URL
https://opencores.org/ocsvn/openrisc/openrisc/trunk
Subversion Repositories openrisc
[/] [openrisc/] [trunk/] [gnu-dev/] [or1k-gcc/] [libstdc++-v3/] [doc/] [html/] [manual/] [source_design_notes.html] - Rev 749
Go to most recent revision | Compare with Previous | Blame | View Log
<?xml version="1.0" encoding="UTF-8" standalone="no"?> <!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.1//EN" "http://www.w3.org/TR/xhtml11/DTD/xhtml11.dtd"> <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 "/><meta name="keywords" content=" ISO C++ , runtime , library "/><link rel="home" href="../index.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. Contributing </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> </p><div class="literallayout"><p><br/> <br/> The Library<br/> -----------<br/> <br/> This paper is covers two major areas:<br/> <br/> - Features and policies not mentioned in the standard that<br/> the quality of the library implementation depends on, including<br/> extensions and "implementation-defined" features;<br/> <br/> - Plans for required but unimplemented library features and<br/> optimizations to them.<br/> <br/> Overhead<br/> --------<br/> <br/> The standard defines a large library, much larger than the standard<br/> C library. A naive implementation would suffer substantial overhead<br/> in compile time, executable size, and speed, rendering it unusable<br/> in many (particularly embedded) applications. The alternative demands<br/> care in construction, and some compiler support, but there is no<br/> need for library subsets.<br/> <br/> What are the sources of this overhead? There are four main causes:<br/> <br/> - The library is specified almost entirely as templates, which<br/> with current compilers must be included in-line, resulting in<br/> very slow builds as tens or hundreds of thousands of lines<br/> of function definitions are read for each user source file.<br/> Indeed, the entire SGI STL, as well as the dos Reis valarray,<br/> are provided purely as header files, largely for simplicity in<br/> porting. Iostream/locale is (or will be) as large again.<br/> <br/> - The library is very flexible, specifying a multitude of hooks<br/> where users can insert their own code in place of defaults.<br/> When these hooks are not used, any time and code expended to<br/> support that flexibility is wasted.<br/> <br/> - Templates are often described as causing to "code bloat". In<br/> practice, this refers (when it refers to anything real) to several<br/> independent processes. First, when a class template is manually<br/> instantiated in its entirely, current compilers place the definitions<br/> for all members in a single object file, so that a program linking<br/> to one member gets definitions of all. Second, template functions<br/> which do not actually depend on the template argument are, under<br/> current compilers, generated anew for each instantiation, rather<br/> than being shared with other instantiations. Third, some of the<br/> flexibility mentioned above comes from virtual functions (both in<br/> regular classes and template classes) which current linkers add<br/> to the executable file even when they manifestly cannot be called.<br/> <br/> - The library is specified to use a language feature, exceptions,<br/> which in the current gcc compiler ABI imposes a run time and<br/> code space cost to handle the possibility of exceptions even when<br/> they are not used. Under the new ABI (accessed with -fnew-abi),<br/> there is a space overhead and a small reduction in code efficiency<br/> resulting from lost optimization opportunities associated with<br/> non-local branches associated with exceptions.<br/> <br/> What can be done to eliminate this overhead? A variety of coding<br/> techniques, and compiler, linker and library improvements and<br/> extensions may be used, as covered below. Most are not difficult,<br/> and some are already implemented in varying degrees.<br/> <br/> Overhead: Compilation Time<br/> --------------------------<br/> <br/> Providing "ready-instantiated" template code in object code archives<br/> allows us to avoid generating and optimizing template instantiations<br/> in each compilation unit which uses them. However, the number of such<br/> instantiations that are useful to provide is limited, and anyway this<br/> is not enough, by itself, to minimize compilation time. In particular,<br/> it does not reduce time spent parsing conforming headers.<br/> <br/> Quicker header parsing will depend on library extensions and compiler<br/> improvements. One approach is some variation on the techniques<br/> previously marketed as "pre-compiled headers", now standardized as<br/> support for the "export" keyword. "Exported" template definitions<br/> can be placed (once) in a "repository" -- really just a library, but<br/> of template definitions rather than object code -- to be drawn upon<br/> at link time when an instantiation is needed, rather than placed in<br/> header files to be parsed along with every compilation unit.<br/> <br/> Until "export" is implemented we can put some of the lengthy template<br/> definitions in #if guards or alternative headers so that users can skip<br/> over the full definitions when they need only the ready-instantiated<br/> specializations.<br/> <br/> To be precise, this means that certain headers which define<br/> templates which users normally use only for certain arguments<br/> can be instrumented to avoid exposing the template definitions<br/> to the compiler unless a macro is defined. For example, in<br/> <string>, we might have:<br/> <br/> template <class _CharT, ... > class basic_string {<br/> ... // member declarations<br/> };<br/> ... // operator declarations<br/> <br/> #ifdef _STRICT_ISO_<br/> # if _G_NO_TEMPLATE_EXPORT<br/> # include <bits/std_locale.h> // headers needed by definitions<br/> # ...