![[BACK]](/cvsweb/icons/back.gif){kind=icon}
Return to gcc.info-10 CVS log ![[TXT]](/cvsweb/icons/text.gif){kind=icon}
![[DIR]](/cvsweb/icons/dir.gif){kind=icon}
Up to [GCC 2.x] / gcc|
Return to gcc.info-10 CVS log | Up to [GCC 2.x] / gcc |
1.1.1.9 ! root 1: This is Info file gcc.info, produced by Makeinfo version 1.67 from the ! 2: input file gcc.texi. 1.1 root 3: 4: This file documents the use and the internals of the GNU compiler. 5: 1.1.1.8 root 6: Published by the Free Software Foundation 59 Temple Place - Suite 330 7: Boston, MA 02111-1307 USA 1.1.1.5 root 8: 1.1.1.8 root 9: Copyright (C) 1988, 1989, 1992, 1993, 1994, 1995 Free Software 10: Foundation, Inc. 1.1 root 11: 1.1.1.3 root 12: Permission is granted to make and distribute verbatim copies of this 13: manual provided the copyright notice and this permission notice are 14: preserved on all copies. 1.1 root 15: 16: Permission is granted to copy and distribute modified versions of 17: this manual under the conditions for verbatim copying, provided also 1.1.1.7 root 18: that the sections entitled "GNU General Public License," "Funding for 19: Free Software," and "Protect Your Freedom--Fight `Look And Feel'" are 20: included exactly as in the original, and provided that the entire 21: resulting derived work is distributed under the terms of a permission 22: notice identical to this one. 1.1 root 23: 24: Permission is granted to copy and distribute translations of this 25: manual into another language, under the above conditions for modified 1.1.1.3 root 26: versions, except that the sections entitled "GNU General Public 1.1.1.7 root 27: License," "Funding for Free Software," and "Protect Your Freedom--Fight 28: `Look And Feel'", and this permission notice, may be included in 29: translations approved by the Free Software Foundation instead of in the 30: original English. 1.1.1.3 root 31: 32: 1.1.1.8 root 33: File: gcc.info, Node: Asm Labels, Next: Explicit Reg Vars, Prev: Extended Asm, Up: C Extensions 1.1.1.6 root 34: 1.1.1.8 root 35: Controlling Names Used in Assembler Code 36: ======================================== 1.1.1.7 root 37: 1.1.1.8 root 38: You can specify the name to be used in the assembler code for a C 39: function or variable by writing the `asm' (or `__asm__') keyword after 40: the declarator as follows: 41: 42: int foo asm ("myfoo") = 2; 43: 44: This specifies that the name to be used for the variable `foo' in the 45: assembler code should be `myfoo' rather than the usual `_foo'. 46: 47: On systems where an underscore is normally prepended to the name of 48: a C function or variable, this feature allows you to define names for 49: the linker that do not start with an underscore. 50: 51: You cannot use `asm' in this way in a function *definition*; but you 52: can get the same effect by writing a declaration for the function 53: before its definition and putting `asm' there, like this: 54: 55: extern func () asm ("FUNC"); 56: 57: func (x, y) 58: int x, y; 59: ... 60: 61: It is up to you to make sure that the assembler names you choose do 62: not conflict with any other assembler symbols. Also, you must not use a 63: register name; that would produce completely invalid assembler code. 64: GNU CC does not as yet have the ability to store static variables in 65: registers. Perhaps that will be added. 66: 67: 68: File: gcc.info, Node: Explicit Reg Vars, Next: Alternate Keywords, Prev: Asm Labels, Up: C Extensions 69: 70: Variables in Specified Registers 71: ================================ 72: 73: GNU C allows you to put a few global variables into specified 74: hardware registers. You can also specify the register in which an 75: ordinary register variable should be allocated. 76: 77: * Global register variables reserve registers throughout the program. 78: This may be useful in programs such as programming language 79: interpreters which have a couple of global variables that are 80: accessed very often. 81: 82: * Local register variables in specific registers do not reserve the 83: registers. The compiler's data flow analysis is capable of 84: determining where the specified registers contain live values, and 85: where they are available for other uses. 86: 87: These local variables are sometimes convenient for use with the 88: extended `asm' feature (*note Extended Asm::.), if you want to 89: write one output of the assembler instruction directly into a 90: particular register. (This will work provided the register you 91: specify fits the constraints specified for that operand in the 92: `asm'.) 1.1.1.7 root 93: 1.1.1.8 root 94: * Menu: 1.1.1.7 root 95: 1.1.1.8 root 96: * Global Reg Vars:: 97: * Local Reg Vars:: 1.1.1.7 root 98: 99: 1.1.1.8 root 100: File: gcc.info, Node: Global Reg Vars, Next: Local Reg Vars, Up: Explicit Reg Vars 1.1.1.7 root 101: 1.1.1.8 root 102: Defining Global Register Variables 103: ---------------------------------- 1.1.1.7 root 104: 1.1.1.8 root 105: You can define a global register variable in GNU C like this: 106: 107: register int *foo asm ("a5"); 108: 109: Here `a5' is the name of the register which should be used. Choose a 110: register which is normally saved and restored by function calls on your 111: machine, so that library routines will not clobber it. 112: 113: Naturally the register name is cpu-dependent, so you would need to 114: conditionalize your program according to cpu type. The register `a5' 115: would be a good choice on a 68000 for a variable of pointer type. On 116: machines with register windows, be sure to choose a "global" register 117: that is not affected magically by the function call mechanism. 