Return to DESIGN CVS log

Up to [ELWIX - Embedded LightWeight unIX -] / embedaddon / libiconv

Annotation of embedaddon/libiconv/DESIGN, revision 1.1.1.1

1.1       misho       1: While some other iconv(3) implementations - like FreeBSD iconv(3) - choose
                      2: the "many small shared libraries" and dlopen(3) approach, this implementation
                      3: packs everything into a single shared library. Here is a comparison of the
                      4: two designs.
                      5: 
                      6: * Run-time efficiency
                      7:   1. A dlopen() based approach needs a cache of loaded shared libraries.
                      8:   Otherwise, every iconv_open() call will result in a call to dlopen()
                      9:   and thus to file system related system calls - which is prohibitive
                     10:   because some applications use the iconv_open/iconv/iconv_close sequence
                     11:   for every single filename, string, or piece of text.
                     12:   2. In terms of virtual memory use, both approaches are on par. Being shared
                     13:   libraries, the tables are shared between any processes that use them.
                     14:   And because of the demand loading used by Unix systems (and because libiconv
                     15:   does not have initialization functions), only those parts of the tables
                     16:   which are needed (typically very few kilobytes) will be read from disk and
                     17:   paged into main memory.
                     18:   3. Even with a cache of loaded shared libraries, the dlopen() based approach
                     19:   makes more system calls, because it has to load one or two shared libraries
                     20:   for every encoding in use.
                     21: 
                     22: * Total size
                     23:   In the dlopen(3) approach, every shared library has a symbol table and
                     24:   relocation offset. All together, FreeBSD iconv installs more than 200 shared
                     25:   libraries with a total size of 2.3 MB. Whereas libiconv installs 0.45 MB.
                     26: 
                     27: * Extensibility
                     28:   The dlopen(3) approach is good for guaranteeing extensibility if the iconv
                     29:   implementation is distributed without source. (Or when, as in glibc, you
                     30:   cannot rebuild iconv without rebuilding your libc, thus possibly
                     31:   destabilizing your system.)
                     32:   The libiconv package achieves extensibility through the LGPL license:
                     33:   Every user has access to the source of the package and can extend and
                     34:   replace just libiconv.so.
                     35:   The places which have to be modified when a new encoding is added are as
                     36:   follows: add an #include statement in iconv.c, add an entry in the table in
                     37:   iconv.c, and of course, update the README and iconv_open.3 manual page.
                     38: 
                     39: * Use within other packages
                     40:   If you want to incorporate an iconv implementation into another package
                     41:   (such as a mail user agent or web browser), the single library approach
                     42:   is easier, because:
                     43:   1. In the shared library approach you have to provide the right directory
                     44:   prefix which will be used at run time.
                     45:   2. Incorporating iconv as a static library into the executable is easy -
                     46:   it won't need dynamic loading. (This assumes that your package is under
                     47:   the LGPL or GPL license.)
                     48: 
                     49: 
                     50: All conversions go through Unicode. This is possible because most of the
                     51: world's characters have already been allocated in the Unicode standard.
                     52: Therefore we have for each encoding two functions:
                     53: - For conversion from the encoding to Unicode, a function called xxx_mbtowc.
                     54: - For conversion from Unicode to the encoding, a function called xxx_wctomb,
                     55:   and for stateful encodings, a function called xxx_reset which returns to
                     56:   the initial shift state.
                     57: 
                     58: 
                     59: All our functions operate on a single Unicode character at a time. This is
                     60: obviously less efficient than operating on an entire buffer of characters at
                     61: a time, but it makes the coding considerably easier and less bug-prone. Those
                     62: who wish best performance should install the Real Thing (TM): GNU libc 2.1
                     63: or newer.
                     64: