1 <?xml version=
"1.0" encoding=
"UTF-8" standalone=
"no"?>
2 <!DOCTYPE html PUBLIC
"-//W3C//DTD XHTML 1.1//EN" "http://www.w3.org/TR/xhtml11/DTD/xhtml11.dtd">
3 <html xmlns=
"http://www.w3.org/1999/xhtml"><head><title>Design Notes
</title><meta name=
"generator" content=
"DocBook XSL-NS Stylesheets V1.76.1"/><meta name=
"keywords" content=
" ISO C++ , library "/><link rel=
"home" href=
"../spine.html" title=
"The GNU C++ Library"/><link rel=
"up" href=
"appendix_contributing.html" title=
"Appendix A. Contributing"/><link rel=
"prev" href=
"source_code_style.html" title=
"Coding Style"/><link rel=
"next" href=
"appendix_porting.html" title=
"Appendix B. Porting and Maintenance"/></head><body><div class=
"navheader"><table width=
"100%" summary=
"Navigation header"><tr><th colspan=
"3" align=
"center">Design Notes
</th></tr><tr><td align=
"left"><a accesskey=
"p" href=
"source_code_style.html">Prev
</a> </td><th width=
"60%" align=
"center">Appendix A.
6 </th><td align=
"right"> <a accesskey=
"n" href=
"appendix_porting.html">Next
</a></td></tr></table><hr/></div><div class=
"section" title=
"Design Notes"><div class=
"titlepage"><div><div><h2 class=
"title"><a id=
"contrib.design_notes"/>Design Notes
</h2></div></div></div><p>
7 </p><div class=
"literallayout"><p><br/>
12 This paper is covers two major areas:
<br/>
14 - Features and policies not mentioned in the standard that
<br/>
15 the quality of the library implementation depends on, including
<br/>
16 extensions and
"implementation-defined" features;
<br/>
18 - Plans for required but unimplemented library features and
<br/>
19 optimizations to them.
<br/>
24 The standard defines a large library, much larger than the standard
<br/>
25 C library. A naive implementation would suffer substantial overhead
<br/>
26 in compile time, executable size, and speed, rendering it unusable
<br/>
27 in many (particularly embedded) applications. The alternative demands
<br/>
28 care in construction, and some compiler support, but there is no
<br/>
29 need for library subsets.
<br/>
31 What are the sources of this overhead? There are four main causes:
<br/>
33 - The library is specified almost entirely as templates, which
<br/>
34 with current compilers must be included in-line, resulting in
<br/>
35 very slow builds as tens or hundreds of thousands of lines
<br/>
36 of function definitions are read for each user source file.
<br/>
37 Indeed, the entire SGI STL, as well as the dos Reis valarray,
<br/>
38 are provided purely as header files, largely for simplicity in
<br/>
39 porting. Iostream/locale is (or will be) as large again.
<br/>
41 - The library is very flexible, specifying a multitude of hooks
<br/>
42 where users can insert their own code in place of defaults.
<br/>
43 When these hooks are not used, any time and code expended to
<br/>
44 support that flexibility is wasted.
<br/>
46 - Templates are often described as causing to
"code bloat". In
<br/>
47 practice, this refers (when it refers to anything real) to several
<br/>
48 independent processes. First, when a class template is manually
<br/>
49 instantiated in its entirely, current compilers place the definitions
<br/>
50 for all members in a single object file, so that a program linking
<br/>
51 to one member gets definitions of all. Second, template functions
<br/>
52 which do not actually depend on the template argument are, under
<br/>
53 current compilers, generated anew for each instantiation, rather
<br/>
54 than being shared with other instantiations. Third, some of the
<br/>
55 flexibility mentioned above comes from virtual functions (both in
<br/>
56 regular classes and template classes) which current linkers add
<br/>
57 to the executable file even when they manifestly cannot be called.
<br/>
59 - The library is specified to use a language feature, exceptions,
<br/>
60 which in the current gcc compiler ABI imposes a run time and
<br/>
61 code space cost to handle the possibility of exceptions even when
<br/>
62 they are not used. Under the new ABI (accessed with -fnew-abi),
<br/>
63 there is a space overhead and a small reduction in code efficiency
<br/>
64 resulting from lost optimization opportunities associated with
<br/>
65 non-local branches associated with exceptions.
<br/>
67 What can be done to eliminate this overhead? A variety of coding
<br/>
68 techniques, and compiler, linker and library improvements and
<br/>
69 extensions may be used, as covered below. Most are not difficult,
<br/>
70 and some are already implemented in varying degrees.
<br/>
72 Overhead: Compilation Time
<br/>
73 --------------------------
<br/>
75 Providing
"ready-instantiated" template code in object code archives
<br/>
76 allows us to avoid generating and optimizing template instantiations
<br/>
77 in each compilation unit which uses them. However, the number of such
<br/>
78 instantiations that are useful to provide is limited, and anyway this
<br/>
79 is not enough, by itself, to minimize compilation time. In particular,
<br/>
80 it does not reduce time spent parsing conforming headers.
<br/>
82 Quicker header parsing will depend on library extensions and compiler
<br/>
83 improvements. One approach is some variation on the techniques
<br/>
84 previously marketed as
"pre-compiled headers", now standardized as
<br/>
85 support for the
"export" keyword.
"Exported" template definitions
<br/>
86 can be placed (once) in a
"repository" -- really just a library, but
<br/>
87 of template definitions rather than object code -- to be drawn upon
<br/>
88 at link time when an instantiation is needed, rather than placed in
<br/>
89 header files to be parsed along with every compilation unit.
