
1322 @node Builtin Type Descriptors
1323 @subsection Defining Builtin Types Using Builtin Type Descriptors
1325 This is the method used by Sun's @code{acc} for defining builtin types.
1326 These are the type descriptors to define builtin types:
1329 @c FIXME: clean up description of width and offset, once we figure out
1331 @item b @var{signed} @var{char-flag} @var{width} ; @var{offset} ; @var{nbits} ;
1332 Define an integral type. @var{signed} is @samp{u} for unsigned or
1333 @samp{s} for signed. @var{char-flag} is @samp{c} which indicates this
1334 is a character type, or is omitted. I assume this is to distinguish an
1335 integral type from a character type of the same size, for example it
1336 might make sense to set it for the C type @code{wchar_t} so the debugger
1337 can print such variables differently (Solaris does not do this). Sun
1338 sets it on the C types @code{signed char} and @code{unsigned char} which
1339 arguably is wrong. @var{width} and @var{offset} appear to be for small
1340 objects stored in larger ones, for example a @code{short} in an
1341 @code{int} register. @var{width} is normally the number of bytes in the
1342 type. @var{offset} seems to always be zero. @var{nbits} is the number
1343 of bits in the type.
1345 Note that type descriptor @samp{b} used for builtin types conflicts with
1346 its use for Pascal space types (@pxref{Miscellaneous Types}); they can
1347 be distinguished because the character following the type descriptor
1348 will be a digit, @samp{(}, or @samp{-} for a Pascal space type, or
1349 @samp{u} or @samp{s} for a builtin type.
1352 Documented by AIX to define a wide character type, but their compiler
1353 actually uses negative type numbers (@pxref{Negative Type Numbers}).
1355 @item R @var{fp-type} ; @var{bytes} ;
1356 Define a floating point type. @var{fp-type} has one of the following values:
1360 IEEE 32-bit (single precision) floating point format.
1363 IEEE 64-bit (double precision) floating point format.
1365 @item 3 (NF_COMPLEX)
1366 @item 4 (NF_COMPLEX16)
1367 @item 5 (NF_COMPLEX32)
1368 @c "GDB source" really means @file{include/aout/stab_gnu.h}, but trying
1369 @c to put that here got an overfull hbox.
1370 These are for complex numbers. A comment in the GDB source describes
1371 them as Fortran @code{complex}, @code{double complex}, and
1372 @code{complex*16}, respectively, but what does that mean? (i.e., Single
1373 precision? Double precison?).
1375 @item 6 (NF_LDOUBLE)
1376 Long double. This should probably only be used for Sun format
1377 @code{long double}, and new codes should be used for other floating
1378 point formats (@code{NF_DOUBLE} can be used if a @code{long double} is
1379 really just an IEEE double, of course).
1382 @var{bytes} is the number of bytes occupied by the type. This allows a
1383 debugger to perform some operations with the type even if it doesn't
1384 understand @var{fp-type}.
1386 @item g @var{type-information} ; @var{nbits}
1387 Documented by AIX to define a floating type, but their compiler actually
1388 uses negative type numbers (@pxref{Negative Type Numbers}).
1390 @item c @var{type-information} ; @var{nbits}
1391 Documented by AIX to define a complex type, but their compiler actually
1392 uses negative type numbers (@pxref{Negative Type Numbers}).
1395 The C @code{void} type is defined as a signed integral type 0 bits long:
1397 .stabs "void:t19=bs0;0;0",128,0,0,0
1399 The Solaris compiler seems to omit the trailing semicolon in this case.
1400 Getting sloppy in this way is not a swift move because if a type is
1401 embedded in a more complex expression it is necessary to be able to tell
1404 I'm not sure how a boolean type is represented.
1406 @node Negative Type Numbers
1407 @subsection Negative Type Numbers
1409 This is the method used in XCOFF for defining builtin types.
1410 Since the debugger knows about the builtin types anyway, the idea of
1411 negative type numbers is simply to give a special type number which
1412 indicates the builtin type. There is no stab defining these types.
1414 There are several subtle issues with negative type numbers.
1416 One is the size of the type. A builtin type (for example the C types
1417 @code{int} or @code{long}) might have different sizes depending on
1418 compiler options, the target architecture, the ABI, etc. This issue
1419 doesn't come up for IBM tools since (so far) they just target the
1420 RS/6000; the sizes indicated below for each size are what the IBM
1421 RS/6000 tools use. To deal with differing sizes, either define separate
1422 negative type numbers for each size (which works but requires changing
1423 the debugger, and, unless you get both AIX dbx and GDB to accept the
1424 change, introduces an incompatibility), or use a type attribute
1425 (@pxref{String Field}) to define a new type with the appropriate size
1426 (which merely requires a debugger which understands type attributes,
1427 like AIX dbx). For example,
1430 .stabs "boolean:t10=@@s8;-16",128,0,0,0
1433 defines an 8-bit boolean type, and
1436 .stabs "boolean:t10=@@s64;-16",128,0,0,0
1439 defines a 64-bit boolean type.
1441 A similar issue is the format of the type. This comes up most often for
1442 floating-point types, which could have various formats (particularly
1443 extended doubles, which vary quite a bit even among IEEE systems).
1444 Again, it is best to define a new negative type number for each
1445 different format; changing the format based on the target system has
1446 various problems. One such problem is that the Alpha has both VAX and
1447 IEEE floating types. One can easily imagine one library using the VAX
1448 types and another library in the same executable using the IEEE types.
1449 Another example is that the interpretation of whether a boolean is true
1450 or false can be based on the least significant bit, most significant
1451 bit, whether it is zero, etc., and different compilers (or different
1452 options to the same compiler) might provide different kinds of boolean.
1454 The last major issue is the names of the types. The name of a given
1455 type depends @emph{only} on the negative type number given; these do not
1456 vary depending on the language, the target system, or anything else.
1457 One can always define separate type numbers---in the following list you
1458 will see for example separate @code{int} and @code{integer*4} types
1459 which are identical except for the name. But compatibility can be
1460 maintained by not inventing new negative type numbers and instead just
1461 defining a new type with a new name. For example:
1464 .stabs "CARDINAL:t10=-8",128,0,0,0
1467 Here is the list of negative type numbers. The phrase @dfn{integral
1468 type} is used to mean twos-complement (I strongly suspect that all
1469 machines which use stabs use twos-complement; most machines use
1470 twos-complement these days).
1474 @code{int}, 32 bit signed integral type.
1477 @code{char}, 8 bit type holding a character. Both GDB and dbx on AIX
1478 treat this as signed. GCC uses this type whether @code{char} is signed
1479 or not, which seems like a bad idea. The AIX compiler (@code{xlc}) seems to
1480 avoid this type; it uses -5 instead for @code{char}.
1483 @code{short}, 16 bit signed integral type.
1486 @code{long}, 32 bit signed integral type.
1489 @code{unsigned char}, 8 bit unsigned integral type.
1492 @code{signed char}, 8 bit signed integral type.
1495 @code{unsigned short}, 16 bit unsigned integral type.
1498 @code{unsigned int}, 32 bit unsigned integral type.
1501 @code{unsigned}, 32 bit unsigned integral type.
1504 @code{unsigned long}, 32 bit unsigned integral type.
1507 @code{void}, type indicating the lack of a value.
1510 @code{float}, IEEE single precision.
1513 @code{double}, IEEE double precision.
1516 @code{long double}, IEEE double precision. The compiler claims the size
1517 will increase in a future release, and for binary compatibility you have
1518 to avoid using @code{long double}. I hope when they increase it they
1519 use a new negative type number.
1522 @code{integer}. 32 bit signed integral type.
1525 @code{boolean}. 32 bit type. How is the truth value encoded? Is it
1526 the least significant bit or is it a question of whether the whole value
1527 is zero or non-zero?
1530 @code{short real}. IEEE single precision.
1533 @code{real}. IEEE double precision.
1536 @code{stringptr}. @xref{Strings}.
1539 @code{character}, 8 bit unsigned character type.
1542 @code{logical*1}, 8 bit type. This Fortran type has a split
1543 personality in that it is used for boolean variables, but can also be
1544 used for unsigned integers. 0 is false, 1 is true, and other values are
1548 @code{logical*2}, 16 bit type. This Fortran type has a split
1549 personality in that it is used for boolean variables, but can also be
1550 used for unsigned integers. 0 is false, 1 is true, and other values are
1554 @code{logical*4}, 32 bit type. This Fortran type has a split
1555 personality in that it is used for boolean variables, but can also be
1556 used for unsigned integers. 0 is false, 1 is true, and other values are
1560 @code{logical}, 32 bit type. This Fortran type has a split
1561 personality in that it is used for boolean variables, but can also be
1562 used for unsigned integers. 0 is false, 1 is true, and other values are
1566 @code{complex}. A complex type consisting of two IEEE single-precision
1567 floating point values.
1570 @code{complex}. A complex type consisting of two IEEE double-precision
1571 floating point values.
1574 @code{integer*1}, 8 bit signed integral type.
1577 @code{integer*2}, 16 bit signed integral type.
1580 @code{integer*4}, 32 bit signed integral type.
1583 @code{wchar}. Wide character, 16 bits wide, unsigned (what format?
1587 @node Miscellaneous Types
1588 @section Miscellaneous Types
1591 @item b @var{type-information} ; @var{bytes}
1592 Pascal space type. This is documented by IBM; what does it mean?
1594 This use of the @samp{b} type descriptor can be distinguished
1595 from its use for builtin integral types (@pxref{Builtin Type
1596 Descriptors}) because the character following the type descriptor is
1597 always a digit, @samp{(}, or @samp{-}.