118: 119: In addition, operating systems on one type of cpu may differ in how 120: they name the registers; then you would need additional conditionals. 121: For example, some 68000 operating systems call this register `%a5'. 122: 123: Eventually there may be a way of asking the compiler to choose a 124: register automatically, but first we need to figure out how it should 125: choose and how to enable you to guide the choice. No solution is 126: evident. 127: 128: Defining a global register variable in a certain register reserves 129: that register entirely for this use, at least within the current 130: compilation. The register will not be allocated for any other purpose 131: in the functions in the current compilation. The register will not be 132: saved and restored by these functions. Stores into this register are 133: never deleted even if they would appear to be dead, but references may 134: be deleted or moved or simplified. 135: 136: It is not safe to access the global register variables from signal 137: handlers, or from more than one thread of control, because the system 138: library routines may temporarily use the register for other things 139: (unless you recompile them specially for the task at hand). 140: 141: It is not safe for one function that uses a global register variable 142: to call another such function `foo' by way of a third function `lose' 143: that was compiled without knowledge of this variable (i.e. in a 144: different source file in which the variable wasn't declared). This is 145: because `lose' might save the register and put some other value there. 146: For example, you can't expect a global register variable to be 147: available in the comparison-function that you pass to `qsort', since 148: `qsort' might have put something else in that register. (If you are 149: prepared to recompile `qsort' with the same global register variable, 150: you can solve this problem.) 151: 152: If you want to recompile `qsort' or other source files which do not 153: actually use your global register variable, so that they will not use 154: that register for any other purpose, then it suffices to specify the 155: compiler option `-ffixed-REG'. You need not actually add a global 156: register declaration to their source code. 157: 158: A function which can alter the value of a global register variable 159: cannot safely be called from a function compiled without this variable, 160: because it could clobber the value the caller expects to find there on 161: return. Therefore, the function which is the entry point into the part 162: of the program that uses the global register variable must explicitly 163: save and restore the value which belongs to its caller. 164: 165: On most machines, `longjmp' will restore to each global register 166: variable the value it had at the time of the `setjmp'. On some 167: machines, however, `longjmp' will not change the value of global 168: register variables. To be portable, the function that called `setjmp' 169: should make other arrangements to save the values of the global register 170: variables, and to restore them in a `longjmp'. This way, the same 171: thing will happen regardless of what `longjmp' does. 172: 173: All global register variable declarations must precede all function 174: definitions. If such a declaration could appear after function 175: definitions, the declaration would be too late to prevent the register 176: from being used for other purposes in the preceding functions. 177: 178: Global register variables may not have initial values, because an 179: executable file has no means to supply initial contents for a register. 180: 181: On the Sparc, there are reports that g3 ... g7 are suitable 182: registers, but certain library functions, such as `getwd', as well as 183: the subroutines for division and remainder, modify g3 and g4. g1 and 184: g2 are local temporaries. 1.1.1.7 root 185: 1.1.1.8 root 186: On the 68000, a2 ... a5 should be suitable, as should d2 ... d7. Of 187: course, it will not do to use more than a few of those. 1.1.1.7 root 188: 1.1.1.8 root 189: 190: File: gcc.info, Node: Local Reg Vars, Prev: Global Reg Vars, Up: Explicit Reg Vars 1.1.1.7 root 191: 1.1.1.8 root 192: Specifying Registers for Local Variables 193: ---------------------------------------- 1.1.1.7 root 194: 1.1.1.8 root 195: You can define a local register variable with a specified register 196: like this: 1.1.1.7 root 197: 1.1.1.8 root 198: register int *foo asm ("a5"); 1.1.1.7 root 199: 1.1.1.8 root 200: Here `a5' is the name of the register which should be used. Note that 201: this is the same syntax used for defining global register variables, 202: but for a local variable it would appear within a function. 