source: trunk/third/gettext2/os2/README.OS2 @ 22167

Welcome!
========

This is the OS/2 port of GNU gettext internationalization library.


Compatibility
=============

The library has been compiled with -Zmt flag, but it doesn't matter as soon
as you use the EMX single-threaded runtime fix (emx-strt-fix-0.0.2.zip).

The library is fully compatible with the previous port of gettext library
(0.10.35) which is largely used especialy by XFree86/2 programs. All the
old programs that I have with gettext support run fine with the new version
of the DLL.


Installation
============

If you set the GNULOCALEDIR environment variable to point to your
x:/xxx/share/locale directory, it will override any other setting. That is,
unpack the binary distribution over /emx, set GNULOCALEDIR=x:/emx/share/locale
(where x: is the drive letter of your EMX installation) and that's all.

If you use the UNIXROOT environment variable, the default catalogue search
paths will be like on Unices, e.g. $(UNIXROOT)/usr/lib and
$(UNIXROOT)/usr/share/locale. GNULOCALEDIR always overrides this.

Now if you haven't did it earlier, set the language identifier that you use.
This is done by adding a "SET LANG=xxx" environment setting to your CONFIG.SYS,
where xxx is the identifier of your language (example: en_UK for English in UK,
ru_RU for Russian in Russia. Also you can use names like "russian", "italian"
and so on - see the share/locale/locale.alias file).

This port of gettext supports character set conversions. This means that if
your .mo files were written using new gettext guidelines, e.g. they contain a
message like this:

msgid ""
msgstr "Content-Type: text/plain; charset=koi8-r\n"

the messages will be properly converted to your active codepage using OS/2
Unicode API. For example, russian message catalog gettext.mo is in the
KOI8-R (codepage 878) encoding while OS/2 uses codepage 866. Now when you
run any of these tools it detects that the active OS/2 codepage is 866 and
performs the translation from CP878 -> CP866 for every message.

If you want to override the character set used to output messages (for example
in XFree86 for Russian the KOI8-R encoding (codepage 878) is used) you can
set the output character set by adding a postfix to the LANG environment
variable, this way:

set LANG=ru_RU.KOI8-R

or (equivalent):

set LANG=ru_RU.CP878

or (same effect):

set LANG=ru_RU.IBM-878

If the output character set is ommited from the LANG variable, the default
codepage is ALWAYS taken from the operating system (e.g. the codepage setting
from locale.alias is always ignored, so "russian" stays just for "ru_RU" and
not for "ru_RU.ISO-8859-5"); you may want to set it just if you want to
override the active OS/2 codepage.


XFree86 setup
=============

If you use XFree86 and the OS/2 default character set is different from the
XFree86 default character set (e.g. for Russain CP866 vs KOI8-R), you can add
the following (or similar) statement to your startx.cmd file (after the
commands dealing with HOME and X11SHELL):

call VALUE 'LANG', 'ru_RU.KOI8-R', env

Otherwise you can get incorrect (wrong codepage) output from programs that
previously worked (e.g. GIMP 1.22). This is because earlier versions of gettext
didn't support character set translations.


Implementation remarks
======================

The codepage conversion code uses OS/2 Unicode API, thus it falls under the
limits that OS/2 Unicode API has. For example, OS/2 Unicode API does not
support the BIG5 East Asian character set nor ISO-8859-X where X > 9 (at
least with Warp4 with fixpack 14 that I have). If someone knows the
OS/2 API identifiers for BIG5 or ISO8859-10,... encodings, please tell me!

Since gettext 0.11 iconv emulation layer supports correctly UTF-8. Also
I have added theoretical support for the following East Asian encodings:
EUC-JP, EUC-KR, EUC-TW, EUC-CN. However, these encodings are (I believe)
supported only on East Asian editions of OS/2. The code pages for them are
listed in the \language\codepage\ucstbl.lst file but the codepage files
themselves are missing; I believe they are ommited from European OS/2's
due to their large size.

Also I have added "support" for the BIG5 codeset as an alias for IBM-950
codepage. However, I'm not very sure about this; in any case OS/2 does not
support (as far as I know) anything closer to BIG5.


Additional API
==============

This package provides additionaly the iconv() API that can be used by
developers for doing more feature-full Unix ports. The iconv() API is used
to convert text between various codepages. The intl.h header file contains
the prototypes and definitions needed for iconv(); if you configure software
with autoconf it possibly will find intl.h and set up the software accordingly.

All these functions are exported from INTL.DLL. The iconv.a import library
imports all the iconv* functions from INTL.DLL. So, like on Unix, now you can
#include <iconv.h>, then link with -liconv and you will get a fully functional
iconv implementation.


Rebuilding the library
======================

The library is quite easy to rebuild. Since the OS/2 support is provided now
out-of-the-box in gettext, you just have to download and unpack the source
archive. Now there are two ways to rebuild the gettext library:

1. If you're a masochist you can go the clumsy configure/make Unix way. This
is not recommended however as I found no way to tell libtool to generate a
slightly non-standard DLL which will be backward compatible with gettext
0.10.35. The compatibility is achieved by prepending backward.def to the
export definition file generated with emximp or somehow else. Thus it is
highly recommended you build using the second way, if it is possible.

2. Go to os2 and just run `make'. If you have all the required tools,
it should painlessly compile. Finally, if you want a binary distribution
archive, do `make distr'. The weak side of building this way is that makefile
is somewhat fragile. This means that if the makefile is left unmodified and
a new version of gettext is rolled out, it *may* not work. But every possible
attempt was made to ensure that the makefile takes most important build
parameters from their autoconf counterparts.

WARNING: Due to bugs in GNU Make 3.76.1 (at least in its OS/2 port) you can
get sometimes (depending on make version and makefile modification :) funny
messages like these:

zip warning: name not matched: emx/src/gettext-0.10.40/support/os2/iconv.h

or even:

*** No rule to make target `out/release/intl.a', needed by `all'.  Stop.

Such messages are a bad joke. Ignore it, and re-run make. This is a
long-standing bug in GNU make, alas.

If you want a debug version of library, you can do `make DEBUG=1'.

If you don't have the LxLite tool installed, do `make LXLITE=0'

NB: For best results, it is highly recommended that you use at least emxbind.exe
and ld.exe from gcc 3.0.2 or later, since they contain a number of fixes that
will help you generate a more optimal DLL.


Contributors
============

Hung-Chi Chu <hcchu@r350.ee.ntu.edu.tw>
        the original port of gettext (0.10.35)

Jun SAWATAISHI <jsawa@attglobal.net>
        some more work on it and submitted the patches to GNU team, although
        they were not completely integrated.

Andrew Zabolotny <zap@cobra.ru>
        Succeeded to remove almost all OS/2-specific #ifdef's from mainstream
        source code, wrote the dedicated OS/2 makefile, wrote the iconv wrapper
        around OS/2 Unicode API, added support for locale translations.