<br/> # include <bits/string.tcc> // member and global template definitions.<br/> # endif<br/> #endif<br/> <br/> Users who compile without specifying a strict-ISO-conforming flag<br/> would not see many of the template definitions they now see, and rely<br/> instead on ready-instantiated specializations in the library. This<br/> technique would be useful for the following substantial components:<br/> string, locale/iostreams, valarray. It would *not* be useful or<br/> usable with the following: containers, algorithms, iterators,<br/> allocator. Since these constitute a large (though decreasing)<br/> fraction of the library, the benefit the technique offers is<br/> limited.<br/> <br/> The language specifies the semantics of the "export" keyword, but<br/> the gcc compiler does not yet support it. When it does, problems<br/> with large template inclusions can largely disappear, given some<br/> minor library reorganization, along with the need for the apparatus<br/> described above.<br/> <br/> Overhead: Flexibility Cost<br/> --------------------------<br/> <br/> The library offers many places where users can specify operations<br/> to be performed by the library in place of defaults. Sometimes<br/> this seems to require that the library use a more-roundabout, and<br/> possibly slower, way to accomplish the default requirements than<br/> would be used otherwise.<br/> <br/> The primary protection against this overhead is thorough compiler<br/> optimization, to crush out layers of inline function interfaces.<br/> Kuck & Associates has demonstrated the practicality of this kind<br/> of optimization.<br/> <br/> The second line of defense against this overhead is explicit<br/> specialization. By defining helper function templates, and writing<br/> specialized code for the default case, overhead can be eliminated<br/> for that case without sacrificing flexibility. This takes full<br/> advantage of any ability of the optimizer to crush out degenerate<br/> code.<br/> <br/> The library specifies many virtual functions which current linkers<br/> load even when they cannot be called. Some minor improvements to the<br/> compiler and to ld would eliminate any such overhead by simply<br/> omitting virtual functions that the complete program does not call.<br/> A prototype of this work has already been done. For targets where<br/> GNU ld is not used, a "pre-linker" could do the same job.<br/> <br/> The main areas in the standard interface where user flexibility<br/> can result in overhead are:<br/> <br/> - Allocators: Containers are specified to use user-definable<br/> allocator types and objects, making tuning for the container<br/> characteristics tricky.<br/> <br/> - Locales: the standard specifies locale objects used to implement<br/> iostream operations, involving many virtual functions which use<br/> streambuf iterators.<br/> <br/> - Algorithms and containers: these may be instantiated on any type,<br/> frequently duplicating code for identical operations.<br/> <br/> - Iostreams and strings: users are permitted to use these on their<br/> own types, and specify the operations the stream must use on these<br/> types.<br/> <br/> Note that these sources of overhead are _avoidable_. The techniques<br/> to avoid them are covered below.<br/> <br/> Code Bloat<br/> ----------<br/> <br/> In the SGI STL, and in some other headers, many of the templates<br/> are defined "inline" -- either explicitly or by their placement<br/> in class definitions -- which should not be inline. This is a<br/> source of code bloat. Matt had remarked that he was relying on<br/> the compiler to recognize what was too big to benefit from inlining,<br/> and generate it out-of-line automatically. However, this also can<br/> result in code bloat except where the linker can eliminate the extra<br/> copies.<br/> <br/> Fixing these cases will require an audit of all inline functions<br/> defined in the library to determine which merit inlining, and moving<br/> the rest out of line. This is an issue mainly in chapters 23, 25, and<br/> 27. Of course it can be done incrementally, and we should generally<br/> accept patches that move large functions out of line and into ".tcc"<br/> files, which can later be pulled into a repository. Compiler/linker<br/> improvements to recognize very large inline functions and move them<br/> out-of-line, but shared among compilation units, could make this<br/> work unnecessary.<br/> <br/> Pre-instantiating template specializations currently produces large<br/> amounts of dead code which bloats statically linked programs. The<br/> current state of the static library, libstdc++.a, is intolerable on<br/> this account, and will fuel further confused speculation about a need<br/> for a library "subset". A compiler improvement that treats each<br/> instantiated function as a separate object file, for linking purposes,<br/> would be one solution to this problem. An alternative would be to<br/> split up the manual instantiation files into dozens upon dozens of<br/> little files, each compiled separately, but an abortive attempt at<br/> this was done for <string> and, though it is far from complete, it<br/> is already a nuisance. A better interim solution (just until we have<br/> "export") is badly needed.