1599 @item B @var{type-information}
1600 A volatile-qualified version of @var{type-information}. This is
1601 a Sun extension. References and stores to a variable with a
1602 volatile-qualified type must not be optimized or cached; they
1603 must occur as the user specifies them.
1605 @item d @var{type-information}
1606 File of type @var{type-information}. As far as I know this is only used
1609 @item k @var{type-information}
1610 A const-qualified version of @var{type-information}. This is a Sun
1611 extension. A variable with a const-qualified type cannot be modified.
1613 @item M @var{type-information} ; @var{length}
1614 Multiple instance type. The type seems to composed of @var{length}
1615 repetitions of @var{type-information}, for example @code{character*3} is
1616 represented by @samp{M-2;3}, where @samp{-2} is a reference to a
1617 character type (@pxref{Negative Type Numbers}). I'm not sure how this
1618 differs from an array. This appears to be a Fortran feature.
1619 @var{length} is a bound, like those in range types; see @ref{Subranges}.
1621 @item S @var{type-information}
1622 Pascal set type. @var{type-information} must be a small type such as an
1623 enumeration or a subrange, and the type is a bitmask whose length is
1624 specified by the number of elements in @var{type-information}.
1626 @item * @var{type-information}
1627 Pointer to @var{type-information}.
1630 @node Cross-References
1631 @section Cross-References to Other Types
1633 A type can be used before it is defined; one common way to deal with
1634 that situation is just to use a type reference to a type which has not
1637 Another way is with the @samp{x} type descriptor, which is followed by
1638 @samp{s} for a structure tag, @samp{u} for a union tag, or @samp{e} for
1639 a enumerator tag, followed by the name of the tag, followed by @samp{:}.
1640 For example, the following C declarations:
1651 .stabs "bar:G16=*17=xsfoo:",32,0,0,0
1654 Not all debuggers support the @samp{x} type descriptor, so on some
1655 machines GCC does not use it. I believe that for the above example it
1656 would just emit a reference to type 17 and never define it, but I
1657 haven't verified that.
1659 Modula-2 imported types, at least on AIX, use the @samp{i} type
1660 descriptor, which is followed by the name of the module from which the
1661 type is imported, followed by @samp{:}, followed by the name of the
1662 type. There is then optionally a comma followed by type information for
1663 the type. This differs from merely naming the type (@pxref{Typedefs}) in
1664 that it identifies the module; I don't understand whether the name of
1665 the type given here is always just the same as the name we are giving
1666 it, or whether this type descriptor is used with a nameless stab
1667 (@pxref{String Field}), or what. The symbol ends with @samp{;}.
1670 @section Subrange Types
1672 The @samp{r} type descriptor defines a type as a subrange of another
1673 type. It is followed by type information for the type of which it is a
1674 subrange, a semicolon, an integral lower bound, a semicolon, an
1675 integral upper bound, and a semicolon. The AIX documentation does not
1676 specify the trailing semicolon, in an effort to specify array indexes
1677 more cleanly, but a subrange which is not an array index has always
1678 included a trailing semicolon (@pxref{Arrays}).
1680 Instead of an integer, either bound can be one of the following:
1683 @item A @var{offset}
1684 The bound is passed by reference on the stack at offset @var{offset}
1685 from the argument list. @xref{Parameters}, for more information on such
1688 @item T @var{offset}
1689 The bound is passed by value on the stack at offset @var{offset} from
1692 @item a @var{register-number}
1693 The bound is pased by reference in register number
1694 @var{register-number}.
1696 @item t @var{register-number}
1697 The bound is passed by value in register number @var{register-number}.
1703 Subranges are also used for builtin types; see @ref{Traditional Builtin Types}.
1706 @section Array Types
1708 Arrays use the @samp{a} type descriptor. Following the type descriptor
1709 is the type of the index and the type of the array elements. If the
1710 index type is a range type, it ends in a semicolon; otherwise
1711 (for example, if it is a type reference), there does not
1712 appear to be any way to tell where the types are separated. In an
1713 effort to clean up this mess, IBM documents the two types as being
1714 separated by a semicolon, and a range type as not ending in a semicolon
1715 (but this is not right for range types which are not array indexes,
1716 @pxref{Subranges}). I think probably the best solution is to specify
1717 that a semicolon ends a range type, and that the index type and element
1718 type of an array are separated by a semicolon, but that if the index
1719 type is a range type, the extra semicolon can be omitted. GDB (at least
1720 through version 4.9) doesn't support any kind of index type other than a
1721 range anyway; I'm not sure about dbx.
1723 It is well established, and widely used, that the type of the index,
1724 unlike most types found in the stabs, is merely a type definition, not
1725 type information (@pxref{String Field}) (that is, it need not start with
1726 @samp{@var{type-number}=} if it is defining a new type). According to a
1727 comment in GDB, this is also true of the type of the array elements; it
1728 gives @samp{ar1;1;10;ar1;1;10;4} as a legitimate way to express a two
1729 dimensional array. According to AIX documentation, the element type
1730 must be type information. GDB accepts either.
1732 The type of the index is often a range type, expressed as the type
1733 descriptor @samp{r} and some parameters. It defines the size of the
1734 array. In the example below, the range @samp{r1;0;2;} defines an index
1735 type which is a subrange of type 1 (integer), with a lower bound of 0
1736 and an upper bound of 2. This defines the valid range of subscripts of
1737 a three-element C array.
1739 For example, the definition:
1742 char char_vec[3] = @{'a','b','c'@};
1746 produces the output:
1749 .stabs "char_vec:G19=ar1;0;2;2",32,0,0,0
1758 If an array is @dfn{packed}, the elements are spaced more
1759 closely than normal, saving memory at the expense of speed. For
1760 example, an array of 3-byte objects might, if unpacked, have each
1761 element aligned on a 4-byte boundary, but if packed, have no padding.
1762 One way to specify that something is packed is with type attributes
1763 (@pxref{String Field}). In the case of arrays, another is to use the
1764 @samp{P} type descriptor instead of @samp{a}. Other than specifying a
1765 packed array, @samp{P} is identical to @samp{a}.
1767 @c FIXME-what is it? A pointer?
1768 An open array is represented by the @samp{A} type descriptor followed by
1769 type information specifying the type of the array elements.
1771 @c FIXME: what is the format of this type? A pointer to a vector of pointers?
1772 An N-dimensional dynamic array is represented by
1775 D @var{dimensions} ; @var{type-information}
1778 @c Does dimensions really have this meaning? The AIX documentation
1780 @var{dimensions} is the number of dimensions; @var{type-information}
1781 specifies the type of the array elements.
1783 @c FIXME: what is the format of this type? A pointer to some offsets in
1785 A subarray of an N-dimensional array is represented by
1788 E @var{dimensions} ; @var{type-information}
1791 @c Does dimensions really have this meaning? The AIX documentation
1793 @var{dimensions} is the number of dimensions; @var{type-information}
1794 specifies the type of the array elements.
1799 Some languages, like C or the original Pascal, do not have string types,
1800 they just have related things like arrays of characters. But most
1801 Pascals and various other languages have string types, which are
1802 indicated as follows:
1805 @item n @var{type-information} ; @var{bytes}
1806 @var{bytes} is the maximum length. I'm not sure what
1807 @var{type-information} is; I suspect that it means that this is a string
1808 of @var{type-information} (thus allowing a string of integers, a string
1809 of wide characters, etc., as well as a string of characters). Not sure
1810 what the format of this type is. This is an AIX feature.
1812 @item z @var{type-information} ; @var{bytes}
1813 Just like @samp{n} except that this is a gstring, not an ordinary
1814 string. I don't know the difference.
1817 Pascal Stringptr. What is this? This is an AIX feature.
1821 @section Enumerations
1823 Enumerations are defined with the @samp{e} type descriptor.
1825 @c FIXME: Where does this information properly go? Perhaps it is
1826 @c redundant with something we already explain.
1827 The source line below declares an enumeration type at file scope.
1828 The type definition is located after the @code{N_RBRAC} that marks the end of
1829 the previous procedure's block scope, and before the @code{N_FUN} that marks
1830 the beginning of the next procedure's block scope. Therefore it does not
1831 describe a block local symbol, but a file local one.
1836 enum e_places @{first,second=3,last@};
1840 generates the following stab:
1843 .stabs "e_places:T22=efirst:0,second:3,last:4,;",128,0,0,0
1846 The symbol descriptor (@samp{T}) says that the stab describes a
1847 structure, enumeration, or union tag. The type descriptor @samp{e},
1848 following the @samp{22=} of the type definition narrows it down to an
1849 enumeration type. Following the @samp{e} is a list of the elements of
1850 the enumeration. The format is @samp{@var{name}:@var{value},}. The
1851 list of elements ends with @samp{;}.
1853 There is no standard way to specify the size of an enumeration type; it
1854 is determined by the architecture (normally all enumerations types are
1855 32 bits). There should be a way to specify an enumeration type of
1856 another size; type attributes would be one way to do this. @xref{Stabs
1862 The encoding of structures in stabs can be shown with an example.
1864 The following source code declares a structure tag and defines an
1865 instance of the structure in global scope. Then a @code{typedef} equates the
1866 structure tag with a new type. Seperate stabs are generated for the
1867 structure tag, the structure @code{typedef}, and the structure instance. The
1868 stabs for the tag and the @code{typedef} are emited when the definitions are
1869 encountered. Since the structure elements are not initialized, the
1870 stab and code for the structure variable itself is located at the end
1871 of the program in the bss section.