1.1.1.7 root 203: 1.1.1.8 root 204: Naturally the register name is cpu-dependent, but this is not a 205: problem, since specific registers are most often useful with explicit 206: assembler instructions (*note Extended Asm::.). Both of these things 207: generally require that you conditionalize your program according to cpu 208: type. 1.1.1.7 root 209: 1.1.1.8 root 210: In addition, operating systems on one type of cpu may differ in how 211: they name the registers; then you would need additional conditionals. 212: For example, some 68000 operating systems call this register `%a5'. 1.1.1.7 root 213: 1.1.1.8 root 214: Eventually there may be a way of asking the compiler to choose a 215: register automatically, but first we need to figure out how it should 216: choose and how to enable you to guide the choice. No solution is 217: evident. 1.1.1.7 root 218: 1.1.1.8 root 219: Defining such a register variable does not reserve the register; it 220: remains available for other uses in places where flow control determines 221: the variable's value is not live. However, these registers are made 222: unavailable for use in the reload pass. I would not be surprised if 223: excessive use of this feature leaves the compiler too few available 224: registers to compile certain functions. 1.1.1.7 root 225: 226: 1.1.1.8 root 227: File: gcc.info, Node: Alternate Keywords, Next: Incomplete Enums, Prev: Explicit Reg Vars, Up: C Extensions 1.1.1.7 root 228: 1.1.1.8 root 229: Alternate Keywords 1.1.1.6 root 230: ================== 231: 1.1.1.8 root 232: The option `-traditional' disables certain keywords; `-ansi' 233: disables certain others. This causes trouble when you want to use GNU C 234: extensions, or ANSI C features, in a general-purpose header file that 235: should be usable by all programs, including ANSI C programs and 236: traditional ones. The keywords `asm', `typeof' and `inline' cannot be 237: used since they won't work in a program compiled with `-ansi', while 238: the keywords `const', `volatile', `signed', `typeof' and `inline' won't 239: work in a program compiled with `-traditional'. 240: 241: The way to solve these problems is to put `__' at the beginning and 242: end of each problematical keyword. For example, use `__asm__' instead 243: of `asm', `__const__' instead of `const', and `__inline__' instead of 244: `inline'. 245: 246: Other C compilers won't accept these alternative keywords; if you 247: want to compile with another compiler, you can define the alternate 248: keywords as macros to replace them with the customary keywords. It 249: looks like this: 250: 251: #ifndef __GNUC__ 252: #define __asm__ asm 253: #endif 254: 255: `-pedantic' causes warnings for many GNU C extensions. You can 256: prevent such warnings within one expression by writing `__extension__' 257: before the expression. `__extension__' has no effect aside from this. 258: 259: 260: File: gcc.info, Node: Incomplete Enums, Next: Function Names, Prev: Alternate Keywords, Up: C Extensions 261: 262: Incomplete `enum' Types 263: ======================= 264: 265: You can define an `enum' tag without specifying its possible values. 266: This results in an incomplete type, much like what you get if you write 267: `struct foo' without describing the elements. A later declaration 268: which does specify the possible values completes the type. 269: 270: You can't allocate variables or storage using the type while it is 271: incomplete. However, you can work with pointers to that type. 272: 273: This extension may not be very useful, but it makes the handling of 274: `enum' more consistent with the way `struct' and `union' are handled. 275: 276: This extension is not supported by GNU C++. 277: 278: 279: File: gcc.info, Node: Function Names, Prev: Incomplete Enums, Up: C Extensions 280: 281: Function Names as Strings 282: ========================= 283: 284: GNU CC predefines two string variables to be the name of the current 285: function. The variable `__FUNCTION__' is the name of the function as 286: it appears in the source. The variable `__PRETTY_FUNCTION__' is the 287: name of the function pretty printed in a language specific fashion. 288: 289: These names are always the same in a C function, but in a C++ 290: function they may be different. For example, this program: 291: 292: extern "C" { 293: extern int printf (char *, ...); 294: } 295: 296: class a { 297: public: 298: sub (int i) 299: { 300: printf ("__FUNCTION__ = %s\n", __FUNCTION__); 301: printf ("__PRETTY_FUNCTION__ = %s\n", __PRETTY_FUNCTION__); 302: } 303: }; 304: 305: int 306: main (void) 307: { 308: a ax; 309: ax.sub (0); 310: return 0; 311: } 312: 313: gives this output: 314: 315: __FUNCTION__ = sub 316: __PRETTY_FUNCTION__ = int a::sub (int) 317: 318: These names are not macros: they are predefined string variables. 319: For example, `#ifdef __FUNCTION__' does not have any special meaning 320: inside a function, since the preprocessor does not do anything special 321: with the identifier `__FUNCTION__'. 322: 323: 324: File: gcc.info, Node: C++ Extensions, Next: Trouble, Prev: C Extensions, Up: Top 325: 326: Extensions to the C++ Language 327: ****************************** 328: 329: The GNU compiler provides these extensions to the C++ language (and 330: you can also use most of the C language extensions in your C++ 331: programs). If you want to write code that checks whether these 332: features are available, you can test for the GNU compiler the same way 333: as for C programs: check for a predefined macro `__GNUC__'. You can 334: also use `__GNUG__' to test specifically for GNU C++ (*note Standard 335: Predefined Macros: (cpp.info)Standard Predefined.). 