<br/>
91 Until
"export" is implemented we can put some of the lengthy template
<br/>
92 definitions in #if guards or alternative headers so that users can skip
<br/>
93 over the full definitions when they need only the ready-instantiated
<br/>
96 To be precise, this means that certain headers which define
<br/>
97 templates which users normally use only for certain arguments
<br/>
98 can be instrumented to avoid exposing the template definitions
<br/>
99 to the compiler unless a macro is defined. For example, in
<br/>
100 <string
>, we might have:
<br/>
102 template
<class _CharT, ...
> class basic_string {
<br/>
103 ... // member declarations
<br/>
105 ... // operator declarations
<br/>
107 #ifdef _STRICT_ISO_
<br/>
108 # if _G_NO_TEMPLATE_EXPORT
<br/>
109 # include
<bits/std_locale.h
> // headers needed by definitions
<br/>
111 # include
<bits/string.tcc
> // member and global template definitions.
<br/>
115 Users who compile without specifying a strict-ISO-conforming flag
<br/>
116 would not see many of the template definitions they now see, and rely
<br/>
117 instead on ready-instantiated specializations in the library. This
<br/>
118 technique would be useful for the following substantial components:
<br/>
119 string, locale/iostreams, valarray. It would *not* be useful or
<br/>
120 usable with the following: containers, algorithms, iterators,
<br/>
121 allocator. Since these constitute a large (though decreasing)
<br/>
122 fraction of the library, the benefit the technique offers is
<br/>
125 The language specifies the semantics of the
"export" keyword, but
<br/>
126 the gcc compiler does not yet support it. When it does, problems
<br/>
127 with large template inclusions can largely disappear, given some
<br/>
128 minor library reorganization, along with the need for the apparatus
<br/>
129 described above.
<br/>
131 Overhead: Flexibility Cost
<br/>
132 --------------------------
<br/>
134 The library offers many places where users can specify operations
<br/>
135 to be performed by the library in place of defaults. Sometimes
<br/>
136 this seems to require that the library use a more-roundabout, and
<br/>
137 possibly slower, way to accomplish the default requirements than
<br/>
138 would be used otherwise.
<br/>
140 The primary protection against this overhead is thorough compiler
<br/>
141 optimization, to crush out layers of inline function interfaces.
<br/>
142 Kuck
& Associates has demonstrated the practicality of this kind
<br/>
143 of optimization.
<br/>
145 The second line of defense against this overhead is explicit
<br/>
146 specialization. By defining helper function templates, and writing
<br/>
147 specialized code for the default case, overhead can be eliminated
<br/>
148 for that case without sacrificing flexibility. This takes full
<br/>
149 advantage of any ability of the optimizer to crush out degenerate
<br/>
152 The library specifies many virtual functions which current linkers
<br/>
153 load even when they cannot be called. Some minor improvements to the
<br/>
154 compiler and to ld would eliminate any such overhead by simply
<br/>
155 omitting virtual functions that the complete program does not call.
<br/>
156 A prototype of this work has already been done. For targets where
<br/>
157 GNU ld is not used, a
"pre-linker" could do the same job.
<br/>
159 The main areas in the standard interface where user flexibility
<br/>
160 can result in overhead are:
<br/>
162 - Allocators: Containers are specified to use user-definable
<br/>
163 allocator types and objects, making tuning for the container
<br/>
164 characteristics tricky.
<br/>
166 - Locales: the standard specifies locale objects used to implement
<br/>
167 iostream operations, involving many virtual functions which use
<br/>
168 streambuf iterators.
<br/>
170 - Algorithms and containers: these may be instantiated on any type,
<br/>
171 frequently duplicating code for identical operations.
<br/>
173 - Iostreams and strings: users are permitted to use these on their
<br/>
174 own types, and specify the operations the stream must use on these
<br/>
177 Note that these sources of overhead are _avoidable_. The techniques
<br/>
178 to avoid them are covered below.
<br/>
183 In the SGI STL, and in some other headers, many of the templates
<br/>
184 are defined
"inline" -- either explicitly or by their placement
<br/>
185 in class definitions -- which should not be inline. This is a
<br/>
186 source of code bloat. Matt had remarked that he was relying on
<br/>
187 the compiler to recognize what was too big to benefit from inlining,
<br/>
188 and generate it out-of-line automatically. However, this also can
<br/>
189 result in code bloat except where the linker can eliminate the extra
<br/>
192 Fixing these cases will require an audit of all inline functions
<br/>
193 defined in the library to determine which merit inlining, and moving
<br/>
194 the rest out of line. This is an issue mainly in chapters
23,
25, and
<br/>
195 27. Of course it can be done incrementally, and we should generally
<br/>
196 accept patches that move large functions out of line and into
".tcc"<br/>
197 files, which can later be pulled into a repository. Compiler/linker
<br/>
198 improvements to recognize very large inline functions and move them
<br/>
199 out-of-line, but shared among compilation units, could make this
<br/>
200 work unnecessary.
<br/>
202 Pre-instantiating template specializations currently produces large
<br/>
203 amounts of dead code which bloats statically linked programs. The
<br/>
204 current state of the static library, libstdc++.a, is intolerable on
<br/>
205 this account, and will fuel further confused speculation about a need
<br/>
206 for a library
"subset". A compiler improvement that treats each
<br/>
207 instantiated function as a separate object file, for linking purposes,
<br/>
208 would be one solution to this problem. An alternative would be to
<br/>
209 split up the manual instantiation files into dozens upon dozens of
<br/>
210 little files, each compiled separately, but an abortive attempt at
<br/>
211 this was done for
<string
> and, though it is far from complete, it
<br/>
212 is already a nuisance. A better interim solution (just until we have
<br/>
213 "export") is badly needed.