1878 struct s_tag* s_next;
1881 typedef struct s_tag s_typedef;
1884 The structure tag has an @code{N_LSYM} stab type because, like the
1885 enumeration, the symbol has file scope. Like the enumeration, the
1886 symbol descriptor is @samp{T}, for enumeration, structure, or tag type.
1887 The type descriptor @samp{s} following the @samp{16=} of the type
1888 definition narrows the symbol type to structure.
1890 Following the @samp{s} type descriptor is the number of bytes the
1891 structure occupies, followed by a description of each structure element.
1892 The structure element descriptions are of the form @var{name:type, bit
1893 offset from the start of the struct, number of bits in the element}.
1895 @c FIXME: phony line break. Can probably be fixed by using an example
1896 @c with fewer fields.
1899 .stabs "s_tag:T16=s20s_int:1,0,32;s_float:12,32,32;
1900 s_char_vec:17=ar1;0;7;2,64,64;s_next:18=*16,128,32;;",128,0,0,0
1903 In this example, the first two structure elements are previously defined
1904 types. For these, the type following the @samp{@var{name}:} part of the
1905 element description is a simple type reference. The other two structure
1906 elements are new types. In this case there is a type definition
1907 embedded after the @samp{@var{name}:}. The type definition for the
1908 array element looks just like a type definition for a standalone array.
1909 The @code{s_next} field is a pointer to the same kind of structure that
1910 the field is an element of. So the definition of structure type 16
1911 contains a type definition for an element which is a pointer to type 16.
1914 @section Giving a Type a Name
1916 To give a type a name, use the @samp{t} symbol descriptor. The type
1917 is specified by the type information (@pxref{String Field}) for the stab.
1921 .stabs "s_typedef:t16",128,0,0,0 # @r{128 is N_LSYM}
1924 specifies that @code{s_typedef} refers to type number 16. Such stabs
1925 have symbol type @code{N_LSYM} (or @code{C_DECL} for XCOFF).
1927 If you are specifying the tag name for a structure, union, or
1928 enumeration, use the @samp{T} symbol descriptor instead. I believe C is
1929 the only language with this feature.
1931 If the type is an opaque type (I believe this is a Modula-2 feature),
1932 AIX provides a type descriptor to specify it. The type descriptor is
1933 @samp{o} and is followed by a name. I don't know what the name
1934 means---is it always the same as the name of the type, or is this type
1935 descriptor used with a nameless stab (@pxref{String Field})? There
1936 optionally follows a comma followed by type information which defines
1937 the type of this type. If omitted, a semicolon is used in place of the
1938 comma and the type information, and the type is much like a generic
1939 pointer type---it has a known size but little else about it is
1953 This code generates a stab for a union tag and a stab for a union
1954 variable. Both use the @code{N_LSYM} stab type. If a union variable is
1955 scoped locally to the procedure in which it is defined, its stab is
1956 located immediately preceding the @code{N_LBRAC} for the procedure's block
1959 The stab for the union tag, however, is located preceding the code for
1960 the procedure in which it is defined. The stab type is @code{N_LSYM}. This
1961 would seem to imply that the union type is file scope, like the struct
1962 type @code{s_tag}. This is not true. The contents and position of the stab
1963 for @code{u_type} do not convey any infomation about its procedure local
1966 @c FIXME: phony line break. Can probably be fixed by using an example
1967 @c with fewer fields.
1970 .stabs "u_tag:T23=u4u_int:1,0,32;u_float:12,0,32;u_char:21,0,32;;",
1974 The symbol descriptor @samp{T}, following the @samp{name:} means that
1975 the stab describes an enumeration, structure, or union tag. The type
1976 descriptor @samp{u}, following the @samp{23=} of the type definition,
1977 narrows it down to a union type definition. Following the @samp{u} is
1978 the number of bytes in the union. After that is a list of union element
1979 descriptions. Their format is @var{name:type, bit offset into the
1980 union, number of bytes for the element;}.
1982 The stab for the union variable is:
1985 .stabs "an_u:23",128,0,0,-20 # @r{128 is N_LSYM}
1988 @samp{-20} specifies where the variable is stored (@pxref{Stack
1991 @node Function Types
1992 @section Function Types
1994 Various types can be defined for function variables. These types are
1995 not used in defining functions (@pxref{Procedures}); they are used for
1996 things like pointers to functions.
1998 The simple, traditional, type is type descriptor @samp{f} is followed by
1999 type information for the return type of the function, followed by a
2002 This does not deal with functions for which the number and types of the
2003 parameters are part of the type, as in Modula-2 or ANSI C. AIX provides
2004 extensions to specify these, using the @samp{f}, @samp{F}, @samp{p}, and
2005 @samp{R} type descriptors.
2007 First comes the type descriptor. If it is @samp{f} or @samp{F}, this
2008 type involves a function rather than a procedure, and the type
2009 information for the return type of the function follows, followed by a
2010 comma. Then comes the number of parameters to the function and a
2011 semicolon. Then, for each parameter, there is the name of the parameter
2012 followed by a colon (this is only present for type descriptors @samp{R}
2013 and @samp{F} which represent Pascal function or procedure parameters),
2014 type information for the parameter, a comma, 0 if passed by reference or
2015 1 if passed by value, and a semicolon. The type definition ends with a
2018 For example, this variable definition:
2025 generates the following code:
2028 .stabs "g_pf:G24=*25=f1",32,0,0,0
2029 .common _g_pf,4,"bss"
2032 The variable defines a new type, 24, which is a pointer to another new
2033 type, 25, which is a function returning @code{int}.
2036 @chapter Symbol Information in Symbol Tables
2038 This chapter describes the format of symbol table entries
2039 and how stab assembler directives map to them. It also describes the
2040 transformations that the assembler and linker make on data from stabs.
2043 * Symbol Table Format::
2044 * Transformations On Symbol Tables::
2047 @node Symbol Table Format
2048 @section Symbol Table Format
2050 Each time the assembler encounters a stab directive, it puts
2051 each field of the stab into a corresponding field in a symbol table
2052 entry of its output file. If the stab contains a string field, the
2053 symbol table entry for that stab points to a string table entry
2054 containing the string data from the stab. Assembler labels become
2055 relocatable addresses. Symbol table entries in a.out have the format:
2057 @c FIXME: should refer to external, not internal.
2059 struct internal_nlist @{
2060 unsigned long n_strx; /* index into string table of name */
2061 unsigned char n_type; /* type of symbol */
2062 unsigned char n_other; /* misc info (usually empty) */
2063 unsigned short n_desc; /* description field */
2064 bfd_vma n_value; /* value of symbol */
2068 If the stab has a string, the @code{n_strx} field holds the offset in
2069 bytes of the string within the string table. The string is terminated
2070 by a NUL character. If the stab lacks a string (for example, it was
2071 produced by a @code{.stabn} or @code{.stabd} directive), the
2072 @code{n_strx} field is zero.
2074 Symbol table entries with @code{n_type} field values greater than 0x1f
2075 originated as stabs generated by the compiler (with one random
2076 exception). The other entries were placed in the symbol table of the
2077 executable by the assembler or the linker.
2079 @node Transformations On Symbol Tables
2080 @section Transformations on Symbol Tables
2082 The linker concatenates object files and does fixups of externally
2085 You can see the transformations made on stab data by the assembler and
2086 linker by examining the symbol table after each pass of the build. To
2087 do this, use @samp{nm -ap}, which dumps the symbol table, including
2088 debugging information, unsorted. For stab entries the columns are:
2089 @var{value}, @var{other}, @var{desc}, @var{type}, @var{string}. For
2090 assembler and linker symbols, the columns are: @var{value}, @var{type},
2093 The low 5 bits of the stab type tell the linker how to relocate the
2094 value of the stab. Thus for stab types like @code{N_RSYM} and
2095 @code{N_LSYM}, where the value is an offset or a register number, the
2096 low 5 bits are @code{N_ABS}, which tells the linker not to relocate the
2099 Where the value of a stab contains an assembly language label,
2100 it is transformed by each build step. The assembler turns it into a
2101 relocatable address and the linker turns it into an absolute address.
2104 * Transformations On Static Variables::
2105 * Transformations On Global Variables::
2106 * ELF Transformations:: In ELF, things are a bit different.
2109 @node Transformations On Static Variables
2110 @subsection Transformations on Static Variables
2112 This source line defines a static variable at file scope:
2115 static int s_g_repeat
2119 The following stab describes the symbol:
2122 .stabs "s_g_repeat:S1",38,0,0,_s_g_repeat
2126 The assembler transforms the stab into this symbol table entry in the
2127 @file{.o} file. The location is expressed as a data segment offset.
2130 00000084 - 00 0000 STSYM s_g_repeat:S1
2134 In the symbol table entry from the executable, the linker has made the
2135 relocatable address absolute.
2138 0000e00c - 00 0000 STSYM s_g_repeat:S1
2141 @node Transformations On Global Variables
2142 @subsection Transformations on Global Variables
2144 Stabs for global variables do not contain location information. In
2145 this case, the debugger finds location information in the assembler or
2146 linker symbol table entry describing the variable. The source line:
2156 .stabs "g_foo:G2",32,0,0,0
2159 The variable is represented by two symbol table entries in the object
2160 file (see below). The first one originated as a stab. The second one
2161 is an external symbol. The upper case @samp{D} signifies that the
2162 @code{n_type} field of the symbol table contains 7, @code{N_DATA} with
2163 local linkage. The stab's value is zero since the value is not used for
2164 @code{N_GSYM} stabs. The value of the linker symbol is the relocatable
2165 address corresponding to the variable.