1.1.1.7 root 336: 337: * Menu: 338: 1.1.1.8 root 339: * Naming Results:: Giving a name to C++ function return values. 340: * Min and Max:: C++ Minimum and maximum operators. 341: * Destructors and Goto:: Goto is safe to use in C++ even when destructors 342: are needed. 343: * C++ Interface:: You can use a single C++ header file for both 344: declarations and definitions. 345: * Template Instantiation:: Methods for ensuring that exactly one copy of 346: each needed template instantiation is emitted. 347: * C++ Signatures:: You can specify abstract types to get subtype 348: polymorphism independent from inheritance. 349: 350: 351: File: gcc.info, Node: Naming Results, Next: Min and Max, Up: C++ Extensions 352: 353: Named Return Values in C++ 354: ========================== 355: 356: GNU C++ extends the function-definition syntax to allow you to 357: specify a name for the result of a function outside the body of the 358: definition, in C++ programs: 1.1.1.6 root 359: 1.1.1.8 root 360: TYPE 361: FUNCTIONNAME (ARGS) return RESULTNAME; 1.1.1.5 root 362: { 363: ... 1.1.1.8 root 364: BODY 365: ... 366: } 1.1.1.7 root 367: 1.1.1.8 root 368: You can use this feature to avoid an extra constructor call when a 369: function result has a class type. For example, consider a function 370: `m', declared as `X v = m ();', whose result is of class `X': 1.1.1.7 root 371: 1.1.1.8 root 372: X 373: m () 374: { 375: X b; 376: b.a = 23; 377: return b; 378: } 379: 380: Although `m' appears to have no arguments, in fact it has one 381: implicit argument: the address of the return value. At invocation, the 382: address of enough space to hold `v' is sent in as the implicit argument. 383: Then `b' is constructed and its `a' field is set to the value 23. 384: Finally, a copy constructor (a constructor of the form `X(X&)') is 385: applied to `b', with the (implicit) return value location as the 386: target, so that `v' is now bound to the return value. 387: 388: But this is wasteful. The local `b' is declared just to hold 389: something that will be copied right out. While a compiler that 390: combined an "elision" algorithm with interprocedural data flow analysis 391: could conceivably eliminate all of this, it is much more practical to 392: allow you to assist the compiler in generating efficient code by 393: manipulating the return value explicitly, thus avoiding the local 394: variable and copy constructor altogether. 395: 396: Using the extended GNU C++ function-definition syntax, you can avoid 397: the temporary allocation and copying by naming `r' as your return value 398: at the outset, and assigning to its `a' field directly: 1.1.1.7 root 399: 1.1.1.8 root 400: X 401: m () return r; 402: { 403: r.a = 23; 404: } 1.1.1.7 root 405: 1.1.1.8 root 406: The declaration of `r' is a standard, proper declaration, whose effects 407: are executed *before* any of the body of `m'. 1.1.1.7 root 408: 1.1.1.8 root 409: Functions of this type impose no additional restrictions; in 410: particular, you can execute `return' statements, or return implicitly by 411: reaching the end of the function body ("falling off the edge"). Cases 412: like 1.1.1.7 root 413: 1.1.1.8 root 414: X 415: m () return r (23); 416: { 417: return; 418: } 1.1.1.7 root 419: 1.1.1.8 root 420: (or even `X m () return r (23); { }') are unambiguous, since the return 421: value `r' has been initialized in either case. The following code may 422: be hard to read, but also works predictably: 1.1.1.7 root 423: 1.1.1.8 root 424: X 425: m () return r; 426: { 427: X b; 428: return b; 429: } 430: 431: The return value slot denoted by `r' is initialized at the outset, 432: but the statement `return b;' overrides this value. The compiler deals 433: with this by destroying `r' (calling the destructor if there is one, or 434: doing nothing if there is not), and then reinitializing `r' with `b'. 435: 436: This extension is provided primarily to help people who use 437: overloaded operators, where there is a great need to control not just 438: the arguments, but the return values of functions. For classes where 439: the copy constructor incurs a heavy performance penalty (especially in 440: the common case where there is a quick default constructor), this is a 441: major savings. The disadvantage of this extension is that you do not 442: control when the default constructor for the return value is called: it 443: is always called at the beginning. 444: 445: 446: File: gcc.info, Node: Min and Max, Next: Destructors and Goto, Prev: Naming Results, Up: C++ Extensions 447: 448: Minimum and Maximum Operators in C++ 449: ==================================== 450: 451: It is very convenient to have operators which return the "minimum" 452: or the "maximum" of two arguments. In GNU C++ (but not in GNU C), 453: 454: `A <? B' 455: is the "minimum", returning the smaller of the numeric values A 456: and B; 457: 458: `A >? B' 459: is the "maximum", returning the larger of the numeric values A and 460: B. 461: 462: These operations are not primitive in ordinary C++, since you can 463: use a macro to return the minimum of two things in C++, as in the 464: following example. 