<br/>
215 When building a shared library, the current compiler/linker cannot
<br/>
216 automatically generate the instantiations needed. This creates a
<br/>
217 miserable situation; it means any time something is changed in the
<br/>
218 library, before a shared library can be built someone must manually
<br/>
219 copy the declarations of all templates that are needed by other parts
<br/>
220 of the library to an
"instantiation" file, and add it to the build
<br/>
221 system to be compiled and linked to the library. This process is
<br/>
222 readily automated, and should be automated as soon as possible.
<br/>
223 Users building their own shared libraries experience identical
<br/>
226 Sharing common aspects of template definitions among instantiations
<br/>
227 can radically reduce code bloat. The compiler could help a great
<br/>
228 deal here by recognizing when a function depends on nothing about
<br/>
229 a template parameter, or only on its size, and giving the resulting
<br/>
230 function a link-name
"equate" that allows it to be shared with other
<br/>
231 instantiations. Implementation code could take advantage of the
<br/>
232 capability by factoring out code that does not depend on the template
<br/>
233 argument into separate functions to be merged by the compiler.
<br/>
235 Until such a compiler optimization is implemented, much can be done
<br/>
236 manually (if tediously) in this direction. One such optimization is
<br/>
237 to derive class templates from non-template classes, and move as much
<br/>
238 implementation as possible into the base class. Another is to partial-
<br/>
239 specialize certain common instantiations, such as vector
<T*
>, to share
<br/>
240 code for instantiations on all types T. While these techniques work,
<br/>
241 they are far from the complete solution that a compiler improvement
<br/>
244 Overhead: Expensive Language Features
<br/>
245 -------------------------------------
<br/>
247 The main
"expensive" language feature used in the standard library
<br/>
248 is exception support, which requires compiling in cleanup code with
<br/>
249 static table data to locate it, and linking in library code to use
<br/>
250 the table. For small embedded programs the amount of such library
<br/>
251 code and table data is assumed by some to be excessive. Under the
<br/>
252 "new" ABI this perception is generally exaggerated, although in some
<br/>
253 cases it may actually be excessive.
<br/>
255 To implement a library which does not use exceptions directly is
<br/>
256 not difficult given minor compiler support (to
"turn off" exceptions
<br/>
257 and ignore exception constructs), and results in no great library
<br/>
258 maintenance difficulties. To be precise, given
"-fno-exceptions",
<br/>
259 the compiler should treat
"try" blocks as ordinary blocks, and
<br/>
260 "catch" blocks as dead code to ignore or eliminate. Compiler
<br/>
261 support is not strictly necessary, except in the case of
"function<br/>
262 try blocks"; otherwise the following macros almost suffice:
<br/>
264 #define throw(X)
<br/>
265 #define try if (true)
<br/>
266 #define catch(X) else if (false)
<br/>
268 However, there may be a need to use function try blocks in the
<br/>
269 library implementation, and use of macros in this way can make
<br/>
270 correct diagnostics impossible. Furthermore, use of this scheme
<br/>
271 would require the library to call a function to re-throw exceptions
<br/>
272 from a try block. Implementing the above semantics in the compiler
<br/>
275 Given the support above (however implemented) it only remains to
<br/>
276 replace code that
"throws" with a call to a well-documented
"handler"<br/>
277 function in a separate compilation unit which may be replaced by
<br/>
278 the user. The main source of exceptions that would be difficult
<br/>
279 for users to avoid is memory allocation failures, but users can
<br/>
280 define their own memory allocation primitives that never throw.
<br/>
281 Otherwise, the complete list of such handlers, and which library
<br/>
282 functions may call them, would be needed for users to be able to
<br/>
283 implement the necessary substitutes. (Fortunately, they have the
<br/>
289 The template capabilities of C++ offer enormous opportunities for
<br/>
290 optimizing common library operations, well beyond what would be
<br/>
291 considered
"eliminating overhead". In particular, many operations
<br/>
292 done in Glibc with macros that depend on proprietary language
<br/>
293 extensions can be implemented in pristine Standard C++. For example,
<br/>
294 the chapter
25 algorithms, and even C library functions such as strchr,
<br/>
295 can be specialized for the case of static arrays of known (small) size.
<br/>
297 Detailed optimization opportunities are identified below where
<br/>
298 the component where they would appear is discussed. Of course new
<br/>
299 opportunities will be identified during implementation.
<br/>
301 Unimplemented Required Library Features
<br/>
302 ---------------------------------------
<br/>
304 The standard specifies hundreds of components, grouped broadly by
<br/>
305 chapter. These are listed in excruciating detail in the CHECKLIST
<br/>
319 Annex D backward compatibility
<br/>
321 Anyone participating in implementation of the library should obtain
<br/>
322 a copy of the standard, ISO
14882. People in the U.S. can obtain an
<br/>
323 electronic copy for US$
18 from ANSI's web site. Those from other
<br/>
324 countries should visit http://www.iso.org/ to find out the location
<br/>
325 of their country's representation in ISO, in order to know who can
<br/>
326 sell them a copy.
<br/>
328 The emphasis in the following sections is on unimplemented features
<br/>
329 and optimization opportunities.
<br/>
331 Chapter
17 General
<br/>
332 -------------------
<br/>
334 Chapter
17 concerns overall library requirements.
<br/>
336 The standard doesn't mention threads. A multi-thread (MT) extension
<br/>
337 primarily affects operators new and delete (
18), allocator (
20),
<br/>
338 string (
21), locale (
22), and iostreams (
27). The common underlying
<br/>
339 support needed for this is discussed under chapter
20.
