lh | 9ed821d | 2023-04-07 01:36:19 -0700 | [diff] [blame] | 1 | Conformance of the GNU libc with various standards |
| 2 | ================================================== |
| 3 | |
| 4 | The GNU libc is designed to be conformant with existing standard as |
| 5 | far as possible. To ensure this I've run various tests. The results |
| 6 | are presented here. |
| 7 | |
| 8 | |
| 9 | Open Group's hdrchk |
| 10 | =================== |
| 11 | |
| 12 | The hdrchk test suite is available from the Open Group at |
| 13 | |
| 14 | ftp://ftp.rdg.opengroup.org/pub/unsupported/stdtools/hdrchk/ |
| 15 | |
| 16 | I've last run the suite on 2004-04-17 on a Linux/x86 system running |
| 17 | a Fedora Core 2 test 2 + updates with the following results [*]: |
| 18 | |
| 19 | FIPS No reported problems |
| 20 | |
| 21 | POSIX90 No reported problems |
| 22 | |
| 23 | XPG3 Prototypes are now in the correct header file |
| 24 | |
| 25 | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ |
| 26 | *** Starting unistd.h |
| 27 | Missing: extern char * cuserid(); |
| 28 | Missing: extern int rename(); |
| 29 | *** Completed unistd.h |
| 30 | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ |
| 31 | |
| 32 | XPG4 Prototype is now in the correct header file |
| 33 | and the _POSIX2_C_VERSION symbol has been removed |
| 34 | |
| 35 | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ |
| 36 | *** Starting unistd.h |
| 37 | Missing: extern char * cuserid(); |
| 38 | Missing: #define _POSIX2_C_VERSION (-1L) |
| 39 | *** Completed unistd.h |
| 40 | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ |
| 41 | |
| 42 | POSIX96 Prototype moved |
| 43 | (using "base realtime threads" subsets) |
| 44 | |
| 45 | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ |
| 46 | *** Starting unistd.h |
| 47 | Missing: extern int pthread_atfork(); |
| 48 | *** Completed unistd.h |
| 49 | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ |
| 50 | |
| 51 | UNIX98 Prototypes moved and _POSIX2_C_VERSION removed |
| 52 | (using "base realtime threads mse lfs" subset) |
| 53 | |
| 54 | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ |
| 55 | *** Starting unistd.h |
| 56 | Missing: extern char * cuserid(); |
| 57 | Missing: #define _POSIX2_C_VERSION (-1L) |
| 58 | Missing: extern int pthread_atfork(); |
| 59 | *** Completed unistd.h |
| 60 | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ |
| 61 | |
| 62 | |
| 63 | That means all the reported issues are due to the headers having been |
| 64 | cleaned up for recent POSIX/Unix specification versions. Duplicated |
| 65 | prototypes have been removed and obsolete symbols have been removed. |
| 66 | Which means that as far as the tests performed by the script go, the |
| 67 | headers files comply to the current POSIX/Unix specification. |
| 68 | |
| 69 | |
| 70 | [*] Since the scripts are not clever enough for the way gcc handles |
| 71 | include files (namely, putting some of them in gcc-local directory) I |
| 72 | copied over the iso646.h, float.h, and stddef.h headers and ignored the |
| 73 | problems resulting from the split limits.h file). |
| 74 | |
| 75 | |
| 76 | Technical C standards conformance issues in glibc |
| 77 | ================================================= |
| 78 | |
| 79 | If you compile programs against glibc with __STRICT_ANSI__ defined |
| 80 | (as, for example, by gcc -ansi, gcc -std=c89, gcc -std=iso1990:199409 |
| 81 | or gcc -std=c99), and use only the headers specified by the version of |
| 82 | the C standard chosen, glibc will attempt to conform to that version |
| 83 | of the C standard (as indicated by __STDC_VERSION__): |
| 84 | |
| 85 | GCC options Standard version |
| 86 | -ansi ISO/IEC 9899:1990 |
| 87 | -std=c89 ISO/IEC 9899:1990 |
| 88 | -std=iso9899:199409 ISO/IEC 9899:1990 as amended by Amd.1:1995 |
| 89 | -std=c99 ISO/IEC 9899:1999 |
| 90 | |
| 91 | (Note that -std=c99 is not available in GCC 2.95.2, and that no |