2168 00000000 - 00 0000 GSYM g_foo:G2
2173 These entries as transformed by the linker. The linker symbol table
2174 entry now holds an absolute address:
2177 00000000 - 00 0000 GSYM g_foo:G2
2182 @node ELF Transformations
2183 @subsection Transformations of Stabs in ELF Files
2185 For ELF files, use @code{objdump --stabs} instead of @code{nm} to show
2186 the stabs in an object or executable file. @code{objdump} is a GNU
2187 utility; Sun does not provide any equivalent.
2189 The following example is for a stab whose value is an address is
2190 relative to the compilation unit (@pxref{Stabs In ELF}). For example,
2197 appears within a function, then the assembly language output from the
2203 .stabs "ld:V(0,3)",0x26,0,4,.L18-Ddata.data # @r{0x26 is N_STSYM}
2210 Because the value is formed by subtracting one symbol from another, the
2211 value is absolute, not relocatable, and so the object file contains
2214 Symnum n_type n_othr n_desc n_value n_strx String
2215 31 STSYM 0 4 00000004 680 ld:V(0,3)
2218 without any relocations, and the executable file also contains
2221 Symnum n_type n_othr n_desc n_value n_strx String
2222 31 STSYM 0 4 00000004 680 ld:V(0,3)
2226 @chapter GNU C++ Stabs
2229 * Basic Cplusplus Types::
2232 * Methods:: Method definition
2234 * Method Modifiers::
2237 * Virtual Base Classes::
2241 Type descriptors added for C++ descriptions:
2245 method type (@code{##} if minimal debug)
2248 Member (class and variable) type. It is followed by type information
2249 for the offset basetype, a comma, and type information for the type of
2250 the field being pointed to. (FIXME: this is acknowledged to be
2251 gibberish. Can anyone say what really goes here?).
2253 Note that there is a conflict between this and type attributes
2254 (@pxref{String Field}); both use type descriptor @samp{@@}.
2255 Fortunately, the @samp{@@} type descriptor used in this C++ sense always
2256 will be followed by a digit, @samp{(}, or @samp{-}, and type attributes
2257 never start with those things.
2260 @node Basic Cplusplus Types
2261 @section Basic Types For C++
2263 << the examples that follow are based on a01.C >>
2266 C++ adds two more builtin types to the set defined for C. These are
2267 the unknown type and the vtable record type. The unknown type, type
2268 16, is defined in terms of itself like the void type.
2270 The vtable record type, type 17, is defined as a structure type and
2271 then as a structure tag. The structure has four fields: delta, index,
2272 pfn, and delta2. pfn is the function pointer.
2274 << In boilerplate $vtbl_ptr_type, what are the fields delta,
2275 index, and delta2 used for? >>
2277 This basic type is present in all C++ programs even if there are no
2278 virtual methods defined.
2281 .stabs "struct_name:sym_desc(type)type_def(17)=type_desc(struct)struct_bytes(8)
2282 elem_name(delta):type_ref(short int),bit_offset(0),field_bits(16);
2283 elem_name(index):type_ref(short int),bit_offset(16),field_bits(16);
2284 elem_name(pfn):type_def(18)=type_desc(ptr to)type_ref(void),
2285 bit_offset(32),field_bits(32);
2286 elem_name(delta2):type_def(short int);bit_offset(32),field_bits(16);;"
2291 .stabs "$vtbl_ptr_type:t17=s8
2292 delta:6,0,16;index:6,16,16;pfn:18=*15,32,32;delta2:6,32,16;;"
2297 .stabs "name:sym_dec(struct tag)type_ref($vtbl_ptr_type)",N_LSYM,NIL,NIL,NIL
2301 .stabs "$vtbl_ptr_type:T17",128,0,0,0
2304 @node Simple Classes
2305 @section Simple Class Definition
2307 The stabs describing C++ language features are an extension of the
2308 stabs describing C. Stabs representing C++ class types elaborate
2309 extensively on the stab format used to describe structure types in C.
2310 Stabs representing class type variables look just like stabs
2311 representing C language variables.
2313 Consider the following very simple class definition.
2319 int Ameth(int in, char other);
2323 The class @code{baseA} is represented by two stabs. The first stab describes
2324 the class as a structure type. The second stab describes a structure
2325 tag of the class type. Both stabs are of stab type @code{N_LSYM}. Since the
2326 stab is not located between an @code{N_FUN} and an @code{N_LBRAC} stab this indicates
2327 that the class is defined at file scope. If it were, then the @code{N_LSYM}
2328 would signify a local variable.
2330 A stab describing a C++ class type is similar in format to a stab
2331 describing a C struct, with each class member shown as a field in the
2332 structure. The part of the struct format describing fields is
2333 expanded to include extra information relevent to C++ class members.
2334 In addition, if the class has multiple base classes or virtual
2335 functions the struct format outside of the field parts is also
2338 In this simple example the field part of the C++ class stab
2339 representing member data looks just like the field part of a C struct
2340 stab. The section on protections describes how its format is
2341 sometimes extended for member data.
2343 The field part of a C++ class stab representing a member function
2344 differs substantially from the field part of a C struct stab. It
2345 still begins with @samp{name:} but then goes on to define a new type number
2346 for the member function, describe its return type, its argument types,
2347 its protection level, any qualifiers applied to the method definition,
2348 and whether the method is virtual or not. If the method is virtual
2349 then the method description goes on to give the vtable index of the
2350 method, and the type number of the first base class defining the
2353 When the field name is a method name it is followed by two colons rather
2354 than one. This is followed by a new type definition for the method.
2355 This is a number followed by an equal sign and the type descriptor
2356 @samp{#}, indicating a method type, and a second @samp{#}, indicating
2357 that this is the @dfn{minimal} type of method definition used by GCC2,
2358 not larger method definitions used by earlier versions of GCC. This is
2359 followed by a type reference showing the return type of the method and a
2362 The format of an overloaded operator method name differs from that of
2363 other methods. It is @samp{op$::@var{operator-name}.} where
2364 @var{operator-name} is the operator name such as @samp{+} or @samp{+=}.
2365 The name ends with a period, and any characters except the period can
2366 occur in the @var{operator-name} string.
2368 The next part of the method description represents the arguments to the
2369 method, preceeded by a colon and ending with a semi-colon. The types of
2370 the arguments are expressed in the same way argument types are expressed
2371 in C++ name mangling. In this example an @code{int} and a @code{char}
2374 This is followed by a number, a letter, and an asterisk or period,
2375 followed by another semicolon. The number indicates the protections
2376 that apply to the member function. Here the 2 means public. The
2377 letter encodes any qualifier applied to the method definition. In
2378 this case, @samp{A} means that it is a normal function definition. The dot
2379 shows that the method is not virtual. The sections that follow
2380 elaborate further on these fields and describe the additional
2381 information present for virtual methods.
2385 .stabs "class_name:sym_desc(type)type_def(20)=type_desc(struct)struct_bytes(4)
2386 field_name(Adat):type(int),bit_offset(0),field_bits(32);
2388 method_name(Ameth)::type_def(21)=type_desc(method)return_type(int);
2389 :arg_types(int char);
2390 protection(public)qualifier(normal)virtual(no);;"
2395 .stabs "baseA:t20=s4Adat:1,0,32;Ameth::21=##1;:ic;2A.;;",128,0,0,0
2397 .stabs "class_name:sym_desc(struct tag)",N_LSYM,NIL,NIL,NIL
2399 .stabs "baseA:T20",128,0,0,0
2402 @node Class Instance
2403 @section Class Instance
2405 As shown above, describing even a simple C++ class definition is
2406 accomplished by massively extending the stab format used in C to
2407 describe structure types. However, once the class is defined, C stabs
2408 with no modifications can be used to describe class instances. The
2418 yields the following stab describing the class instance. It looks no
2419 different from a standard C stab describing a local variable.
2422 .stabs "name:type_ref(baseA)", N_LSYM, NIL, NIL, frame_ptr_offset
2426 .stabs "AbaseA:20",128,0,0,-20
2430 @section Method Definition
2432 The class definition shown above declares Ameth. The C++ source below
2437 baseA::Ameth(int in, char other)
2444 This method definition yields three stabs following the code of the
2445 method. One stab describes the method itself and following two describe
2446 its parameters. Although there is only one formal argument all methods
2447 have an implicit argument which is the @code{this} pointer. The @code{this}
2448 pointer is a pointer to the object on which the method was called. Note
2449 that the method name is mangled to encode the class name and argument
2450 types. Name mangling is described in the @sc{arm} (@cite{The Annotated
2451 C++ Reference Manual}, by Ellis and Stroustrup, @sc{isbn}
2452 0-201-51459-1); @file{gpcompare.texi} in Cygnus GCC distributions
2453 describes the differences between GNU mangling and @sc{arm}
2455 @c FIXME: Use @xref, especially if this is generally installed in the
2457 @c FIXME: This information should be in a net release, either of GCC or
2458 @c GDB. But gpcompare.texi doesn't seem to be in the FSF GCC.