<br/> <br/> When building a shared library, the current compiler/linker cannot<br/> automatically generate the instantiations needed. This creates a<br/> miserable situation; it means any time something is changed in the<br/> library, before a shared library can be built someone must manually<br/> copy the declarations of all templates that are needed by other parts<br/> of the library to an "instantiation" file, and add it to the build<br/> system to be compiled and linked to the library. This process is<br/> readily automated, and should be automated as soon as possible.<br/> Users building their own shared libraries experience identical<br/> frustrations.<br/> <br/> Sharing common aspects of template definitions among instantiations<br/> can radically reduce code bloat. The compiler could help a great<br/> deal here by recognizing when a function depends on nothing about<br/> a template parameter, or only on its size, and giving the resulting<br/> function a link-name "equate" that allows it to be shared with other<br/> instantiations. Implementation code could take advantage of the<br/> capability by factoring out code that does not depend on the template<br/> argument into separate functions to be merged by the compiler.<br/> <br/> Until such a compiler optimization is implemented, much can be done<br/> manually (if tediously) in this direction. One such optimization is<br/> to derive class templates from non-template classes, and move as much<br/> implementation as possible into the base class. Another is to partial-<br/> specialize certain common instantiations, such as vector<T*>, to share<br/> code for instantiations on all types T. While these techniques work,<br/> they are far from the complete solution that a compiler improvement<br/> would afford.<br/> <br/> Overhead: Expensive Language Features<br/> -------------------------------------<br/> <br/> The main "expensive" language feature used in the standard library<br/> is exception support, which requires compiling in cleanup code with<br/> static table data to locate it, and linking in library code to use<br/> the table. For small embedded programs the amount of such library<br/> code and table data is assumed by some to be excessive. Under the<br/> "new" ABI this perception is generally exaggerated, although in some<br/> cases it may actually be excessive.<br/> <br/> To implement a library which does not use exceptions directly is<br/> not difficult given minor compiler support (to "turn off" exceptions<br/> and ignore exception constructs), and results in no great library<br/> maintenance difficulties. To be precise, given "-fno-exceptions",<br/> the compiler should treat "try" blocks as ordinary blocks, and<br/> "catch" blocks as dead code to ignore or eliminate. Compiler<br/> support is not strictly necessary, except in the case of "function<br/> try blocks"; otherwise the following macros almost suffice:<br/> <br/> #define throw(X)<br/> #define try if (true)<br/> #define catch(X) else if (false)<br/> <br/> However, there may be a need to use function try blocks in the<br/> library implementation, and use of macros in this way can make<br/> correct diagnostics impossible. Furthermore, use of this scheme<br/> would require the library to call a function to re-throw exceptions<br/> from a try block. Implementing the above semantics in the compiler<br/> is preferable.<br/> <br/> Given the support above (however implemented) it only remains to<br/> replace code that "throws" with a call to a well-documented "handler"<br/> function in a separate compilation unit which may be replaced by<br/> the user. The main source of exceptions that would be difficult<br/> for users to avoid is memory allocation failures, but users can<br/> define their own memory allocation primitives that never throw.<br/> Otherwise, the complete list of such handlers, and which library<br/> functions may call them, would be needed for users to be able to<br/> implement the necessary substitutes. (Fortunately, they have the<br/> source code.)<br/> <br/> Opportunities<br/> -------------<br/> <br/> The template capabilities of C++ offer enormous opportunities for<br/> optimizing common library operations, well beyond what would be<br/> considered "eliminating overhead". In particular, many operations<br/> done in Glibc with macros that depend on proprietary language<br/> extensions can be implemented in pristine Standard C++. For example,<br/> the chapter 25 algorithms, and even C library functions such as strchr,<br/> can be specialized for the case of static arrays of known (small) size.<br/> <br/> Detailed optimization opportunities are identified below where<br/> the component where they would appear is discussed. Of course new<br/> opportunities will be identified during implementation.<br/> <br/> Unimplemented Required Library Features<br/> ---------------------------------------<br/> <br/> The standard specifies hundreds of components, grouped broadly by<br/> chapter. These are listed in excruciating detail in the CHECKLIST<br/> file.<br/> <br/> 17 general<br/> 18 support<br/> 19 diagnostics<br/> 20 utilities<br/> 21 string<br/> 22 locale<br/> 23 containers<br/> 24 iterators<br/> 25 algorithms<br/> 26 numerics<br/> 27 iostreams<br/> Annex D backward compatibility<br/> <br/> Anyone participating in implementation of the library should obtain<br/> a copy of the standard, ISO 14882. People in the U.S. can obtain an<br/> electronic copy for US$18 from ANSI's web site. Those from other<br/> countries should visit http://www.iso.org/ to find out the location<br/> of their country's representation in ISO, in order to know who can<br/> sell them a copy.