<br/>
341 The standard requirements on names from the C headers create a
<br/>
342 lot of work, mostly done. Names in the C headers must be visible
<br/>
343 in the std:: and sometimes the global namespace; the names in the
<br/>
344 two scopes must refer to the same object. More stringent is that
<br/>
345 Koenig lookup implies that any types specified as defined in std::
<br/>
346 really are defined in std::. Names optionally implemented as
<br/>
347 macros in C cannot be macros in C++. (An overview may be read at
<br/>
348 <http://www.cantrip.org/cheaders.html
>). The scripts
"inclosure"<br/>
349 and
"mkcshadow", and the directories shadow/ and cshadow/, are the
<br/>
350 beginning of an effort to conform in this area.
<br/>
352 A correct conforming definition of C header names based on underlying
<br/>
353 C library headers, and practical linking of conforming namespaced
<br/>
354 customer code with third-party C libraries depends ultimately on
<br/>
355 an ABI change, allowing namespaced C type names to be mangled into
<br/>
356 type names as if they were global, somewhat as C function names in a
<br/>
357 namespace, or C++ global variable names, are left unmangled. Perhaps
<br/>
358 another
"extern" mode, such as 'extern
"C-global"' would be an
<br/>
359 appropriate place for such type definitions. Such a type would
<br/>
360 affect mangling as follows:
<br/>
364 extern
"C-global" { // or maybe just 'extern
"C"'
<br/>
368 void f(A::X*); // mangles to f__FPQ21A1X
<br/>
369 void f(A::Y*); // mangles to f__FP1Y
<br/>
371 (It may be that this is really the appropriate semantics for regular
<br/>
372 'extern
"C"', and 'extern
"C-global"', as an extension, would not be
<br/>
373 necessary.) This would allow functions declared in non-standard C headers
<br/>
374 (and thus fixable by neither us nor users) to link properly with functions
<br/>
375 declared using C types defined in properly-namespaced headers. The
<br/>
376 problem this solves is that C headers (which C++ programmers do persist
<br/>
377 in using) frequently forward-declare C struct tags without including
<br/>
378 the header where the type is defined, as in
<br/>
381 void munge(tm*);
<br/>
383 Without some compiler accommodation, munge cannot be called by correct
<br/>
384 C++ code using a pointer to a correctly-scoped tm* value.
<br/>
386 The current C headers use the preprocessor extension
"#include_next",
<br/>
387 which the compiler complains about when run
"-pedantic".
<br/>
388 (Incidentally, it appears that
"-fpedantic" is currently ignored,
<br/>
389 probably a bug.) The solution in the C compiler is to use
<br/>
390 "-isystem" rather than
"-I", but unfortunately in g++ this seems
<br/>
391 also to wrap the whole header in an 'extern
"C"' block, so it's
<br/>
392 unusable for C++ headers. The correct solution appears to be to
<br/>
393 allow the various special include-directory options, if not given
<br/>
394 an argument, to affect subsequent include-directory options additively,
<br/>
395 so that if one said
<br/>
397 -pedantic -iprefix $(prefix) \
<br/>
398 -idirafter -ino-pedantic -ino-extern-c -iwithprefix -I g++-v3 \
<br/>
399 -iwithprefix -I g++-v3/ext
<br/>
401 the compiler would search $(prefix)/g++-v3 and not report
<br/>
402 pedantic warnings for files found there, but treat files in
<br/>
403 $(prefix)/g++-v3/ext pedantically. (The undocumented semantics
<br/>
404 of
"-isystem" in g++ stink. Can they be rescinded? If not it
<br/>
405 must be replaced with something more rationally behaved.)
<br/>
407 All the C headers need the treatment above; in the standard these
<br/>
408 headers are mentioned in various chapters. Below, I have only
<br/>
409 mentioned those that present interesting implementation issues.
<br/>
411 The components identified as
"mostly complete", below, have not been
<br/>
412 audited for conformance. In many cases where the library passes
<br/>
413 conformance tests we have non-conforming extensions that must be
<br/>
414 wrapped in #if guards for
"pedantic" use, and in some cases renamed
<br/>
415 in a conforming way for continued use in the implementation regardless
<br/>
416 of conformance flags.
<br/>
418 The STL portion of the library still depends on a header
<br/>
419 stl/bits/stl_config.h full of #ifdef clauses. This apparatus
<br/>
420 should be replaced with autoconf/automake machinery.
<br/>
422 The SGI STL defines a type_traits
<> template, specialized for
<br/>
423 many types in their code including the built-in numeric and
<br/>
424 pointer types and some library types, to direct optimizations of
<br/>
425 standard functions. The SGI compiler has been extended to generate
<br/>
426 specializations of this template automatically for user types,
<br/>
427 so that use of STL templates on user types can take advantage of
<br/>
428 these optimizations. Specializations for other, non-STL, types
<br/>
429 would make more optimizations possible, but extending the gcc
<br/>
430 compiler in the same way would be much better. Probably the next
<br/>
431 round of standardization will ratify this, but probably with
<br/>
432 changes, so it probably should be renamed to place it in the
<br/>
433 implementation namespace.
<br/>
435 The SGI STL also defines a large number of extensions visible in
<br/>
436 standard headers. (Other extensions that appear in separate headers
<br/>
437 have been sequestered in subdirectories ext/ and backward/.) All
<br/>
438 these extensions should be moved to other headers where possible,
<br/>
439 and in any case wrapped in a namespace (not std!), and (where kept
<br/>
440 in a standard header) girded about with macro guards. Some cannot be
<br/>
441 moved out of standard headers because they are used to implement
<br/>
442 standard features. The canonical method for accommodating these
<br/>
443 is to use a protected name, aliased in macro guards to a user-space
<br/>
444 name. Unfortunately C++ offers no satisfactory template typedef
<br/>
445 mechanism, so very ad-hoc and unsatisfactory aliasing must be used
<br/>
448 Implementation of a template typedef mechanism should have the highest
<br/>
449 priority among possible extensions, on the same level as implementation
<br/>
450 of the template
"export" feature.