465: 466: #define MIN(X,Y) ((X) < (Y) ? : (X) : (Y)) 467: 468: You might then use `int min = MIN (i, j);' to set MIN to the minimum 469: value of variables I and J. 470: 471: However, side effects in `X' or `Y' may cause unintended behavior. 472: For example, `MIN (i++, j++)' will fail, incrementing the smaller 473: counter twice. A GNU C extension allows you to write safe macros that 474: avoid this kind of problem (*note Naming an Expression's Type: Naming 475: Types.). However, writing `MIN' and `MAX' as macros also forces you to 476: use function-call notation notation for a fundamental arithmetic 477: operation. Using GNU C++ extensions, you can write `int min = i <? j;' 478: instead. 479: 480: Since `<?' and `>?' are built into the compiler, they properly 481: handle expressions with side-effects; `int min = i++ <? j++;' works 482: correctly. 483: 484: 485: File: gcc.info, Node: Destructors and Goto, Next: C++ Interface, Prev: Min and Max, Up: C++ Extensions 486: 487: `goto' and Destructors in GNU C++ 488: ================================= 489: 490: In C++ programs, you can safely use the `goto' statement. When you 491: use it to exit a block which contains aggregates requiring destructors, 492: the destructors will run before the `goto' transfers control. (In ANSI 493: C++, `goto' is restricted to targets within the current block.) 494: 495: The compiler still forbids using `goto' to *enter* a scope that 496: requires constructors. 497: 498: 499: File: gcc.info, Node: C++ Interface, Next: Template Instantiation, Prev: Destructors and Goto, Up: C++ Extensions 500: 501: Declarations and Definitions in One Header 502: ========================================== 503: 504: C++ object definitions can be quite complex. In principle, your 505: source code will need two kinds of things for each object that you use 506: across more than one source file. First, you need an "interface" 507: specification, describing its structure with type declarations and 508: function prototypes. Second, you need the "implementation" itself. It 509: can be tedious to maintain a separate interface description in a header 510: file, in parallel to the actual implementation. It is also dangerous, 511: since separate interface and implementation definitions may not remain 512: parallel. 513: 514: With GNU C++, you can use a single header file for both purposes. 515: 516: *Warning:* The mechanism to specify this is in transition. For the 517: nonce, you must use one of two `#pragma' commands; in a future 518: release of GNU C++, an alternative mechanism will make these 519: `#pragma' commands unnecessary. 520: 521: The header file contains the full definitions, but is marked with 522: `#pragma interface' in the source code. This allows the compiler to 523: use the header file only as an interface specification when ordinary 524: source files incorporate it with `#include'. In the single source file 525: where the full implementation belongs, you can use either a naming 526: convention or `#pragma implementation' to indicate this alternate use 527: of the header file. 528: 529: `#pragma interface' 530: `#pragma interface "SUBDIR/OBJECTS.h"' 531: Use this directive in *header files* that define object classes, 532: to save space in most of the object files that use those classes. 533: Normally, local copies of certain information (backup copies of 534: inline member functions, debugging information, and the internal 535: tables that implement virtual functions) must be kept in each 536: object file that includes class definitions. You can use this 537: pragma to avoid such duplication. When a header file containing 538: `#pragma interface' is included in a compilation, this auxiliary 539: information will not be generated (unless the main input source 540: file itself uses `#pragma implementation'). Instead, the object 541: files will contain references to be resolved at link time. 542: 543: The second form of this directive is useful for the case where you 544: have multiple headers with the same name in different directories. 545: If you use this form, you must specify the same string to `#pragma 546: implementation'. 547: 548: `#pragma implementation' 549: `#pragma implementation "OBJECTS.h"' 550: Use this pragma in a *main input file*, when you want full output 551: from included header files to be generated (and made globally 552: visible). The included header file, in turn, should use `#pragma 553: interface'. Backup copies of inline member functions, debugging 554: information, and the internal tables used to implement virtual 555: functions are all generated in implementation files. 556: 557: If you use `#pragma implementation' with no argument, it applies to 558: an include file with the same basename(1) as your source file. 559: For example, in `allclass.cc', `#pragma implementation' by itself 560: is equivalent to `#pragma implementation "allclass.h"'. 561: 562: In versions of GNU C++ prior to 2.6.0 `allclass.h' was treated as 563: an implementation file whenever you would include it from 564: `allclass.cc' even if you never specified `#pragma 565: implementation'. This was deemed to be more trouble than it was 566: worth, however, and disabled. 