2461 .stabs "name:symbol_desriptor(global function)return_type(int)",
2462 N_FUN, NIL, NIL, code_addr_of_method_start
2464 .stabs "Ameth__5baseAic:F1",36,0,0,_Ameth__5baseAic
2467 Here is the stab for the @code{this} pointer implicit argument. The
2468 name of the @code{this} pointer is always @code{this}. Type 19, the
2469 @code{this} pointer is defined as a pointer to type 20, @code{baseA},
2470 but a stab defining @code{baseA} has not yet been emited. Since the
2471 compiler knows it will be emited shortly, here it just outputs a cross
2472 reference to the undefined symbol, by prefixing the symbol name with
2476 .stabs "name:sym_desc(register param)type_def(19)=
2477 type_desc(ptr to)type_ref(baseA)=
2478 type_desc(cross-reference to)baseA:",N_RSYM,NIL,NIL,register_number
2480 .stabs "this:P19=*20=xsbaseA:",64,0,0,8
2483 The stab for the explicit integer argument looks just like a parameter
2484 to a C function. The last field of the stab is the offset from the
2485 argument pointer, which in most systems is the same as the frame
2489 .stabs "name:sym_desc(value parameter)type_ref(int)",
2490 N_PSYM,NIL,NIL,offset_from_arg_ptr
2492 .stabs "in:p1",160,0,0,72
2495 << The examples that follow are based on A1.C >>
2498 @section Protections
2501 In the simple class definition shown above all member data and
2502 functions were publicly accessable. The example that follows
2503 contrasts public, protected and privately accessable fields and shows
2504 how these protections are encoded in C++ stabs.
2506 @c FIXME: What does "part of the string" mean?
2507 Protections for class member data are signified by two characters
2508 embedded in the stab defining the class type. These characters are
2509 located after the name: part of the string. @samp{/0} means private,
2510 @samp{/1} means protected, and @samp{/2} means public. If these
2511 characters are omited this means that the member is public. The
2512 following C++ source:
2526 generates the following stab to describe the class type all_data.
2529 .stabs "class_name:sym_desc(type)type_def(19)=type_desc(struct)struct_bytes
2530 data_name:/protection(private)type_ref(int),bit_offset,num_bits;
2531 data_name:/protection(protected)type_ref(char),bit_offset,num_bits;
2532 data_name:(/num omited, private)type_ref(float),bit_offset,num_bits;;"
2537 .stabs "all_data:t19=s12
2538 priv_dat:/01,0,32;prot_dat:/12,32,8;pub_dat:12,64,32;;",128,0,0,0
2541 Protections for member functions are signified by one digit embeded in
2542 the field part of the stab describing the method. The digit is 0 if
2543 private, 1 if protected and 2 if public. Consider the C++ class
2547 class all_methods @{
2549 int priv_meth(int in)@{return in;@};
2551 char protMeth(char in)@{return in;@};
2553 float pubMeth(float in)@{return in;@};
2557 It generates the following stab. The digit in question is to the left
2558 of an @samp{A} in each case. Notice also that in this case two symbol
2559 descriptors apply to the class name struct tag and struct type.
2562 .stabs "class_name:sym_desc(struct tag&type)type_def(21)=
2563 sym_desc(struct)struct_bytes(1)
2564 meth_name::type_def(22)=sym_desc(method)returning(int);
2565 :args(int);protection(private)modifier(normal)virtual(no);
2566 meth_name::type_def(23)=sym_desc(method)returning(char);
2567 :args(char);protection(protected)modifier(normal)virual(no);
2568 meth_name::type_def(24)=sym_desc(method)returning(float);
2569 :args(float);protection(public)modifier(normal)virtual(no);;",
2574 .stabs "all_methods:Tt21=s1priv_meth::22=##1;:i;0A.;protMeth::23=##2;:c;1A.;
2575 pubMeth::24=##12;:f;2A.;;",128,0,0,0
2578 @node Method Modifiers
2579 @section Method Modifiers (@code{const}, @code{volatile}, @code{const volatile})
2583 In the class example described above all the methods have the normal
2584 modifier. This method modifier information is located just after the
2585 protection information for the method. This field has four possible
2586 character values. Normal methods use @samp{A}, const methods use
2587 @samp{B}, volatile methods use @samp{C}, and const volatile methods use
2588 @samp{D}. Consider the class definition below:
2593 int ConstMeth (int arg) const @{ return arg; @};
2594 char VolatileMeth (char arg) volatile @{ return arg; @};
2595 float ConstVolMeth (float arg) const volatile @{return arg; @};
2599 This class is described by the following stab:
2602 .stabs "class(A):sym_desc(struct)type_def(20)=type_desc(struct)struct_bytes(1)
2603 meth_name(ConstMeth)::type_def(21)sym_desc(method)
2604 returning(int);:arg(int);protection(public)modifier(const)virtual(no);
2605 meth_name(VolatileMeth)::type_def(22)=sym_desc(method)
2606 returning(char);:arg(char);protection(public)modifier(volatile)virt(no)
2607 meth_name(ConstVolMeth)::type_def(23)=sym_desc(method)
2608 returning(float);:arg(float);protection(public)modifer(const volatile)
2609 virtual(no);;", @dots{}
2613 .stabs "A:T20=s1ConstMeth::21=##1;:i;2B.;VolatileMeth::22=##2;:c;2C.;
2614 ConstVolMeth::23=##12;:f;2D.;;",128,0,0,0
2617 @node Virtual Methods
2618 @section Virtual Methods
2620 << The following examples are based on a4.C >>
2622 The presence of virtual methods in a class definition adds additional
2623 data to the class description. The extra data is appended to the
2624 description of the virtual method and to the end of the class
2625 description. Consider the class definition below:
2631 virtual int A_virt (int arg) @{ return arg; @};
2635 This results in the stab below describing class A. It defines a new
2636 type (20) which is an 8 byte structure. The first field of the class
2637 struct is @samp{Adat}, an integer, starting at structure offset 0 and
2640 The second field in the class struct is not explicitly defined by the
2641 C++ class definition but is implied by the fact that the class
2642 contains a virtual method. This field is the vtable pointer. The
2643 name of the vtable pointer field starts with @samp{$vf} and continues with a
2644 type reference to the class it is part of. In this example the type
2645 reference for class A is 20 so the name of its vtable pointer field is
2646 @samp{$vf20}, followed by the usual colon.
2648 Next there is a type definition for the vtable pointer type (21).
2649 This is in turn defined as a pointer to another new type (22).
2651 Type 22 is the vtable itself, which is defined as an array, indexed by
2652 a range of integers between 0 and 1, and whose elements are of type
2653 17. Type 17 was the vtable record type defined by the boilerplate C++
2654 type definitions, as shown earlier.
2656 The bit offset of the vtable pointer field is 32. The number of bits
2657 in the field are not specified when the field is a vtable pointer.
2659 Next is the method definition for the virtual member function @code{A_virt}.
2660 Its description starts out using the same format as the non-virtual
2661 member functions described above, except instead of a dot after the
2662 @samp{A} there is an asterisk, indicating that the function is virtual.
2663 Since is is virtual some addition information is appended to the end
2664 of the method description.
2666 The first number represents the vtable index of the method. This is a
2667 32 bit unsigned number with the high bit set, followed by a
2670 The second number is a type reference to the first base class in the
2671 inheritence hierarchy defining the virtual member function. In this
2672 case the class stab describes a base class so the virtual function is
2673 not overriding any other definition of the method. Therefore the
2674 reference is to the type number of the class that the stab is
2677 This is followed by three semi-colons. One marks the end of the
2678 current sub-section, one marks the end of the method field, and the
2679 third marks the end of the struct definition.
2681 For classes containing virtual functions the very last section of the
2682 string part of the stab holds a type reference to the first base
2683 class. This is preceeded by @samp{~%} and followed by a final semi-colon.
2686 .stabs "class_name(A):type_def(20)=sym_desc(struct)struct_bytes(8)
2687 field_name(Adat):type_ref(int),bit_offset(0),field_bits(32);
2688 field_name(A virt func ptr):type_def(21)=type_desc(ptr to)type_def(22)=
2689 sym_desc(array)index_type_ref(range of int from 0 to 1);
2690 elem_type_ref(vtbl elem type),
2692 meth_name(A_virt)::typedef(23)=sym_desc(method)returning(int);
2693 :arg_type(int),protection(public)normal(yes)virtual(yes)
2694 vtable_index(1);class_first_defining(A);;;~%first_base(A);",
2698 @c FIXME: bogus line break.
2700 .stabs "A:t20=s8Adat:1,0,32;$vf20:21=*22=ar1;0;1;17,32;
2701 A_virt::23=##1;:i;2A*-2147483647;20;;;~%20;",128,0,0,0
2705 @section Inheritence
2707 Stabs describing C++ derived classes include additional sections that
2708 describe the inheritence hierarchy of the class. A derived class stab
2709 also encodes the number of base classes. For each base class it tells
2710 if the base class is virtual or not, and if the inheritence is private
2711 or public. It also gives the offset into the object of the portion of
2712 the object corresponding to each base class.
2714 This additional information is embeded in the class stab following the
2715 number of bytes in the struct. First the number of base classes
2716 appears bracketed by an exclamation point and a comma.
2718 Then for each base type there repeats a series: two digits, a number,
2719 a comma, another number, and a semi-colon.
2721 The first of the two digits is 1 if the base class is virtual and 0 if
2722 not. The second digit is 2 if the derivation is public and 0 if not.
2724 The number following the first two digits is the offset from the start
2725 of the object to the part of the object pertaining to the base class.
2727 After the comma, the second number is a type_descriptor for the base
2728 type. Finally a semi-colon ends the series, which repeats for each
2731 The source below defines three base classes @code{A}, @code{B}, and
2732 @code{C} and the derived class @code{D}.
2739 virtual int A_virt (int arg) @{ return arg; @};
2745 virtual int B_virt (int arg) @{return arg; @};
2751 virtual int C_virt (int arg) @{return arg; @};
2754 class D : A, virtual B, public C @{
2757 virtual int A_virt (int arg ) @{ return arg+1; @};
2758 virtual int B_virt (int arg) @{ return arg+2; @};
2759 virtual int C_virt (int arg) @{ return arg+3; @};
2760 virtual int D_virt (int arg) @{ return arg; @};
2764 Class stabs similar to the ones described earlier are generated for
2767 @c FIXME!!! the linebreaks in the following example probably make the
2768 @c examples literally unusable, but I don't know any other way to get
2769 @c them on the page.