<br/> <br/> The emphasis in the following sections is on unimplemented features<br/> and optimization opportunities.<br/> <br/> Chapter 17 General<br/> -------------------<br/> <br/> Chapter 17 concerns overall library requirements.<br/> <br/> The standard doesn't mention threads. A multi-thread (MT) extension<br/> primarily affects operators new and delete (18), allocator (20),<br/> string (21), locale (22), and iostreams (27). The common underlying<br/> support needed for this is discussed under chapter 20.<br/> <br/> The standard requirements on names from the C headers create a<br/> lot of work, mostly done. Names in the C headers must be visible<br/> in the std:: and sometimes the global namespace; the names in the<br/> two scopes must refer to the same object. More stringent is that<br/> Koenig lookup implies that any types specified as defined in std::<br/> really are defined in std::. Names optionally implemented as<br/> macros in C cannot be macros in C++. (An overview may be read at<br/> <http://www.cantrip.org/cheaders.html>). The scripts "inclosure"<br/> and "mkcshadow", and the directories shadow/ and cshadow/, are the<br/> beginning of an effort to conform in this area.<br/> <br/> A correct conforming definition of C header names based on underlying<br/> C library headers, and practical linking of conforming namespaced<br/> customer code with third-party C libraries depends ultimately on<br/> an ABI change, allowing namespaced C type names to be mangled into<br/> type names as if they were global, somewhat as C function names in a<br/> namespace, or C++ global variable names, are left unmangled. Perhaps<br/> another "extern" mode, such as 'extern "C-global"' would be an<br/> appropriate place for such type definitions. Such a type would<br/> affect mangling as follows:<br/> <br/> namespace A {<br/> struct X {};<br/> extern "C-global" { // or maybe just 'extern "C"'<br/> struct Y {};<br/> };<br/> }<br/> void f(A::X*); // mangles to f__FPQ21A1X<br/> void f(A::Y*); // mangles to f__FP1Y<br/> <br/> (It may be that this is really the appropriate semantics for regular<br/> 'extern "C"', and 'extern "C-global"', as an extension, would not be<br/> necessary.) This would allow functions declared in non-standard C headers<br/> (and thus fixable by neither us nor users) to link properly with functions<br/> declared using C types defined in properly-namespaced headers. The<br/> problem this solves is that C headers (which C++ programmers do persist<br/> in using) frequently forward-declare C struct tags without including<br/> the header where the type is defined, as in<br/> <br/> struct tm;<br/> void munge(tm*);<br/> <br/> Without some compiler accommodation, munge cannot be called by correct<br/> C++ code using a pointer to a correctly-scoped tm* value.<br/> <br/> The current C headers use the preprocessor extension "#include_next",<br/> which the compiler complains about when run "-pedantic".<br/> (Incidentally, it appears that "-fpedantic" is currently ignored,<br/> probably a bug.) The solution in the C compiler is to use<br/> "-isystem" rather than "-I", but unfortunately in g++ this seems<br/> also to wrap the whole header in an 'extern "C"' block, so it's<br/> unusable for C++ headers. The correct solution appears to be to<br/> allow the various special include-directory options, if not given<br/> an argument, to affect subsequent include-directory options additively,<br/> so that if one said<br/> <br/> -pedantic -iprefix $(prefix) \<br/> -idirafter -ino-pedantic -ino-extern-c -iwithprefix -I g++-v3 \<br/> -iwithprefix -I g++-v3/ext<br/> <br/> the compiler would search $(prefix)/g++-v3 and not report<br/> pedantic warnings for files found there, but treat files in<br/> $(prefix)/g++-v3/ext pedantically. (The undocumented semantics<br/> of "-isystem" in g++ stink. Can they be rescinded? If not it<br/> must be replaced with something more rationally behaved.)<br/> <br/> All the C headers need the treatment above; in the standard these<br/> headers are mentioned in various chapters. Below, I have only<br/> mentioned those that present interesting implementation issues.<br/> <br/> The components identified as "mostly complete", below, have not been<br/> audited for conformance. In many cases where the library passes<br/> conformance tests we have non-conforming extensions that must be<br/> wrapped in #if guards for "pedantic" use, and in some cases renamed<br/> in a conforming way for continued use in the implementation regardless<br/> of conformance flags.<br/> <br/> The STL portion of the library still depends on a header<br/> stl/bits/stl_config.h full of #ifdef clauses. This apparatus<br/> should be replaced with autoconf/automake machinery.<br/> <br/> The SGI STL defines a type_traits<> template, specialized for<br/> many types in their code including the built-in numeric and<br/> pointer types and some library types, to direct optimizations of<br/> standard functions. The SGI compiler has been extended to generate<br/> specializations of this template automatically for user types,<br/> so that use of STL templates on user types can take advantage of<br/> these optimizations. Specializations for other, non-STL, types<br/> would make more optimizations possible, but extending the gcc<br/> compiler in the same way would be much better. Probably the next<br/> round of standardization will ratify this, but probably with<br/> changes, so it probably should be renamed to place it in the<br/> implementation namespace.