<br/>
452 Chapter
18 Language support
<br/>
453 ----------------------------
<br/>
455 Headers:
<limits
> <new
> <typeinfo
> <exception
><br/>
456 C headers:
<cstddef
> <climits
> <cfloat
> <cstdarg
> <csetjmp
><br/>
457 <ctime
> <csignal
> <cstdlib
> (also
21,
25,
26)
<br/>
459 This defines the built-in exceptions, rtti, numeric_limits
<>,
<br/>
460 operator new and delete. Much of this is provided by the
<br/>
461 compiler in its static runtime library.
<br/>
463 Work to do includes defining numeric_limits
<> specializations in
<br/>
464 separate files for all target architectures. Values for integer types
<br/>
465 except for bool and wchar_t are readily obtained from the C header
<br/>
466 <limits.h
>, but values for the remaining numeric types (bool, wchar_t,
<br/>
467 float, double, long double) must be entered manually. This is
<br/>
468 largely dog work except for those members whose values are not
<br/>
469 easily deduced from available documentation. Also, this involves
<br/>
470 some work in target configuration to identify the correct choice of
<br/>
471 file to build against and to install.
<br/>
473 The definitions of the various operators new and delete must be
<br/>
474 made thread-safe, which depends on a portable exclusion mechanism,
<br/>
475 discussed under chapter
20. Of course there is always plenty of
<br/>
476 room for improvements to the speed of operators new and delete.
<br/>
478 <cstdarg
>, in Glibc, defines some macros that gcc does not allow to
<br/>
479 be wrapped into an inline function. Probably this header will demand
<br/>
480 attention whenever a new target is chosen. The functions atexit(),
<br/>
481 exit(), and abort() in cstdlib have different semantics in C++, so
<br/>
482 must be re-implemented for C++.
<br/>
484 Chapter
19 Diagnostics
<br/>
485 -----------------------
<br/>
487 Headers:
<stdexcept
><br/>
488 C headers:
<cassert
> <cerrno
><br/>
490 This defines the standard exception objects, which are
"mostly complete".
<br/>
491 Cygnus has a version, and now SGI provides a slightly different one.
<br/>
492 It makes little difference which we use.
<br/>
494 The C global name
"errno", which C allows to be a variable or a macro,
<br/>
495 is required in C++ to be a macro. For MT it must typically result in
<br/>
496 a function call.
<br/>
498 Chapter
20 Utilities
<br/>
499 ---------------------
<br/>
500 Headers:
<utility
> <functional
> <memory
><br/>
501 C header:
<ctime
> (also in
18)
<br/>
503 SGI STL provides
"mostly complete" versions of all the components
<br/>
504 defined in this chapter. However, the auto_ptr
<> implementation
<br/>
505 is known to be wrong. Furthermore, the standard definition of it
<br/>
506 is known to be unimplementable as written. A minor change to the
<br/>
507 standard would fix it, and auto_ptr
<> should be adjusted to match.
<br/>
509 Multi-threading affects the allocator implementation, and there must
<br/>
510 be configuration/installation choices for different users' MT
<br/>
511 requirements. Anyway, users will want to tune allocator options
<br/>
512 to support different target conditions, MT or no.
<br/>
514 The primitives used for MT implementation should be exposed, as an
<br/>
515 extension, for users' own work. We need cross-CPU
"mutex" support,
<br/>
516 multi-processor shared-memory atomic integer operations, and single-
<br/>
517 processor uninterruptible integer operations, and all three configurable
<br/>
518 to be stubbed out for non-MT use, or to use an appropriately-loaded
<br/>
519 dynamic library for the actual runtime environment, or statically
<br/>
520 compiled in for cases where the target architecture is known.
<br/>
522 Chapter
21 String
<br/>
523 ------------------
<br/>
524 Headers:
<string
><br/>
525 C headers:
<cctype
> <cwctype
> <cstring
> <cwchar
> (also in
27)
<br/>
526 <cstdlib
> (also in
18,
25,
26)
<br/>
528 We have
"mostly-complete" char_traits
<> implementations. Many of the
<br/>
529 char_traits
<char
> operations might be optimized further using existing
<br/>
530 proprietary language extensions.
<br/>
532 We have a
"mostly-complete" basic_string
<> implementation. The work
<br/>
533 to manually instantiate char and wchar_t specializations in object
<br/>
534 files to improve link-time behavior is extremely unsatisfactory,
<br/>
535 literally tripling library-build time with no commensurate improvement
<br/>
536 in static program link sizes. It must be redone. (Similar work is
<br/>
537 needed for some components in chapters
22 and
27.)
<br/>
539 Other work needed for strings is MT-safety, as discussed under the
<br/>
540 chapter
20 heading.
<br/>
542 The standard C type mbstate_t from
<cwchar
> and used in char_traits
<><br/>
543 must be different in C++ than in C, because in C++ the default constructor
<br/>
544 value mbstate_t() must be the
"base" or
"ground" sequence state.
<br/>
545 (According to the likely resolution of a recently raised Core issue,
<br/>
546 this may become unnecessary. However, there are other reasons to
<br/>
547 use a state type not as limited as whatever the C library provides.)
<br/>
548 If we might want to provide conversions from (e.g.) internally-
<br/>
549 represented EUC-wide to externally-represented Unicode, or vice-
<br/>
550 versa, the mbstate_t we choose will need to be more accommodating
<br/>
551 than what might be provided by an underlying C library.
<br/>
553 There remain some basic_string template-member functions which do
<br/>
554 not overload properly with their non-template brethren. The infamous
<br/>
555 hack akin to what was done in vector
<> is needed, to conform to
<br/>
556 23.1.1 para
10. The CHECKLIST items for basic_string marked 'X',
<br/>
557 or incomplete, are so marked for this reason.