2770 @c One solution would be to put some of the type definitions into
2771 @c separate stabs, even if that's not exactly what the compiler actually
2774 .stabs "A:T20=s8Adat:1,0,32;$vf20:21=*22=ar1;0;1;17,32;
2775 A_virt::23=##1;:i;2A*-2147483647;20;;;~%20;",128,0,0,0
2777 .stabs "B:Tt25=s8Bdat:1,0,32;$vf25:21,32;B_virt::26=##1;
2778 :i;2A*-2147483647;25;;;~%25;",128,0,0,0
2780 .stabs "C:Tt28=s8Cdat:1,0,32;$vf28:21,32;C_virt::29=##1;
2781 :i;2A*-2147483647;28;;;~%28;",128,0,0,0
2784 In the stab describing derived class @code{D} below, the information about
2785 the derivation of this class is encoded as follows.
2788 .stabs "derived_class_name:symbol_descriptors(struct tag&type)=
2789 type_descriptor(struct)struct_bytes(32)!num_bases(3),
2790 base_virtual(no)inheritence_public(no)base_offset(0),
2791 base_class_type_ref(A);
2792 base_virtual(yes)inheritence_public(no)base_offset(NIL),
2793 base_class_type_ref(B);
2794 base_virtual(no)inheritence_public(yes)base_offset(64),
2795 base_class_type_ref(C); @dots{}
2798 @c FIXME! fake linebreaks.
2800 .stabs "D:Tt31=s32!3,000,20;100,25;0264,28;$vb25:24,128;Ddat:
2801 1,160,32;A_virt::32=##1;:i;2A*-2147483647;20;;B_virt:
2802 :32:i;2A*-2147483647;25;;C_virt::32:i;2A*-2147483647;
2803 28;;D_virt::32:i;2A*-2147483646;31;;;~%20;",128,0,0,0
2806 @node Virtual Base Classes
2807 @section Virtual Base Classes
2809 A derived class object consists of a concatination in memory of the data
2810 areas defined by each base class, starting with the leftmost and ending
2811 with the rightmost in the list of base classes. The exception to this
2812 rule is for virtual inheritence. In the example above, class @code{D}
2813 inherits virtually from base class @code{B}. This means that an
2814 instance of a @code{D} object will not contain its own @code{B} part but
2815 merely a pointer to a @code{B} part, known as a virtual base pointer.
2817 In a derived class stab, the base offset part of the derivation
2818 information, described above, shows how the base class parts are
2819 ordered. The base offset for a virtual base class is always given as 0.
2820 Notice that the base offset for @code{B} is given as 0 even though
2821 @code{B} is not the first base class. The first base class @code{A}
2824 The field information part of the stab for class @code{D} describes the field
2825 which is the pointer to the virtual base class @code{B}. The vbase pointer
2826 name is @samp{$vb} followed by a type reference to the virtual base class.
2827 Since the type id for @code{B} in this example is 25, the vbase pointer name
2830 @c FIXME!! fake linebreaks below
2832 .stabs "D:Tt31=s32!3,000,20;100,25;0264,28;$vb25:24,128;Ddat:1,
2833 160,32;A_virt::32=##1;:i;2A*-2147483647;20;;B_virt::32:i;
2834 2A*-2147483647;25;;C_virt::32:i;2A*-2147483647;28;;D_virt:
2835 :32:i;2A*-2147483646;31;;;~%20;",128,0,0,0
2838 Following the name and a semicolon is a type reference describing the
2839 type of the virtual base class pointer, in this case 24. Type 24 was
2840 defined earlier as the type of the @code{B} class @code{this} pointer. The
2841 @code{this} pointer for a class is a pointer to the class type.
2844 .stabs "this:P24=*25=xsB:",64,0,0,8
2847 Finally the field offset part of the vbase pointer field description
2848 shows that the vbase pointer is the first field in the @code{D} object,
2849 before any data fields defined by the class. The layout of a @code{D}
2850 class object is a follows, @code{Adat} at 0, the vtable pointer for
2851 @code{A} at 32, @code{Cdat} at 64, the vtable pointer for C at 96, the
2852 virtual base pointer for @code{B} at 128, and @code{Ddat} at 160.
2855 @node Static Members
2856 @section Static Members
2858 The data area for a class is a concatenation of the space used by the
2859 data members of the class. If the class has virtual methods, a vtable
2860 pointer follows the class data. The field offset part of each field
2861 description in the class stab shows this ordering.
2863 << How is this reflected in stabs? See Cygnus bug #677 for some info. >>
2866 @appendix Table of Stab Types
2868 The following are all the possible values for the stab type field, for
2869 @code{a.out} files, in numeric order. This does not apply to XCOFF, but
2870 it does apply to stabs in ELF. Stabs in ECOFF use these values but add
2871 0x8f300 to distinguish them from non-stab symbols.
2873 The symbolic names are defined in the file @file{include/aout/stabs.def}.
2876 * Non-Stab Symbol Types:: Types from 0 to 0x1f
2877 * Stab Symbol Types:: Types from 0x20 to 0xff
2880 @node Non-Stab Symbol Types
2881 @appendixsec Non-Stab Symbol Types
2883 The following types are used by the linker and assembler, not by stab
2884 directives. Since this document does not attempt to describe aspects of
2885 object file format other than the debugging format, no details are
2888 @c Try to get most of these to fit on a single line.
2898 File scope absolute symbol
2900 @item 0x3 N_ABS | N_EXT
2901 External absolute symbol
2904 File scope text symbol
2906 @item 0x5 N_TEXT | N_EXT
2907 External text symbol
2910 File scope data symbol
2912 @item 0x7 N_DATA | N_EXT
2913 External data symbol
2916 File scope BSS symbol
2918 @item 0x9 N_BSS | N_EXT
2922 Same as @code{N_FN}, for Sequent compilers
2925 Symbol is indirected to another symbol
2928 Common---visible after shared library dynamic link
2931 Absolute set element
2934 Text segment set element
2937 Data segment set element
2940 BSS segment set element
2943 Pointer to set vector
2945 @item 0x1e N_WARNING
2946 Print a warning message during linking
2949 File name of a @file{.o} file
2952 @node Stab Symbol Types
2953 @appendixsec Stab Symbol Types
2955 The following symbol types indicate that this is a stab. This is the
2956 full list of stab numbers, including stab types that are used in
2957 languages other than C.
2961 Global symbol; see @ref{Global Variables}.
2964 Function name (for BSD Fortran); see @ref{Procedures}.
2967 Function name (@pxref{Procedures}) or text segment variable
2971 Data segment file-scope variable; see @ref{Statics}.
2974 BSS segment file-scope variable; see @ref{Statics}.
2977 Name of main routine; see @ref{Main Program}.
2980 Variable in @code{.rodata} section; see @ref{Statics}.
2983 Global symbol (for Pascal); see @ref{N_PC}.
2986 Number of symbols (according to Ultrix V4.0); see @ref{N_NSYMS}.
2989 No DST map; see @ref{N_NOMAP}.
2991 @c FIXME: describe this solaris feature in the body of the text (see
2992 @c comments in include/aout/stab.def).
2994 Object file (Solaris2).
2996 @c See include/aout/stab.def for (a little) more info.
2998 Debugger options (Solaris2).
3001 Register variable; see @ref{Register Variables}.
3004 Modula-2 compilation unit; see @ref{N_M2C}.
3007 Line number in text segment; see @ref{Line Numbers}.
3010 Line number in data segment; see @ref{Line Numbers}.
3013 Line number in bss segment; see @ref{Line Numbers}.
3016 Sun source code browser, path to @file{.cb} file; see @ref{N_BROWS}.
3019 GNU Modula2 definition module dependency; see @ref{N_DEFD}.
3022 Function start/body/end line numbers (Solaris2).
3025 GNU C++ exception variable; see @ref{N_EHDECL}.
3028 Modula2 info "for imc" (according to Ultrix V4.0); see @ref{N_MOD2}.
3031 GNU C++ @code{catch} clause; see @ref{N_CATCH}.
3034 Structure of union element; see @ref{N_SSYM}.
3037 Last stab for module (Solaris2).
3040 Path and name of source file; see @ref{Source Files}.
3043 Stack variable (@pxref{Stack Variables}) or type (@pxref{Typedefs}).
3046 Beginning of an include file (Sun only); see @ref{Include Files}.
3049 Name of include file; see @ref{Include Files}.
3052 Parameter variable; see @ref{Parameters}.
3055 End of an include file; see @ref{Include Files}.
3058 Alternate entry point; see @ref{N_ENTRY}.
3061 Beginning of a lexical block; see @ref{Block Structure}.
3064 Place holder for a deleted include file; see @ref{Include Files}.
3067 Modula2 scope information (Sun linker); see @ref{N_SCOPE}.
3070 End of a lexical block; see @ref{Block Structure}.
3073 Begin named common block; see @ref{Common Blocks}.
3076 End named common block; see @ref{Common Blocks}.
3079 Member of a common block; see @ref{Common Blocks}.
3081 @c FIXME: How does this really work? Move it to main body of document.
3083 Pascal @code{with} statement: type,,0,0,offset (Solaris2).
3086 Gould non-base registers; see @ref{Gould}.
3089 Gould non-base registers; see @ref{Gould}.
3092 Gould non-base registers; see @ref{Gould}.
3095 Gould non-base registers; see @ref{Gould}.
3098 Gould non-base registers; see @ref{Gould}.
3101 @c Restore the default table indent
3106 @node Symbol Descriptors
3107 @appendix Table of Symbol Descriptors
3109 The symbol descriptor is the character which follows the colon in many
3110 stabs, and which tells what kind of stab it is. @xref{String Field},
3111 for more information about their use.