<br/> <br/> The SGI STL also defines a large number of extensions visible in<br/> standard headers. (Other extensions that appear in separate headers<br/> have been sequestered in subdirectories ext/ and backward/.) All<br/> these extensions should be moved to other headers where possible,<br/> and in any case wrapped in a namespace (not std!), and (where kept<br/> in a standard header) girded about with macro guards. Some cannot be<br/> moved out of standard headers because they are used to implement<br/> standard features. The canonical method for accommodating these<br/> is to use a protected name, aliased in macro guards to a user-space<br/> name. Unfortunately C++ offers no satisfactory template typedef<br/> mechanism, so very ad-hoc and unsatisfactory aliasing must be used<br/> instead.<br/> <br/> Implementation of a template typedef mechanism should have the highest<br/> priority among possible extensions, on the same level as implementation<br/> of the template "export" feature.<br/> <br/> Chapter 18 Language support<br/> ----------------------------<br/> <br/> Headers: <limits> <new> <typeinfo> <exception><br/> C headers: <cstddef> <climits> <cfloat> <cstdarg> <csetjmp><br/> <ctime> <csignal> <cstdlib> (also 21, 25, 26)<br/> <br/> This defines the built-in exceptions, rtti, numeric_limits<>,<br/> operator new and delete. Much of this is provided by the<br/> compiler in its static runtime library.<br/> <br/> Work to do includes defining numeric_limits<> specializations in<br/> separate files for all target architectures. Values for integer types<br/> except for bool and wchar_t are readily obtained from the C header<br/> <limits.h>, but values for the remaining numeric types (bool, wchar_t,<br/> float, double, long double) must be entered manually. This is<br/> largely dog work except for those members whose values are not<br/> easily deduced from available documentation. Also, this involves<br/> some work in target configuration to identify the correct choice of<br/> file to build against and to install.<br/> <br/> The definitions of the various operators new and delete must be<br/> made thread-safe, which depends on a portable exclusion mechanism,<br/> discussed under chapter 20. Of course there is always plenty of<br/> room for improvements to the speed of operators new and delete.<br/> <br/> <cstdarg>, in Glibc, defines some macros that gcc does not allow to<br/> be wrapped into an inline function. Probably this header will demand<br/> attention whenever a new target is chosen. The functions atexit(),<br/> exit(), and abort() in cstdlib have different semantics in C++, so<br/> must be re-implemented for C++.<br/> <br/> Chapter 19 Diagnostics<br/> -----------------------<br/> <br/> Headers: <stdexcept><br/> C headers: <cassert> <cerrno><br/> <br/> This defines the standard exception objects, which are "mostly complete".<br/> Cygnus has a version, and now SGI provides a slightly different one.<br/> It makes little difference which we use.<br/> <br/> The C global name "errno", which C allows to be a variable or a macro,<br/> is required in C++ to be a macro. For MT it must typically result in<br/> a function call.<br/> <br/> Chapter 20 Utilities<br/> ---------------------<br/> Headers: <utility> <functional> <memory><br/> C header: <ctime> (also in 18)<br/> <br/> SGI STL provides "mostly complete" versions of all the components<br/> defined in this chapter. However, the auto_ptr<> implementation<br/> is known to be wrong. Furthermore, the standard definition of it<br/> is known to be unimplementable as written. A minor change to the<br/> standard would fix it, and auto_ptr<> should be adjusted to match.<br/> <br/> Multi-threading affects the allocator implementation, and there must<br/> be configuration/installation choices for different users' MT<br/> requirements. Anyway, users will want to tune allocator options<br/> to support different target conditions, MT or no.<br/> <br/> The primitives used for MT implementation should be exposed, as an<br/> extension, for users' own work. We need cross-CPU "mutex" support,<br/> multi-processor shared-memory atomic integer operations, and single-<br/> processor uninterruptible integer operations, and all three configurable<br/> to be stubbed out for non-MT use, or to use an appropriately-loaded<br/> dynamic library for the actual runtime environment, or statically<br/> compiled in for cases where the target architecture is known.<br/> <br/> Chapter 21 String<br/> ------------------<br/> Headers: <string><br/> C headers: <cctype> <cwctype> <cstring> <cwchar> (also in 27)<br/> <cstdlib> (also in 18, 25, 26)<br/> <br/> We have "mostly-complete" char_traits<> implementations. Many of the<br/> char_traits<char> operations might be optimized further using existing<br/> proprietary language extensions.<br/> <br/> We have a "mostly-complete" basic_string<> implementation. The work<br/> to manually instantiate char and wchar_t specializations in object<br/> files to improve link-time behavior is extremely unsatisfactory,<br/> literally tripling library-build time with no commensurate improvement<br/> in static program link sizes. It must be redone. (Similar work is<br/> needed for some components in chapters 22 and 27.)<br/> <br/> Other work needed for strings is MT-safety, as discussed under the<br/> chapter 20 heading.<br/> <br/> The standard C type mbstate_t from <cwchar> and used in char_traits<><br/> must be different in C++ than in C, because in C++ the default constructor<br/> value mbstate_t() must be the "base" or "ground" sequence state.<br/> (According to the likely resolution of a recently raised Core issue,<br/> this may become unnecessary. However, there are other reasons to<br/> use a state type not as limited as whatever the C library provides.)