<br/>
559 Replacing the string iterators, which currently are simple character
<br/>
560 pointers, with class objects would greatly increase the safety of the
<br/>
561 client interface, and also permit a
"debug" mode in which range,
<br/>
562 ownership, and validity are rigorously checked. The current use of
<br/>
563 raw pointers as string iterators is evil. vector
<> iterators need the
<br/>
564 same treatment. Note that the current implementation freely mixes
<br/>
565 pointers and iterators, and that must be fixed before safer iterators
<br/>
566 can be introduced.
<br/>
568 Some of the functions in
<cstring
> are different from the C version.
<br/>
569 generally overloaded on const and non-const argument pointers. For
<br/>
570 example, in
<cstring
> strchr is overloaded. The functions isupper
<br/>
571 etc. in
<cctype
> typically implemented as macros in C are functions
<br/>
572 in C++, because they are overloaded with others of the same name
<br/>
573 defined in
<locale
>.
<br/>
575 Many of the functions required in
<cwctype
> and
<cwchar
> cannot be
<br/>
576 implemented using underlying C facilities on intended targets because
<br/>
577 such facilities only partly exist.
<br/>
579 Chapter
22 Locale
<br/>
580 ------------------
<br/>
581 Headers:
<locale
><br/>
582 C headers:
<clocale
><br/>
584 We have a
"mostly complete" class locale, with the exception of
<br/>
585 code for constructing, and handling the names of, named locales.
<br/>
586 The ways that locales are named (particularly when categories
<br/>
587 (e.g. LC_TIME, LC_COLLATE) are different) varies among all target
<br/>
588 environments. This code must be written in various versions and
<br/>
589 chosen by configuration parameters.
<br/>
591 Members of many of the facets defined in
<locale
> are stubs. Generally,
<br/>
592 there are two sets of facets: the base class facets (which are supposed
<br/>
593 to implement the
"C" locale) and the
"byname" facets, which are supposed
<br/>
594 to read files to determine their behavior. The base ctype
<>, collate
<>,
<br/>
595 and numpunct
<> facets are
"mostly complete", except that the table of
<br/>
596 bitmask values used for
"is" operations, and corresponding mask values,
<br/>
597 are still defined in libio and just included/linked. (We will need to
<br/>
598 implement these tables independently, soon, but should take advantage
<br/>
599 of libio where possible.) The num_put
<>::put members for integer types
<br/>
600 are
"mostly complete".
<br/>
602 A complete list of what has and has not been implemented may be
<br/>
603 found in CHECKLIST. However, note that the current definition of
<br/>
604 codecvt
<wchar_t,char,mbstate_t
> is wrong. It should simply write
<br/>
605 out the raw bytes representing the wide characters, rather than
<br/>
606 trying to convert each to a corresponding single
"char" value.
<br/>
608 Some of the facets are more important than others. Specifically,
<br/>
609 the members of ctype
<>, numpunct
<>, num_put
<>, and num_get
<> facets
<br/>
610 are used by other library facilities defined in
<string
>,
<istream
>,
<br/>
611 and
<ostream
>, and the codecvt
<> facet is used by basic_filebuf
<><br/>
612 in
<fstream
>, so a conforming iostream implementation depends on
<br/>
615 The
"long long" type eventually must be supported, but code mentioning
<br/>
616 it should be wrapped in #if guards to allow pedantic-mode compiling.
<br/>
618 Performance of num_put
<> and num_get
<> depend critically on
<br/>
619 caching computed values in ios_base objects, and on extensions
<br/>
620 to the interface with streambufs.
<br/>
622 Specifically: retrieving a copy of the locale object, extracting
<br/>
623 the needed facets, and gathering data from them, for each call to
<br/>
624 (e.g.) operator
<< would be prohibitively slow. To cache format
<br/>
625 data for use by num_put
<> and num_get
<> we have a _Format_cache
<><br/>
626 object stored in the ios_base::pword() array. This is constructed
<br/>
627 and initialized lazily, and is organized purely for utility. It
<br/>
628 is discarded when a new locale with different facets is imbued.
<br/>
630 Using only the public interfaces of the iterator arguments to the
<br/>
631 facet functions would limit performance by forbidding
"vector-style"<br/>
632 character operations. The streambuf iterator optimizations are
<br/>
633 described under chapter
24, but facets can also bypass the streambuf
<br/>
634 iterators via explicit specializations and operate directly on the
<br/>
635 streambufs, and use extended interfaces to get direct access to the
<br/>
636 streambuf internal buffer arrays. These extensions are mentioned
<br/>
637 under chapter
27. These optimizations are particularly important
<br/>
638 for input parsing.
<br/>
640 Unused virtual members of locale facets can be omitted, as mentioned
<br/>
641 above, by a smart linker.
<br/>
643 Chapter
23 Containers
<br/>
644 ----------------------
<br/>
645 Headers:
<deque
> <list
> <queue
> <stack
> <vector
> <map
> <set
> <bitset
><br/>
647 All the components in chapter
23 are implemented in the SGI STL.
<br/>
648 They are
"mostly complete"; they include a large number of
<br/>
649 nonconforming extensions which must be wrapped. Some of these
<br/>
650 are used internally and must be renamed or duplicated.
<br/>
652 The SGI components are optimized for large-memory environments. For
<br/>
653 embedded targets, different criteria might be more appropriate. Users
<br/>
654 will want to be able to tune this behavior. We should provide
<br/>
655 ways for users to compile the library with different memory usage
<br/>
656 characteristics.