3113 @c Please keep this alphabetical
3115 @c In TeX, this looks great, digit is in italics. But makeinfo insists
3116 @c on putting it in `', not realizing that @var should override @code.
3117 @c I don't know of any way to make makeinfo do the right thing. Seems
3118 @c like a makeinfo bug to me.
3122 Variable on the stack; see @ref{Stack Variables}.
3125 Parameter passed by reference in register; see @ref{Reference Parameters}.
3128 Based variable; see @ref{Based Variables}.
3131 Constant; see @ref{Constants}.
3134 Conformant array bound (Pascal, maybe other languages); @ref{Conformant
3135 Arrays}. Name of a caught exception (GNU C++). These can be
3136 distinguished because the latter uses @code{N_CATCH} and the former uses
3137 another symbol type.
3140 Floating point register variable; see @ref{Register Variables}.
3143 Parameter in floating point register; see @ref{Register Parameters}.
3146 File scope function; see @ref{Procedures}.
3149 Global function; see @ref{Procedures}.
3152 Global variable; see @ref{Global Variables}.
3155 @xref{Register Parameters}.
3158 Internal (nested) procedure; see @ref{Nested Procedures}.
3161 Internal (nested) function; see @ref{Nested Procedures}.
3164 Label name (documented by AIX, no further information known).
3167 Module; see @ref{Procedures}.
3170 Argument list parameter; see @ref{Parameters}.
3176 Fortran Function parameter; see @ref{Parameters}.
3179 Unfortunately, three separate meanings have been independently invented
3180 for this symbol descriptor. At least the GNU and Sun uses can be
3181 distinguished by the symbol type. Global Procedure (AIX) (symbol type
3182 used unknown); see @ref{Procedures}. Register parameter (GNU) (symbol
3183 type @code{N_PSYM}); see @ref{Parameters}. Prototype of function
3184 referenced by this file (Sun @code{acc}) (symbol type @code{N_FUN}).
3187 Static Procedure; see @ref{Procedures}.
3190 Register parameter; see @ref{Register Parameters}.
3193 Register variable; see @ref{Register Variables}.
3196 File scope variable; see @ref{Statics}.
3199 Type name; see @ref{Typedefs}.
3202 Enumeration, structure, or union tag; see @ref{Typedefs}.
3205 Parameter passed by reference; see @ref{Reference Parameters}.
3208 Procedure scope static variable; see @ref{Statics}.
3211 Conformant array; see @ref{Conformant Arrays}.
3214 Function return variable; see @ref{Parameters}.
3217 @node Type Descriptors
3218 @appendix Table of Type Descriptors
3220 The type descriptor is the character which follows the type number and
3221 an equals sign. It specifies what kind of type is being defined.
3222 @xref{String Field}, for more information about their use.
3227 Type reference; see @ref{String Field}.
3230 Reference to builtin type; see @ref{Negative Type Numbers}.
3233 Method (C++); see @ref{Cplusplus}.
3236 Pointer; see @ref{Miscellaneous Types}.
3242 Type Attributes (AIX); see @ref{String Field}. Member (class and variable)
3243 type (GNU C++); see @ref{Cplusplus}.
3246 Array; see @ref{Arrays}.
3249 Open array; see @ref{Arrays}.
3252 Pascal space type (AIX); see @ref{Miscellaneous Types}. Builtin integer
3253 type (Sun); see @ref{Builtin Type Descriptors}.
3256 Volatile-qualified type; see @ref{Miscellaneous Types}.
3259 Complex builtin type; see @ref{Builtin Type Descriptors}.
3262 COBOL Picture type. See AIX documentation for details.
3265 File type; see @ref{Miscellaneous Types}.
3268 N-dimensional dynamic array; see @ref{Arrays}.
3271 Enumeration type; see @ref{Enumerations}.
3274 N-dimensional subarray; see @ref{Arrays}.
3277 Function type; see @ref{Function Types}.
3280 Pascal function parameter; see @ref{Function Types}
3283 Builtin floating point type; see @ref{Builtin Type Descriptors}.
3286 COBOL Group. See AIX documentation for details.
3289 Imported type; see @ref{Cross-References}.
3292 Const-qualified type; see @ref{Miscellaneous Types}.
3295 COBOL File Descriptor. See AIX documentation for details.
3298 Multiple instance type; see @ref{Miscellaneous Types}.
3301 String type; see @ref{Strings}.
3304 Stringptr; see @ref{Strings}.
3307 Opaque type; see @ref{Typedefs}.
3310 Procedure; see @ref{Function Types}.
3313 Packed array; see @ref{Arrays}.
3316 Range type; see @ref{Subranges}.
3319 Builtin floating type; see @ref{Builtin Type Descriptors} (Sun). Pascal
3320 subroutine parameter; see @ref{Function Types} (AIX). Detecting this
3321 conflict is possible with careful parsing (hint: a Pascal subroutine
3322 parameter type will always contain a comma, and a builtin type
3323 descriptor never will).
3326 Structure type; see @ref{Structures}.
3329 Set type; see @ref{Miscellaneous Types}.
3332 Union; see @ref{Unions}.
3335 Variant record. This is a Pascal and Modula-2 feature which is like a
3336 union within a struct in C. See AIX documentation for details.
3339 Wide character; see @ref{Builtin Type Descriptors}.
3342 Cross-reference; see @ref{Cross-References}.
3345 gstring; see @ref{Strings}.
3348 @node Expanded Reference
3349 @appendix Expanded Reference by Stab Type
3351 @c FIXME: This appendix should go away; see N_PSYM or N_SO for an example.
3353 For a full list of stab types, and cross-references to where they are
3354 described, see @ref{Stab Types}. This appendix just duplicates certain
3355 information from the main body of this document; eventually the
3356 information will all be in one place.
3360 The first line is the symbol type (see @file{include/aout/stab.def}).
3362 The second line describes the language constructs the symbol type
3365 The third line is the stab format with the significant stab fields
3366 named and the rest NIL.
3368 Subsequent lines expand upon the meaning and possible values for each
3369 significant stab field. @samp{#} stands in for the type descriptor.
3371 Finally, any further information.
3374 * N_PC:: Pascal global symbol
3375 * N_NSYMS:: Number of symbols
3376 * N_NOMAP:: No DST map
3377 * N_M2C:: Modula-2 compilation unit
3378 * N_BROWS:: Path to .cb file for Sun source code browser
3379 * N_DEFD:: GNU Modula2 definition module dependency
3380 * N_EHDECL:: GNU C++ exception variable
3381 * N_MOD2:: Modula2 information "for imc"
3382 * N_CATCH:: GNU C++ "catch" clause
3383 * N_SSYM:: Structure or union element
3384 * N_ENTRY:: Alternate entry point
3385 * N_SCOPE:: Modula2 scope information (Sun only)
3386 * Gould:: non-base register symbols used on Gould systems
3387 * N_LENG:: Length of preceding entry
3393 @deffn @code{.stabs} N_PC
3395 Global symbol (for Pascal).
3398 "name" -> "symbol_name" <<?>>
3399 value -> supposedly the line number (stab.def is skeptical)
3403 @file{stabdump.c} says:
3405 global pascal symbol: name,,0,subtype,line
3413 @deffn @code{.stabn} N_NSYMS
3415 Number of symbols (according to Ultrix V4.0).
3418 0, files,,funcs,lines (stab.def)
3425 @deffn @code{.stabs} N_NOMAP
3427 No DST map for symbol (according to Ultrix V4.0). I think this means a
3428 variable has been optimized out.
3431 name, ,0,type,ignored (stab.def)
3438 @deffn @code{.stabs} N_M2C
3440 Modula-2 compilation unit.
3443 "string" -> "unit_name,unit_time_stamp[,code_time_stamp]"
3445 value -> 0 (main unit)
3453 @deffn @code{.stabs} N_BROWS
3455 Sun source code browser, path to @file{.cb} file
3458 "path to associated @file{.cb} file"
3460 Note: N_BROWS has the same value as N_BSLINE.
3466 @deffn @code{.stabn} N_DEFD
3468 GNU Modula2 definition module dependency.
3470 GNU Modula-2 definition module dependency. The value is the
3471 modification time of the definition file. The other field is non-zero
3472 if it is imported with the GNU M2 keyword @code{%INITIALIZE}. Perhaps
3473 @code{N_M2C} can be used if there are enough empty fields?
3479 @deffn @code{.stabs} N_EHDECL
3481 GNU C++ exception variable <<?>>.
3483 "@var{string} is variable name"
3485 Note: conflicts with @code{N_MOD2}.
3491 @deffn @code{.stab?} N_MOD2
3493 Modula2 info "for imc" (according to Ultrix V4.0)
3495 Note: conflicts with @code{N_EHDECL} <<?>>
3501 @deffn @code{.stabn} N_CATCH
3503 GNU C++ @code{catch} clause
3505 GNU C++ @code{catch} clause. The value is its address. The desc field
3506 is nonzero if this entry is immediately followed by a @code{CAUGHT} stab
3507 saying what exception was caught. Multiple @code{CAUGHT} stabs means
3508 that multiple exceptions can be caught here. If desc is 0, it means all
3509 exceptions are caught here.