<br/> If we might want to provide conversions from (e.g.) internally-<br/> represented EUC-wide to externally-represented Unicode, or vice-<br/> versa, the mbstate_t we choose will need to be more accommodating<br/> than what might be provided by an underlying C library.<br/> <br/> There remain some basic_string template-member functions which do<br/> not overload properly with their non-template brethren. The infamous<br/> hack akin to what was done in vector<> is needed, to conform to<br/> 23.1.1 para 10. The CHECKLIST items for basic_string marked 'X',<br/> or incomplete, are so marked for this reason.<br/> <br/> Replacing the string iterators, which currently are simple character<br/> pointers, with class objects would greatly increase the safety of the<br/> client interface, and also permit a "debug" mode in which range,<br/> ownership, and validity are rigorously checked. The current use of<br/> raw pointers as string iterators is evil. vector<> iterators need the<br/> same treatment. Note that the current implementation freely mixes<br/> pointers and iterators, and that must be fixed before safer iterators<br/> can be introduced.<br/> <br/> Some of the functions in <cstring> are different from the C version.<br/> generally overloaded on const and non-const argument pointers. For<br/> example, in <cstring> strchr is overloaded. The functions isupper<br/> etc. in <cctype> typically implemented as macros in C are functions<br/> in C++, because they are overloaded with others of the same name<br/> defined in <locale>.<br/> <br/> Many of the functions required in <cwctype> and <cwchar> cannot be<br/> implemented using underlying C facilities on intended targets because<br/> such facilities only partly exist.<br/> <br/> Chapter 22 Locale<br/> ------------------<br/> Headers: <locale><br/> C headers: <clocale><br/> <br/> We have a "mostly complete" class locale, with the exception of<br/> code for constructing, and handling the names of, named locales.<br/> The ways that locales are named (particularly when categories<br/> (e.g. LC_TIME, LC_COLLATE) are different) varies among all target<br/> environments. This code must be written in various versions and<br/> chosen by configuration parameters.<br/> <br/> Members of many of the facets defined in <locale> are stubs. Generally,<br/> there are two sets of facets: the base class facets (which are supposed<br/> to implement the "C" locale) and the "byname" facets, which are supposed<br/> to read files to determine their behavior. The base ctype<>, collate<>,<br/> and numpunct<> facets are "mostly complete", except that the table of<br/> bitmask values used for "is" operations, and corresponding mask values,<br/> are still defined in libio and just included/linked. (We will need to<br/> implement these tables independently, soon, but should take advantage<br/> of libio where possible.) The num_put<>::put members for integer types<br/> are "mostly complete".<br/> <br/> A complete list of what has and has not been implemented may be<br/> found in CHECKLIST. However, note that the current definition of<br/> codecvt<wchar_t,char,mbstate_t> is wrong. It should simply write<br/> out the raw bytes representing the wide characters, rather than<br/> trying to convert each to a corresponding single "char" value.<br/> <br/> Some of the facets are more important than others. Specifically,<br/> the members of ctype<>, numpunct<>, num_put<>, and num_get<> facets<br/> are used by other library facilities defined in <string>, <istream>,<br/> and <ostream>, and the codecvt<> facet is used by basic_filebuf<><br/> in <fstream>, so a conforming iostream implementation depends on<br/> these.<br/> <br/> The "long long" type eventually must be supported, but code mentioning<br/> it should be wrapped in #if guards to allow pedantic-mode compiling.<br/> <br/> Performance of num_put<> and num_get<> depend critically on<br/> caching computed values in ios_base objects, and on extensions<br/> to the interface with streambufs.<br/> <br/> Specifically: retrieving a copy of the locale object, extracting<br/> the needed facets, and gathering data from them, for each call to<br/> (e.g.) operator<< would be prohibitively slow. To cache format<br/> data for use by num_put<> and num_get<> we have a _Format_cache<><br/> object stored in the ios_base::pword() array. This is constructed<br/> and initialized lazily, and is organized purely for utility. It<br/> is discarded when a new locale with different facets is imbued.<br/> <br/> Using only the public interfaces of the iterator arguments to the<br/> facet functions would limit performance by forbidding "vector-style"<br/> character operations. The streambuf iterator optimizations are<br/> described under chapter 24, but facets can also bypass the streambuf<br/> iterators via explicit specializations and operate directly on the<br/> streambufs, and use extended interfaces to get direct access to the<br/> streambuf internal buffer arrays. These extensions are mentioned<br/> under chapter 27. These optimizations are particularly important<br/> for input parsing.<br/> <br/> Unused virtual members of locale facets can be omitted, as mentioned<br/> above, by a smart linker.<br/> <br/> Chapter 23 Containers<br/> ----------------------<br/> Headers: <deque> <list> <queue> <stack> <vector> <map> <set> <bitset><br/> <br/> All the components in chapter 23 are implemented in the SGI STL.<br/> They are "mostly complete"; they include a large number of<br/> nonconforming extensions which must be wrapped. Some of these<br/> are used internally and must be renamed or duplicated.