<br/>
658 A lot more work is needed on factoring out common code from different
<br/>
659 specializations to reduce code size here and in chapter
25. The
<br/>
660 easiest fix for this would be a compiler/ABI improvement that allows
<br/>
661 the compiler to recognize when a specialization depends only on the
<br/>
662 size (or other gross quality) of a template argument, and allow the
<br/>
663 linker to share the code with similar specializations. In its
<br/>
664 absence, many of the algorithms and containers can be partial-
<br/>
665 specialized, at least for the case of pointers, but this only solves
<br/>
666 a small part of the problem. Use of a type_traits-style template
<br/>
667 allows a few more optimization opportunities, more if the compiler
<br/>
668 can generate the specializations automatically.
<br/>
670 As an optimization, containers can specialize on the default allocator
<br/>
671 and bypass it, or take advantage of details of its implementation
<br/>
672 after it has been improved upon.
<br/>
674 Replacing the vector iterators, which currently are simple element
<br/>
675 pointers, with class objects would greatly increase the safety of the
<br/>
676 client interface, and also permit a
"debug" mode in which range,
<br/>
677 ownership, and validity are rigorously checked. The current use of
<br/>
678 pointers for iterators is evil.
<br/>
680 As mentioned for chapter
24, the deque iterator is a good example of
<br/>
681 an opportunity to implement a
"staged" iterator that would benefit
<br/>
682 from specializations of some algorithms.
<br/>
684 Chapter
24 Iterators
<br/>
685 ---------------------
<br/>
686 Headers:
<iterator
><br/>
688 Standard iterators are
"mostly complete", with the exception of
<br/>
689 the stream iterators, which are not yet templatized on the
<br/>
690 stream type. Also, the base class template iterator
<> appears
<br/>
691 to be wrong, so everything derived from it must also be wrong,
<br/>
694 The streambuf iterators (currently located in stl/bits/std_iterator.h,
<br/>
695 but should be under bits/) can be rewritten to take advantage of
<br/>
696 friendship with the streambuf implementation.
<br/>
698 Matt Austern has identified opportunities where certain iterator
<br/>
699 types, particularly including streambuf iterators and deque
<br/>
700 iterators, have a
"two-stage" quality, such that an intermediate
<br/>
701 limit can be checked much more quickly than the true limit on
<br/>
702 range operations. If identified with a member of iterator_traits,
<br/>
703 algorithms may be specialized for this case. Of course the
<br/>
704 iterators that have this quality can be identified by specializing
<br/>
707 Many of the algorithms must be specialized for the streambuf
<br/>
708 iterators, to take advantage of block-mode operations, in order
<br/>
709 to allow iostream/locale operations' performance not to suffer.
<br/>
710 It may be that they could be treated as staged iterators and
<br/>
711 take advantage of those optimizations.
<br/>
713 Chapter
25 Algorithms
<br/>
714 ----------------------
<br/>
715 Headers:
<algorithm
><br/>
716 C headers:
<cstdlib
> (also in
18,
21,
26))
<br/>
718 The algorithms are
"mostly complete". As mentioned above, they
<br/>
719 are optimized for speed at the expense of code and data size.
<br/>
721 Specializations of many of the algorithms for non-STL types would
<br/>
722 give performance improvements, but we must use great care not to
<br/>
723 interfere with fragile template overloading semantics for the
<br/>
724 standard interfaces. Conventionally the standard function template
<br/>
725 interface is an inline which delegates to a non-standard function
<br/>
726 which is then overloaded (this is already done in many places in
<br/>
727 the library). Particularly appealing opportunities for the sake of
<br/>
728 iostream performance are for copy and find applied to streambuf
<br/>
729 iterators or (as noted elsewhere) for staged iterators, of which
<br/>
730 the streambuf iterators are a good example.
<br/>
732 The bsearch and qsort functions cannot be overloaded properly as
<br/>
733 required by the standard because gcc does not yet allow overloading
<br/>
734 on the extern-
"C"-ness of a function pointer.
<br/>
736 Chapter
26 Numerics
<br/>
737 --------------------
<br/>
738 Headers:
<complex
> <valarray
> <numeric
><br/>
739 C headers:
<cmath
>,
<cstdlib
> (also
18,
21,
25)
<br/>
741 Numeric components: Gabriel dos Reis's valarray, Drepper's complex,
<br/>
742 and the few algorithms from the STL are
"mostly done". Of course
<br/>
743 optimization opportunities abound for the numerically literate. It
<br/>
744 is not clear whether the valarray implementation really conforms
<br/>
745 fully, in the assumptions it makes about aliasing (and lack thereof)
<br/>
746 in its arguments.
<br/>
748 The C div() and ldiv() functions are interesting, because they are the
<br/>
749 only case where a C library function returns a class object by value.
<br/>
750 Since the C++ type div_t must be different from the underlying C type
<br/>
751 (which is in the wrong namespace) the underlying functions div() and
<br/>
752 ldiv() cannot be re-used efficiently. Fortunately they are trivial to
<br/>
755 Chapter
27 Iostreams
<br/>
756 ---------------------
<br/>
757 Headers:
<iosfwd
> <streambuf
> <ios
> <ostream
> <istream
> <iostream
><br/>
758 <iomanip
> <sstream
> <fstream
><br/>
759 C headers:
<cstdio
> <cwchar
> (also in
21)
<br/>
761 Iostream is currently in a very incomplete state.
<iosfwd
>,
<iomanip
>,
<br/>
762 ios_base, and basic_ios
<> are
"mostly complete". basic_streambuf
<> and
<br/>
763 basic_ostream
<> are well along, but basic_istream
<> has had little work
<br/>
764 done. The standard stream objects,
<sstream
> and
<fstream
> have been
<br/>
765 started; basic_filebuf
<> "write" functions have been implemented just
<br/>
766 enough to do
"hello, world".
<br/>
768 Most of the istream and ostream operators
<< and
>> (with the exception
<br/>
769 of the op
<<(integer) ones) have not been changed to use locale primitives,
<br/>
770 sentry objects, or char_traits members.