567: 568: If you use an explicit `#pragma implementation', it must appear in 569: your source file *before* you include the affected header files. 570: 571: Use the string argument if you want a single implementation file to 572: include code from multiple header files. (You must also use 573: `#include' to include the header file; `#pragma implementation' 574: only specifies how to use the file--it doesn't actually include 575: it.) 576: 577: There is no way to split up the contents of a single header file 578: into multiple implementation files. 579: 580: `#pragma implementation' and `#pragma interface' also have an effect 581: on function inlining. 582: 583: If you define a class in a header file marked with `#pragma 584: interface', the effect on a function defined in that class is similar to 585: an explicit `extern' declaration--the compiler emits no code at all to 586: define an independent version of the function. Its definition is used 587: only for inlining with its callers. 588: 589: Conversely, when you include the same header file in a main source 590: file that declares it as `#pragma implementation', the compiler emits 591: code for the function itself; this defines a version of the function 592: that can be found via pointers (or by callers compiled without 593: inlining). If all calls to the function can be inlined, you can avoid 594: emitting the function by compiling with `-fno-implement-inlines'. If 595: any calls were not inlined, you will get linker errors. 596: 597: ---------- Footnotes ---------- 1.1.1.7 root 598: 1.1.1.8 root 599: (1) A file's "basename" was the name stripped of all leading path 600: information and of trailing suffixes, such as `.h' or `.C' or `.cc'. 1.1.1.7 root 601: 602: 1.1.1.8 root 603: File: gcc.info, Node: Template Instantiation, Next: C++ Signatures, Prev: C++ Interface, Up: C++ Extensions 1.1.1.7 root 604: 1.1.1.8 root 605: Where's the Template? 606: ===================== 1.1.1.7 root 607: 1.1.1.8 root 608: C++ templates are the first language feature to require more 609: intelligence from the environment than one usually finds on a UNIX 610: system. Somehow the compiler and linker have to make sure that each 611: template instance occurs exactly once in the executable if it is needed, 612: and not at all otherwise. There are two basic approaches to this 613: problem, which I will refer to as the Borland model and the Cfront 614: model. 615: 616: Borland model 617: Borland C++ solved the template instantiation problem by adding 618: the code equivalent of common blocks to their linker; template 619: instances are emitted in each translation unit that uses them, and 620: they are collapsed together at run time. The advantage of this 621: model is that the linker only has to consider the object files 622: themselves; there is no external complexity to worry about. This 623: disadvantage is that compilation time is increased because the 624: template code is being compiled repeatedly. Code written for this 625: model tends to include definitions of all member templates in the 626: header file, since they must be seen to be compiled. 627: 628: Cfront model 629: The AT&T C++ translator, Cfront, solved the template instantiation 630: problem by creating the notion of a template repository, an 631: automatically maintained place where template instances are 632: stored. As individual object files are built, notes are placed in 633: the repository to record where templates and potential type 634: arguments were seen so that the subsequent instantiation step 635: knows where to find them. At link time, any needed instances are 636: generated and linked in. The advantages of this model are more 637: optimal compilation speed and the ability to use the system 638: linker; to implement the Borland model a compiler vendor also 639: needs to replace the linker. The disadvantages are vastly 640: increased complexity, and thus potential for error; theoretically, 641: this should be just as transparent, but in practice it has been 642: very difficult to build multiple programs in one directory and one 643: program in multiple directories using Cfront. Code written for 644: this model tends to separate definitions of non-inline member 645: templates into a separate file, which is magically found by the 646: link preprocessor when a template needs to be instantiated. 647: 648: Currently, g++ implements neither automatic model. In the mean time, 649: you have three options for dealing with template instantiations: 650: 651: 1. Do nothing. Pretend g++ does implement automatic instantiation 652: management. Code written for the Borland model will work fine, but 653: each translation unit will contain instances of each of the 654: templates it uses. In a large program, this can lead to an 655: unacceptable amount of code duplication. 