3515 @deffn @code{.stabn} N_SSYM
3517 Structure or union element.
3519 The value is the offset in the structure.
3521 <<?looking at structs and unions in C I didn't see these>>
3527 @deffn @code{.stabn} N_ENTRY
3529 Alternate entry point.
3530 The value is its address.
3537 @deffn @code{.stab?} N_SCOPE
3539 Modula2 scope information (Sun linker)
3544 @section Non-base registers on Gould systems
3546 @deffn @code{.stab?} N_NBTEXT
3547 @deffnx @code{.stab?} N_NBDATA
3548 @deffnx @code{.stab?} N_NBBSS
3549 @deffnx @code{.stab?} N_NBSTS
3550 @deffnx @code{.stab?} N_NBLCS
3556 These are used on Gould systems for non-base registers syms.
3558 However, the following values are not the values used by Gould; they are
3559 the values which GNU has been documenting for these values for a long
3560 time, without actually checking what Gould uses. I include these values
3561 only because perhaps some someone actually did something with the GNU
3562 information (I hope not, why GNU knowingly assigned wrong values to
3563 these in the header file is a complete mystery to me).
3566 240 0xf0 N_NBTEXT ??
3567 242 0xf2 N_NBDATA ??
3577 @deffn @code{.stabn} N_LENG
3579 Second symbol entry containing a length-value for the preceding entry.
3580 The value is the length.
3584 @appendix Questions and Anomalies
3588 @c I think this is changed in GCC 2.4.5 to put the line number there.
3589 For GNU C stabs defining local and global variables (@code{N_LSYM} and
3590 @code{N_GSYM}), the desc field is supposed to contain the source
3591 line number on which the variable is defined. In reality the desc
3592 field is always 0. (This behavior is defined in @file{dbxout.c} and
3593 putting a line number in desc is controlled by @samp{#ifdef
3594 WINNING_GDB}, which defaults to false). GDB supposedly uses this
3595 information if you say @samp{list @var{var}}. In reality, @var{var} can
3596 be a variable defined in the program and GDB says @samp{function
3597 @var{var} not defined}.
3600 In GNU C stabs, there seems to be no way to differentiate tag types:
3601 structures, unions, and enums (symbol descriptor @samp{T}) and typedefs
3602 (symbol descriptor @samp{t}) defined at file scope from types defined locally
3603 to a procedure or other more local scope. They all use the @code{N_LSYM}
3604 stab type. Types defined at procedure scope are emited after the
3605 @code{N_RBRAC} of the preceding function and before the code of the
3606 procedure in which they are defined. This is exactly the same as
3607 types defined in the source file between the two procedure bodies.
3608 GDB overcompensates by placing all types in block #1, the block for
3609 symbols of file scope. This is true for default, @samp{-ansi} and
3610 @samp{-traditional} compiler options. (Bugs gcc/1063, gdb/1066.)
3613 What ends the procedure scope? Is it the proc block's @code{N_RBRAC} or the
3614 next @code{N_FUN}? (I believe its the first.)
3617 @c FIXME: This should go with the other stuff about global variables.
3618 Global variable stabs don't have location information. This comes
3619 from the external symbol for the same variable. The external symbol
3620 has a leading underbar on the _name of the variable and the stab does
3621 not. How do we know these two symbol table entries are talking about
3622 the same symbol when their names are different? (Answer: the debugger
3623 knows that external symbols have leading underbars).
3625 @c FIXME: This is absurdly vague; there all kinds of differences, some
3626 @c of which are the same between gnu & sun, and some of which aren't.
3627 @c In particular, I'm pretty sure GCC works with Sun dbx by default.
3629 @c Can GCC be configured to output stabs the way the Sun compiler
3630 @c does, so that their native debugging tools work? <NO?> It doesn't by
3631 @c default. GDB reads either format of stab. (GCC or SunC). How about
3635 @node XCOFF Differences
3636 @appendix Differences Between GNU Stabs in a.out and GNU Stabs in XCOFF
3638 @c FIXME: Merge *all* these into the main body of the document.
3639 The AIX/RS6000 native object file format is XCOFF with stabs. This
3640 appendix only covers those differences which are not covered in the main
3641 body of this document.
3645 BSD a.out stab types correspond to AIX XCOFF storage classes. In general
3646 the mapping is @code{N_@var{stabtype}} becomes @code{C_@var{stabtype}}.
3647 Some stab types in a.out are not supported in XCOFF; most of these use
3650 @c FIXME: Get C_* types for the block, figure out whether it is always
3651 @c used (I suspect not), explain clearly, and move to node Statics.
3652 Exception: initialised static @code{N_STSYM} and un-initialized static
3653 @code{N_LCSYM} both map to the @code{C_STSYM} storage class. But the
3654 distinction is preserved because in XCOFF @code{N_STSYM} and
3655 @code{N_LCSYM} must be emited in a named static block. Begin the block
3656 with @samp{.bs s[RW] data_section_name} for @code{N_STSYM} or @samp{.bs
3657 s bss_section_name} for @code{N_LCSYM}. End the block with @samp{.es}.
3659 @c FIXME: I think they are trying to say something about whether the
3660 @c assembler defaults the value to the location counter.
3662 If the XCOFF stab is an @code{N_FUN} (@code{C_FUN}) then follow the
3663 string field with @samp{,.} instead of just @samp{,}.
3666 I think that's it for @file{.s} file differences. They could stand to be
3667 better presented. This is just a list of what I have noticed so far.
3668 There are a @emph{lot} of differences in the information in the symbol
3669 tables of the executable and object files.
3671 Mapping of a.out stab types to XCOFF storage classes:
3674 stab type storage class
3675 -------------------------------
3711 @node Sun Differences
3712 @appendix Differences Between GNU Stabs and Sun Native Stabs
3714 @c FIXME: Merge all this stuff into the main body of the document.
3718 GNU C stabs define @emph{all} types, file or procedure scope, as
3719 @code{N_LSYM}. Sun doc talks about using @code{N_GSYM} too.
3722 Sun C stabs use type number pairs in the format
3723 (@var{file-number},@var{type-number}) where @var{file-number} is a
3724 number starting with 1 and incremented for each sub-source file in the
3725 compilation. @var{type-number} is a number starting with 1 and
3726 incremented for each new type defined in the compilation. GNU C stabs
3727 use the type number alone, with no source file number.
3731 @appendix Using Stabs With The ELF Object File Format
3733 The ELF object file format allows tools to create object files with
3734 custom sections containing any arbitrary data. To use stabs in ELF
3735 object files, the tools create two custom sections, a section named
3736 @code{.stab} which contains an array of fixed length structures, one
3737 struct per stab, and a section named @code{.stabstr} containing all the
3738 variable length strings that are referenced by stabs in the @code{.stab}
3739 section. The byte order of the stabs binary data matches the byte order
3740 of the ELF file itself, as determined from the @code{EI_DATA} field in
3741 the @code{e_ident} member of the ELF header.
3743 The first stab in the @code{.stab} section for each compilation unit is
3744 synthetic, generated entirely by the assembler, with no corresponding
3745 @code{.stab} directive as input to the assembler. This stab contains
3746 the following fields:
3750 Offset in the @code{.stabstr} section to the source filename.
3756 Unused field, always zero.
3759 Count of upcoming symbols, i.e., the number of remaining stabs for this
3763 Size of the string table fragment associated with this source file, in
3767 The @code{.stabstr} section always starts with a null byte (so that string
3768 offsets of zero reference a null string), followed by random length strings,
3769 each of which is null byte terminated.
3771 The ELF section header for the @code{.stab} section has its
3772 @code{sh_link} member set to the section number of the @code{.stabstr}
3773 section, and the @code{.stabstr} section has its ELF section
3774 header @code{sh_type} member set to @code{SHT_STRTAB} to mark it as a
3777 To keep linking fast, it is a bad idea to have the linker relocating
3778 stabs, so (except for a few cases, see below) none of the addresses in
3779 the @code{n_value} field of the stabs are relocated by the linker.
3780 Instead they are relative to the source file (or some entity smaller
3781 than a source file, like a function). To find the address of each
3782 section corresponding to a given source file, the compiler puts out
3783 symbols giving the address of each section for a given source file.
3784 Since these are ELF (not stab) symbols, the linker relocates them
3785 correctly without having to touch the stabs section. They are named
3786 @code{Bbss.bss} for the bss section, @code{Ddata.data} for the data
3787 section, and @code{Drodata.rodata} for the rodata section. For the text
3788 section, there is no such symbol (but there should be, see below). For
3789 an example of how these symbols work, @xref{ELF Transformations}. GCC
3790 does not provide these symbols; it instead relies on the stabs getting
3791 relocated, which slows down linking. Thus addresses which would
3792 normally be relative to @code{Bbss.bss}, etc., are already relocated.
3793 The Sun linker provided with Solaris 2.2 and earlier relocates stabs
3794 using normal ELF relocation information, as it would do for any section.
3795 Sun has been threatening to kludge their linker to not do this (to speed
3796 up linking), even though the correct way to avoid having the linker do
3797 these relocations is to have the compiler no longer output relocatable
3798 values. Last I heard they had been talked out of the linker kludge.
3799 See Sun point patch 101052-01 and Sun bug 1142109. This affects
3800 @samp{S} symbol descriptor stabs (@pxref{Statics}) and functions
3801 (@pxref{Procedures}). In the latter case, to adopt the clean solution
3802 (making the value of the stab relative to the start of the compilation
3803 unit), it would be necessary to invent a @code{Ttext.text} symbol,
3804 analogous to the @code{Bbss.bss}, etc., symbols. I recommend this
3805 rather than using a zero value and getting the address from the ELF
3808 @node Symbol Types Index
3809 @unnumbered Symbol Types Index