Hi, I have observed this behavior since the early 21.5 version days, but I always thought it would go away. Apparently not. When you load an UTF-8 file in xemacs, some characters do not get displayed - instead, a GETA MARK (U+3013) is displayed (looks like this 〓). This is generally ok for display purposes, but the problem is that when I save that file, all the characters that cannot be displayed get replaced by the GETA MARK, which is unacceptable. To reproduce this problem, one can use the UCA rules file, located at http://oss.software.ibm.com/cvs/icu/~checkout~/icu/source/data/unidata/UC... Load that file in xemacs and save it under a different name. A significant number of code points will go to U+3013. This file has a nice selection of Unicode characters, so the problem should be easily reproduced. The first code point that gets smashed is COMBINING GRAPHEME JOINER (U+034F). There is a number of code points that cause a ~ to be displayed, like some Tibetan characters. These get preserved properly. I'm building on Win2K, no special configuration changes. My config.inc is attached. Thank you! Regards, v. # -*- mode: makefile -*- ############################################################################ # Install options # ############################################################################ INSTALL_DIR=c:\Program Files\XEmacs\XEmacs-$(XEMACS_VERSION_STRING) PACKAGE_PREFIX=c:\Program Files\XEmacs ############################################################################ # Compiled-in features: basic # ############################################################################ # Multilingual support. MULE=1 # Native MS Windows support. HAVE_MS_WINDOWS=1 # GTK support. Do NOT set this to 1; this does not currently work. HAVE_GTK=0 GTK_DIR= ############################################################################ # Compiled-in features: graphics formats # ############################################################################ # Set this to enable XPM support (virtually mandatory), and specify # the directory containing xpm. Get the library from # http://ftp.xemacs.org/aux/xpm-3.4k.tar.gz. HAVE_XPM=1 XPM_DIR=c:\src\libs\xpm-3.4k # Set this to enable GIF support (built-in). HAVE_GIF=1 # Set this to enable PNG support (virtually mandatory), and specify # the directories containing png and zlib. Get the latest version from # ftp://ftp.uu.net/graphics/png/. You will have to rename the zlib directory # from zlib-1.1.3 or whatever to just `zlib' for the build to work. HAVE_PNG=1 PNG_DIR=c:\src\libs\png-1.0.3 ZLIB_DIR=c:\src\libs\zlib-1.1.3 # Set this to enable TIFF support, and specify the directory containing tiff. # Get the latest version from ftp://ftp.uu.net/graphics/tiff/. Not on by # default since TIFF isn't really very important and those TIFF wankers # couldn't be bothered to incorporate minimal MS-Windows patches they've # had sitting around for years, so getting it to build is a major pain in # the ass. HAVE_TIFF=0 TIFF_DIR=c:\src\tiff-v3.4 # Set this to enable JPEG support, and specify the directory containing jpeg. # Get the latest version from ftp://ftp.uu.net/graphics/jpeg/. HAVE_JPEG=1 JPEG_DIR=c:\src\libs\jpeg-6b # Set this to enable XFace support, and specify the directory containing # compface. Get the library from http://ftp.xemacs.org/aux/compface.tar.gz. HAVE_XFACE=0 COMPFACE_DIR= ############################################################################ # Build settings # ############################################################################ # If you want to the built files to be placed outside of the source tree # (e.g. this allows you to build multiple versions of XEmacs, with # different configuration settings, from the same source tree), run # `make-build-dir' to create a skeleton build tree, giving it the name of a # path. This creates the specified directory and the `nt' directory below # it, copies config.inc (if it exists), config.inc.samp and xemacs.mak into # the `nt' directory, and modifies the config files to contain the path of # the source tree in SOURCE_DIR. This will not overwrite files that # already exist, so it can safely be run more than once on the same tree. # # Running nmake in the skeleton build tree will then build XEmacs in that # directory tree, using the source files as specified. The paths of the # `lisp' and `etc' directories in the source tree will be compiled into the # executable as "last-resort" values -- i.e. they will be used if you # simply run the executable as-is, but will not override any local copy of # the `lisp' and/or `etc' directories that you may have made. # # Alternatively, you can just uncomment the line below for BUILD_DIR and # specify a (possibly non-existent) path. Running nmake will then put its # build files into a parallel directory structure underneath the specified # path, creating the directories as necessary. The problem with this is # that the first method above allows you to have a different copy of # `config.inc' for each build directory, but doing it this way means