656: 657: 2. Add `#pragma interface' to all files containing template 658: definitions. For each of these files, add `#pragma implementation 659: "FILENAME"' to the top of some `.C' file which `#include's it. 660: Then compile everything with -fexternal-templates. The templates 661: will then only be expanded in the translation unit which 662: implements them (i.e. has a `#pragma implementation' line for the 663: file where they live); all other files will use external 664: references. If you're lucky, everything should work properly. If 665: you get undefined symbol errors, you need to make sure that each 666: template instance which is used in the program is used in the file 667: which implements that template. If you don't have any use for a 668: particular instance in that file, you can just instantiate it 669: explicitly, using the syntax from the latest C++ working paper: 670: 671: template class A<int>; 672: template ostream& operator << (ostream&, const A<int>&); 673: 674: This strategy will work with code written for either model. If 675: you are using code written for the Cfront model, the file 676: containing a class template and the file containing its member 677: templates should be implemented in the same translation unit. 678: 679: A slight variation on this approach is to use the flag 680: -falt-external-templates instead; this flag causes template 681: instances to be emitted in the translation unit that implements 682: the header where they are first instantiated, rather than the one 683: which implements the file where the templates are defined. This 684: header must be the same in all translation units, or things are 685: likely to break. 686: 687: *Note Declarations and Definitions in One Header: C++ Interface, 688: for more discussion of these pragmas. 689: 690: 3. Explicitly instantiate all the template instances you use, and 691: compile with -fno-implicit-templates. This is probably your best 692: bet; it may require more knowledge of exactly which templates you 693: are using, but it's less mysterious than the previous approach, 694: and it doesn't require any `#pragma's or other g++-specific code. 695: You can scatter the instantiations throughout your program, you 696: can create one big file to do all the instantiations, or you can 697: create tiny files like 1.1.1.7 root 698: 1.1.1.8 root 699: #include "Foo.h" 700: #include "Foo.cc" 701: 702: template class Foo<int>; 1.1.1.7 root 703: 1.1.1.8 root 704: for each instance you need, and create a template instantiation 705: library from those. I'm partial to the last, but your mileage may 706: vary. If you are using Cfront-model code, you can probably get 707: away with not using -fno-implicit-templates when compiling files 708: that don't `#include' the member template definitions. 709: 710: 711: File: gcc.info, Node: C++ Signatures, Prev: Template Instantiation, Up: C++ Extensions 712: 713: Type Abstraction using Signatures 714: ================================= 715: 716: In GNU C++, you can use the keyword `signature' to define a 717: completely abstract class interface as a datatype. You can connect this 718: abstraction with actual classes using signature pointers. If you want 719: to use signatures, run the GNU compiler with the `-fhandle-signatures' 720: command-line option. (With this option, the compiler reserves a second 721: keyword `sigof' as well, for a future extension.) 722: 723: Roughly, signatures are type abstractions or interfaces of classes. 724: Some other languages have similar facilities. C++ signatures are 725: related to ML's signatures, Haskell's type classes, definition modules 726: in Modula-2, interface modules in Modula-3, abstract types in Emerald, 727: type modules in Trellis/Owl, categories in Scratchpad II, and types in 728: POOL-I. For a more detailed discussion of signatures, see `Signatures: 729: A Language Extension for Improving Type Abstraction and Subtype 730: Polymorphism in C++' by Gerald Baumgartner and Vincent F. Russo (Tech 731: report CSD-TR-95-051, Dept. of Computer Sciences, Purdue University, 732: August 1995, a slightly improved version appeared in 733: *Software--Practice & Experience*, 25(8), pp. 863-889, August 1995). 734: You can get the tech report by anonymous FTP from `ftp.cs.purdue.edu' 735: in `pub/gb/Signature-design.ps.gz'. 736: 737: Syntactically, a signature declaration is a collection of member 738: function declarations and nested type declarations. For example, this 739: signature declaration defines a new abstract type `S' with member 740: functions `int foo ()' and `int bar (int)': 1.1.1.7 root 741: 1.1.1.8 root 742: signature S 743: { 744: int foo (); 745: int bar (int); 746: }; 1.1.1.7 root 747: 1.1.1.8 root 748: Since signature types do not include implementation definitions, you 749: cannot write an instance of a signature directly. Instead, you can 750: define a pointer to any class that contains the required interfaces as a 751: "signature pointer". Such a class "implements" the signature type. 752: 753: To use a class as an implementation of `S', you must ensure that the 754: class has public member functions `int foo ()' and `int bar (int)'. 755: The class can have other member functions as well, public or not; as 756: long as it offers what's declared in the signature, it is suitable as 757: an implementation of that signature type. 758: 759: For example, suppose that `C' is a class that meets the requirements 760: of signature `S' (`C' "conforms to" `S'). Then 761: 762: C obj; 763: S * p = &obj; 764: 765: defines a signature pointer `p' and initializes it to point to an 766: object of type `C'. The member function call `int i = p->foo ();' 767: executes `obj.foo ()'. 