<br/> <br/> The SGI components are optimized for large-memory environments. For<br/> embedded targets, different criteria might be more appropriate. Users<br/> will want to be able to tune this behavior. We should provide<br/> ways for users to compile the library with different memory usage<br/> characteristics.<br/> <br/> A lot more work is needed on factoring out common code from different<br/> specializations to reduce code size here and in chapter 25. The<br/> easiest fix for this would be a compiler/ABI improvement that allows<br/> the compiler to recognize when a specialization depends only on the<br/> size (or other gross quality) of a template argument, and allow the<br/> linker to share the code with similar specializations. In its<br/> absence, many of the algorithms and containers can be partial-<br/> specialized, at least for the case of pointers, but this only solves<br/> a small part of the problem. Use of a type_traits-style template<br/> allows a few more optimization opportunities, more if the compiler<br/> can generate the specializations automatically.<br/> <br/> As an optimization, containers can specialize on the default allocator<br/> and bypass it, or take advantage of details of its implementation<br/> after it has been improved upon.<br/> <br/> Replacing the vector iterators, which currently are simple element<br/> pointers, with class objects would greatly increase the safety of the<br/> client interface, and also permit a "debug" mode in which range,<br/> ownership, and validity are rigorously checked. The current use of<br/> pointers for iterators is evil.<br/> <br/> As mentioned for chapter 24, the deque iterator is a good example of<br/> an opportunity to implement a "staged" iterator that would benefit<br/> from specializations of some algorithms.<br/> <br/> Chapter 24 Iterators<br/> ---------------------<br/> Headers: <iterator><br/> <br/> Standard iterators are "mostly complete", with the exception of<br/> the stream iterators, which are not yet templatized on the<br/> stream type. Also, the base class template iterator<> appears<br/> to be wrong, so everything derived from it must also be wrong,<br/> currently.<br/> <br/> The streambuf iterators (currently located in stl/bits/std_iterator.h,<br/> but should be under bits/) can be rewritten to take advantage of<br/> friendship with the streambuf implementation.<br/> <br/> Matt Austern has identified opportunities where certain iterator<br/> types, particularly including streambuf iterators and deque<br/> iterators, have a "two-stage" quality, such that an intermediate<br/> limit can be checked much more quickly than the true limit on<br/> range operations. If identified with a member of iterator_traits,<br/> algorithms may be specialized for this case. Of course the<br/> iterators that have this quality can be identified by specializing<br/> a traits class.<br/> <br/> Many of the algorithms must be specialized for the streambuf<br/> iterators, to take advantage of block-mode operations, in order<br/> to allow iostream/locale operations' performance not to suffer.<br/> It may be that they could be treated as staged iterators and<br/> take advantage of those optimizations.<br/> <br/> Chapter 25 Algorithms<br/> ----------------------<br/> Headers: <algorithm><br/> C headers: <cstdlib> (also in 18, 21, 26))<br/> <br/> The algorithms are "mostly complete". As mentioned above, they<br/> are optimized for speed at the expense of code and data size.<br/> <br/> Specializations of many of the algorithms for non-STL types would<br/> give performance improvements, but we must use great care not to<br/> interfere with fragile template overloading semantics for the<br/> standard interfaces. Conventionally the standard function template<br/> interface is an inline which delegates to a non-standard function<br/> which is then overloaded (this is already done in many places in<br/> the library). Particularly appealing opportunities for the sake of<br/> iostream performance are for copy and find applied to streambuf<br/> iterators or (as noted elsewhere) for staged iterators, of which<br/> the streambuf iterators are a good example.<br/> <br/> The bsearch and qsort functions cannot be overloaded properly as<br/> required by the standard because gcc does not yet allow overloading<br/> on the extern-"C"-ness of a function pointer.<br/> <br/> Chapter 26 Numerics<br/> --------------------<br/> Headers: <complex> <valarray> <numeric><br/> C headers: <cmath>, <cstdlib> (also 18, 21, 25)<br/> <br/> Numeric components: Gabriel dos Reis's valarray, Drepper's complex,<br/> and the few algorithms from the STL are "mostly done". Of course<br/> optimization opportunities abound for the numerically literate. It<br/> is not clear whether the valarray implementation really conforms<br/> fully, in the assumptions it makes about aliasing (and lack thereof)<br/> in its arguments.<br/> <br/> The C div() and ldiv() functions are interesting, because they are the<br/> only case where a C library function returns a class object by value.<br/> Since the C++ type div_t must be different from the underlying C type<br/> (which is in the wrong namespace) the underlying functions div() and<br/> ldiv() cannot be re-used efficiently. Fortunately they are trivial to<br/> re-implement.