<br/>
772 All these templates should be manually instantiated for char and
<br/>
773 wchar_t in a way that links only used members into user programs.
<br/>
775 Streambuf is fertile ground for optimization extensions. An extended
<br/>
776 interface giving iterator access to its internal buffer would be very
<br/>
777 useful for other library components.
<br/>
779 Iostream operations (primarily operators
<< and
>>) can take advantage
<br/>
780 of the case where user code has not specified a locale, and bypass locale
<br/>
781 operations entirely. The current implementation of op
<</num_put
<>::put,
<br/>
782 for the integer types, demonstrates how they can cache encoding details
<br/>
783 from the locale on each operation. There is lots more room for
<br/>
784 optimization in this area.
<br/>
786 The definition of the relationship between the standard streams
<br/>
787 cout et al. and stdout et al. requires something like a
"stdiobuf".
<br/>
788 The SGI solution of using double-indirection to actually use a
<br/>
789 stdio FILE object for buffering is unsatisfactory, because it
<br/>
790 interferes with peephole loop optimizations.
<br/>
792 The
<sstream
> header work has begun. stringbuf can benefit from
<br/>
793 friendship with basic_string
<> and basic_string
<>::_Rep to use
<br/>
794 those objects directly as buffers, and avoid allocating and making
<br/>
797 The basic_filebuf
<> template is a complex beast. It is specified to
<br/>
798 use the locale facet codecvt
<> to translate characters between native
<br/>
799 files and the locale character encoding. In general this involves
<br/>
800 two buffers, one of
"char" representing the file and another of
<br/>
801 "char_type", for the stream, with codecvt
<> translating. The process
<br/>
802 is complicated by the variable-length nature of the translation, and
<br/>
803 the need to seek to corresponding places in the two representations.
<br/>
804 For the case of basic_filebuf
<char
>, when no translation is needed,
<br/>
805 a single buffer suffices. A specialized filebuf can be used to reduce
<br/>
806 code space overhead when no locale has been imbued. Matt Austern's
<br/>
807 work at SGI will be useful, perhaps directly as a source of code, or
<br/>
808 at least as an example to draw on.
<br/>
810 Filebuf, almost uniquely (cf. operator new), depends heavily on
<br/>
811 underlying environmental facilities. In current releases iostream
<br/>
812 depends fairly heavily on libio constant definitions, but it should
<br/>
813 be made independent. It also depends on operating system primitives
<br/>
814 for file operations. There is immense room for optimizations using
<br/>
815 (e.g.) mmap for reading. The shadow/ directory wraps, besides the
<br/>
816 standard C headers, the libio.h and unistd.h headers, for use mainly
<br/>
817 by filebuf. These wrappings have not been completed, though there
<br/>
818 is scaffolding in place.
<br/>
820 The encapsulation of certain C header
<cstdio
> names presents an
<br/>
821 interesting problem. It is possible to define an inline std::fprintf()
<br/>
822 implemented in terms of the 'extern
"C"' vfprintf(), but there is no
<br/>
823 standard vfscanf() to use to implement std::fscanf(). It appears that
<br/>
824 vfscanf but be re-implemented in C++ for targets where no vfscanf
<br/>
825 extension has been defined. This is interesting in that it seems
<br/>
826 to be the only significant case in the C library where this kind of
<br/>
827 rewriting is necessary. (Of course Glibc provides the vfscanf()
<br/>
828 extension.) (The functions related to exit() must be rewritten
<br/>
829 for other reasons.)
<br/>
834 Headers:
<strstream
><br/>
836 Annex D defines many non-library features, and many minor
<br/>
837 modifications to various headers, and a complete header.
<br/>
838 It is
"mostly done", except that the libstdc++-
2 <strstream
><br/>
839 header has not been adopted into the library, or checked to
<br/>
840 verify that it matches the draft in those details that were
<br/>
841 clarified by the committee. Certainly it must at least be
<br/>
842 moved into the std namespace.
<br/>
844 We still need to wrap all the deprecated features in #if guards
<br/>
845 so that pedantic compile modes can detect their use.
<br/>
847 Nonstandard Extensions
<br/>
848 ----------------------
<br/>
849 Headers:
<iostream.h
> <strstream.h
> <hash
> <rbtree
><br/>
850 <pthread_alloc
> <stdiobuf
> (etc.)
<br/>
852 User code has come to depend on a variety of nonstandard components
<br/>
853 that we must not omit. Much of this code can be adopted from
<br/>
854 libstdc++-v2 or from the SGI STL. This particularly includes
<br/>
855 <iostream.h
>,
<strstream.h
>, and various SGI extensions such
<br/>
856 as
<hash_map.h
>. Many of these are already placed in the
<br/>
857 subdirectories ext/ and backward/. (Note that it is better to
<br/>
858 include them via
"<backward/hash_map.h>" or
"<ext/hash_map>" than
<br/>
859 to search the subdirectory itself via a
"-I" directive.
<br/>
860 </p></div></div><div class=
"navfooter"><hr/><table width=
"100%" summary=
"Navigation footer"><tr><td align=
"left"><a accesskey=
"p" href=
"source_code_style.html">Prev
</a> </td><td align=
"center"><a accesskey=
"u" href=
"appendix_contributing.html">Up
</a></td><td align=
"right"> <a accesskey=
"n" href=
"appendix_porting.html">Next
</a></td></tr><tr><td align=
"left" valign=
"top">Coding Style
</td><td align=
"center"><a accesskey=
"h" href=
"../spine.html">Home
</a></td><td align=
"right" valign=
"top"> Appendix B.
861 Porting and Maintenance
863 </td></tr></table></div></body></html>