768: 769: Abstract virtual classes provide somewhat similar facilities in 770: standard C++. There are two main advantages to using signatures 771: instead: 772: 773: 1. Subtyping becomes independent from inheritance. A class or 774: signature type `T' is a subtype of a signature type `S' 775: independent of any inheritance hierarchy as long as all the member 776: functions declared in `S' are also found in `T'. So you can 777: define a subtype hierarchy that is completely independent from any 778: inheritance (implementation) hierarchy, instead of being forced to 779: use types that mirror the class inheritance hierarchy. 780: 781: 2. Signatures allow you to work with existing class hierarchies as 782: implementations of a signature type. If those class hierarchies 783: are only available in compiled form, you're out of luck with 784: abstract virtual classes, since an abstract virtual class cannot 785: be retrofitted on top of existing class hierarchies. So you would 786: be required to write interface classes as subtypes of the abstract 787: virtual class. 788: 789: There is one more detail about signatures. A signature declaration 790: can contain member function *definitions* as well as member function 791: declarations. A signature member function with a full definition is 792: called a *default implementation*; classes need not contain that 793: particular interface in order to conform. For example, a class `C' can 794: conform to the signature 1.1.1.7 root 795: 1.1.1.8 root 796: signature T 797: { 798: int f (int); 799: int f0 () { return f (0); }; 800: }; 1.1.1.7 root 801: 1.1.1.8 root 802: whether or not `C' implements the member function `int f0 ()'. If you 803: define `C::f0', that definition takes precedence; otherwise, the 804: default implementation `S::f0' applies. 1.1.1.7 root 805: 1.1.1.8 root 806: 807: File: gcc.info, Node: Trouble, Next: Bugs, Prev: C++ Extensions, Up: Top 1.1.1.7 root 808: 1.1.1.8 root 809: Known Causes of Trouble with GNU CC 810: *********************************** 1.1.1.7 root 811: 1.1.1.8 root 812: This section describes known problems that affect users of GNU CC. 813: Most of these are not GNU CC bugs per se--if they were, we would fix 814: them. But the result for a user may be like the result of a bug. 1.1.1.7 root 815: 1.1.1.8 root 816: Some of these problems are due to bugs in other software, some are 817: missing features that are too much work to add, and some are places 818: where people's opinions differ as to what is best. 1.1.1.7 root 819: 820: * Menu: 821: 1.1.1.8 root 822: * Actual Bugs:: Bugs we will fix later. 823: * Installation Problems:: Problems that manifest when you install GNU CC. 824: * Cross-Compiler Problems:: Common problems of cross compiling with GNU CC. 825: * Interoperation:: Problems using GNU CC with other compilers, 826: and with certain linkers, assemblers and debuggers. 827: * External Bugs:: Problems compiling certain programs. 828: * Incompatibilities:: GNU CC is incompatible with traditional C. 829: * Fixed Headers:: GNU C uses corrected versions of system header files. 830: This is necessary, but doesn't always work smoothly. 831: * Standard Libraries:: GNU C uses the system C library, which might not be 832: compliant with the ISO/ANSI C standard. 833: * Disappointments:: Regrettable things we can't change, but not quite bugs. 834: * C++ Misunderstandings:: Common misunderstandings with GNU C++. 835: * Protoize Caveats:: Things to watch out for when using `protoize'. 836: * Non-bugs:: Things we think are right, but some others disagree. 837: * Warnings and Errors:: Which problems in your code get warnings, 838: and which get errors. 839: 840: 841: File: gcc.info, Node: Actual Bugs, Next: Installation Problems, Up: Trouble 842: 843: Actual Bugs We Haven't Fixed Yet 844: ================================ 845: 846: * The `fixincludes' script interacts badly with automounters; if the 847: directory of system header files is automounted, it tends to be 848: unmounted while `fixincludes' is running. This would seem to be a 849: bug in the automounter. We don't know any good way to work around 850: it. 851: 852: * The `fixproto' script will sometimes add prototypes for the 853: `sigsetjmp' and `siglongjmp' functions that reference the 854: `jmp_buf' type before that type is defined. To work around this, 855: edit the offending file and place the typedef in front of the 856: prototypes. 857: 858: * There are several obscure case of mis-using struct, union, and 859: enum tags that are not detected as errors by the compiler. 860: 861: * When `-pedantic-errors' is specified, GNU C will incorrectly give 862: an error message when a function name is specified in an expression 863: involving the comma operator. 864: 865: * Loop unrolling doesn't work properly for certain C++ programs. 866: This is a bug in the C++ front end. It sometimes emits incorrect 867: debug info, and the loop unrolling code is unable to recover from 868: this error. 1.1.1.2 root 869:
This archive runs on limited infrastructure. Preserving old code on modern bandwidth. Automated agents are requested to crawl responsibly.