<br/> <br/> Chapter 27 Iostreams<br/> ---------------------<br/> Headers: <iosfwd> <streambuf> <ios> <ostream> <istream> <iostream><br/> <iomanip> <sstream> <fstream><br/> C headers: <cstdio> <cwchar> (also in 21)<br/> <br/> Iostream is currently in a very incomplete state. <iosfwd>, <iomanip>,<br/> ios_base, and basic_ios<> are "mostly complete". basic_streambuf<> and<br/> basic_ostream<> are well along, but basic_istream<> has had little work<br/> done. The standard stream objects, <sstream> and <fstream> have been<br/> started; basic_filebuf<> "write" functions have been implemented just<br/> enough to do "hello, world".<br/> <br/> Most of the istream and ostream operators << and >> (with the exception<br/> of the op<<(integer) ones) have not been changed to use locale primitives,<br/> sentry objects, or char_traits members.<br/> <br/> All these templates should be manually instantiated for char and<br/> wchar_t in a way that links only used members into user programs.<br/> <br/> Streambuf is fertile ground for optimization extensions. An extended<br/> interface giving iterator access to its internal buffer would be very<br/> useful for other library components.<br/> <br/> Iostream operations (primarily operators << and >>) can take advantage<br/> of the case where user code has not specified a locale, and bypass locale<br/> operations entirely. The current implementation of op<</num_put<>::put,<br/> for the integer types, demonstrates how they can cache encoding details<br/> from the locale on each operation. There is lots more room for<br/> optimization in this area.<br/> <br/> The definition of the relationship between the standard streams<br/> cout et al. and stdout et al. requires something like a "stdiobuf".<br/> The SGI solution of using double-indirection to actually use a<br/> stdio FILE object for buffering is unsatisfactory, because it<br/> interferes with peephole loop optimizations.<br/> <br/> The <sstream> header work has begun. stringbuf can benefit from<br/> friendship with basic_string<> and basic_string<>::_Rep to use<br/> those objects directly as buffers, and avoid allocating and making<br/> copies.<br/> <br/> The basic_filebuf<> template is a complex beast. It is specified to<br/> use the locale facet codecvt<> to translate characters between native<br/> files and the locale character encoding. In general this involves<br/> two buffers, one of "char" representing the file and another of<br/> "char_type", for the stream, with codecvt<> translating. The process<br/> is complicated by the variable-length nature of the translation, and<br/> the need to seek to corresponding places in the two representations.<br/> For the case of basic_filebuf<char>, when no translation is needed,<br/> a single buffer suffices. A specialized filebuf can be used to reduce<br/> code space overhead when no locale has been imbued. Matt Austern's<br/> work at SGI will be useful, perhaps directly as a source of code, or<br/> at least as an example to draw on.<br/> <br/> Filebuf, almost uniquely (cf. operator new), depends heavily on<br/> underlying environmental facilities. In current releases iostream<br/> depends fairly heavily on libio constant definitions, but it should<br/> be made independent. It also depends on operating system primitives<br/> for file operations. There is immense room for optimizations using<br/> (e.g.) mmap for reading. The shadow/ directory wraps, besides the<br/> standard C headers, the libio.h and unistd.h headers, for use mainly<br/> by filebuf. These wrappings have not been completed, though there<br/> is scaffolding in place.<br/> <br/> The encapsulation of certain C header <cstdio> names presents an<br/> interesting problem. It is possible to define an inline std::fprintf()<br/> implemented in terms of the 'extern "C"' vfprintf(), but there is no<br/> standard vfscanf() to use to implement std::fscanf(). It appears that<br/> vfscanf but be re-implemented in C++ for targets where no vfscanf<br/> extension has been defined. This is interesting in that it seems<br/> to be the only significant case in the C library where this kind of<br/> rewriting is necessary. (Of course Glibc provides the vfscanf()<br/> extension.) (The functions related to exit() must be rewritten<br/> for other reasons.)<br/> <br/> <br/> Annex D<br/> -------<br/> Headers: <strstream><br/> <br/> Annex D defines many non-library features, and many minor<br/> modifications to various headers, and a complete header.<br/> It is "mostly done", except that the libstdc++-2 <strstream><br/> header has not been adopted into the library, or checked to<br/> verify that it matches the draft in those details that were<br/> clarified by the committee. Certainly it must at least be<br/> moved into the std namespace.<br/> <br/> We still need to wrap all the deprecated features in #if guards<br/> so that pedantic compile modes can detect their use.<br/> <br/> Nonstandard Extensions<br/> ----------------------<br/> Headers: <iostream.h> <strstream.h> <hash> <rbtree><br/> <pthread_alloc> <stdiobuf> (etc.)<br/> <br/> User code has come to depend on a variety of nonstandard components<br/> that we must not omit. Much of this code can be adopted from<br/> libstdc++-v2 or from the SGI STL. This particularly includes<br/> <iostream.h>, <strstream.h>, and various SGI extensions such<br/> as <hash_map.h>. Many of these are already placed in the<br/> subdirectories ext/ and backward/. (Note that it is better to<br/> include them via "<backward/hash_map.h>" or "<ext/hash_map>" than<br/> to search the subdirectory itself via a "-I" directive.<br/> </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="../index.html">Home</a></td><td align="right" valign="top"> Appendix B. Porting and Maintenance </td></tr></table></div></body></html>
Go to most recent revision | Compare with Previous | Blame | View Log