| 92 | version of GCC presently existing implements the full C99 standard.) |
| 93 | |
| 94 | You may then define additional feature test macros to enable the |
| 95 | features from other standards, and use the headers defined in those |
| 96 | standards (for example, defining _POSIX_C_SOURCE to be 199506L to |
| 97 | enable features from ISO/IEC 9945-1:1996). |
| 98 | |
| 99 | There are some technical ways in which glibc is known not to conform |
| 100 | to the supported versions of the C standard, as detailed below. Some |
| 101 | of these relate to defects in the standard that are expected to be |
| 102 | fixed, or to compiler limitations. |
| 103 | |
| 104 | |
| 105 | Defects in the C99 standard |
| 106 | =========================== |
| 107 | |
| 108 | Some defects in C99 were corrected in Technical Corrigendum 1 to that |
| 109 | standard. glibc follows the corrected specification. |
| 110 | |
| 111 | |
| 112 | Implementation of library functions |
| 113 | =================================== |
| 114 | |
| 115 | The implementation of some library functions does not fully follow the |
| 116 | standard specification: |
| 117 | |
| 118 | C99 added additional forms of floating point constants (hexadecimal |
| 119 | constants, NaNs and infinities) to be recognised by strtod() and |
| 120 | scanf(). The effect is to change the behavior of some strictly |
| 121 | conforming C90 programs; glibc implements the C99 versions only |
| 122 | irrespective of the standard version selected. |
| 123 | |
| 124 | C99 added %a as another scanf format specifier for floating point |
| 125 | values. This conflicts with the glibc extension where %as, %a[ and |
| 126 | %aS mean to allocate the string for the data read. A strictly |
| 127 | conforming C99 program using %as, %a[ or %aS in a scanf format string |
| 128 | will misbehave under glibc if it does not include <stdio.h> and |
| 129 | instead declares scanf itself; if it gets the declaration of scanf |
| 130 | from <stdio.h>, it will use a C99-conforming version. |
| 131 | |
| 132 | |
| 133 | Compiler limitations |
| 134 | ==================== |
| 135 | |
| 136 | The macros __STDC_IEC_559__, __STDC_IEC_559_COMPLEX__ and |
| 137 | __STDC_ISO_10646__ are properly supposed to be constant throughout the |
| 138 | translation unit (before and after any library headers are included). |
| 139 | However, they mainly relate to library features, and GCC only knows to |
| 140 | preinclude <stdc-predef.h> to get their definitions in version 4.8 and |
| 141 | later. Programs that test them before including any standard headers |
| 142 | may misbehave with older compilers. |
| 143 | |
| 144 | GCC doesn't support the optional imaginary types. Nor does it |
| 145 | understand the keyword _Complex before GCC 3.0. This has the |
| 146 | corresponding impact on the relevant headers. |
| 147 | |
| 148 | glibc's <tgmath.h> implementation is arcane but thought to work |
| 149 | correctly; a clean and comprehensible version requires compiler |
| 150 | builtins. |
| 151 | |
| 152 | For most of the headers required of freestanding implementations, |
| 153 | glibc relies on GCC to provide correct versions. (At present, glibc |
| 154 | provides <stdint.h>, and GCC doesn't before version 4.5.) |
| 155 | |
| 156 | The definition of math_errhandling conforms so long as no translation |
| 157 | unit using math_errhandling is compiled with -fno-math-errno, |
| 158 | -fno-trapping-math or options such as -ffast-math that imply these |
| 159 | options. math_errhandling is only conditionally defined depending on |
| 160 | __FAST_MATH__; the compiler does not provide the information needed |
| 161 | for more exact definitions based on settings of -fno-math-errno and |
| 162 | -fno-trapping-math, possibly for only some source files in a program. |
| 163 | |
| 164 | |
| 165 | Issues with headers |
| 166 | =================== |
| 167 | |
| 168 | None known. |