you # have only one version of config.inc, and have to manually change it for # each different build. # NOTE: These cannot be relative paths. If you want the source and build to # be relatives of each other, use $(MAKEROOT) to refer to the root of the # current tree -- that's one level up from where xemacs.mak is located. # SOURCE_DIR=c:\src\xemacs\working # BUILD_DIR=c:\src\xemacs\msbuilds\working # Set this to specify the location of makeinfo. (If not set, XEmacs will # attempt to use its built-in, much slower texinfo support when building # info files.) If you are building XEmacs yourself, you probably have # Cygwin sitting around already. If not, you should. Cygwin provides a # `makeinfo.exe' in /usr/bin/makeinfo (/usr/bin is virtual, it's /bin in # the actual file system). #MAKEINFO=c:\cygwin\bin\makeinfo.exe MAKEINFO=c:\src\libs\texinfo-4.2\makeinfo\makeinfo.exe # Set this to turn on optimization when compiling. OPTIMIZED_BUILD=0 # Set this to build with the fastcall calling convention, which uses registers # instead of the stack and should speed things up a bit # #### Change to 1 when I check in the ws with support for fastcall USE_FASTCALL=0 ############################################################################ # Development options # ############################################################################ # Set this to compile in support for profiling. If you want line-by-line # profiling under VC++, you also need debugging turned on. PROFILE_SUPPORT=0 # Set this to enable debug code in XEmacs that doesn't slow things down, # and to add debugging information to the executable. (The code that's # enabled in XEmacs is primarily extra commands that aid in debugging # problems. The kind of debugging code that slows things down -- # i.e. internal error-checking -- is controlled by the ERROR_CHECK_ALL # variable, below.) DEBUG_XEMACS=1 # Set this to enable support for edit-and-continue under VC++. # WARNING: This turns on incremental linking, which is known to lead to # occasional weird crashes in pdump loading. If that happens, do a # nmake -f xemacs.mak clean so that temacs.exe and xemacs.exe get removed. SUPPORT_EDIT_AND_CONTINUE=0 # Uncomment this to turn off or on the error-checking code, which adds # abundant internal error checking (and slows things down a lot). Normally, # leave this alone -- it will be on for beta builds and off for release # builds. # ERROR_CHECK_ALL=0 # Uncomment this to turn on or off whether we compile source files as C++ # files. This turns on additional error checking of various sorts. Normally, # leave it alone -- it will be on when ERROR_CHECK_ALL is on. # CPLUSPLUS_COMPILE=0 # Set this to speed up building, for development purposes. # WARNING: This may not completely rebuild all targets. In particular, # DOC is not rebuilt, and changes to lisp.h and config.h do not trigger # mass rebuilding. Other things may also be enabled that are not safe # for release builds. QUICK_BUILD=0 # Set this to see exactly which compilation commands are being run (not # generally recommended). VERBOSECC=0 # Set this to get nmake to use dependency info (recommended for development). # Requires cygwin or ActiveState versions of Perl to be installed. DEPEND=0 # Set this to use the portable dumper for dumping the preloaded Lisp # routines, instead of the older "unexec" routines in unexnt.c. USE_PORTABLE_DUMPER=1 # Set this to use the new experimental garbage-collection routines instead # of the traditional XEmacs garbage-collection routines. USE_KKCC=0 # Set this to turn on the use of the union type, which gets you improved # type checking of Lisp_Objects -- they're declared as unions instead of # ints, and so places where a Lisp_Object is mistakenly passed to a routine # expecting an int (or vice-versa), or a check is written `if (foo)' # instead of `if (!NILP (foo))', will be flagged as errors. (All of these # do NOT lead to the expected results! Qnil is not represented as 0 [so if # (foo) will *ALWAYS* be true for a Lisp_Object], and the representation of # an integer as a Lisp_Object is not just the integer's numeric value, but # usually 2x the integer +/- 1.) # There used to be a claim that it simplified debugging. There may have # been a grain of truth to this pre-19.8, when there was no lrecord type # and all objects had a separate type appearing in the tag. Nowadays, # however, there is no debugging gain, and in fact frequent debugging *LOSS*, # since many debuggers don't handle unions very well, and usually there is # no way to directly specify a union from a debugging prompt. # Furthermore, release builds should *NOT* be done this way because (a) you # may get less efficiency, with compilers that can't figure out how to # optimize the union into a machine word; (b) even worse, the union type # often triggers compiler bugs, especially when combined with Mule and # error-checking. This has been the case with various times using GCC, # *AND CURRENTLY HAPPENS WITH VC++*, at least when using pdump. Therefore, # be warned! USE_UNION_TYPE=0