uclibc-ng/TODO

TODO list for the uClibc 0.9.28 release:
-------------------------------------------------
    *) Audit header files.  Remove prototypes for all functions that
	are not supported -- especially needed for the libm headers.
    *) Audit header files.  When options are disabled, also disable
	them in the include files as well by checking for the proper
	define from include/bits/uClibc_config.h (pulled in from features.h)
    *) Test each architecture with the latest LTP testsuite, and post
	LTP testsuite results for each architecture on uclibc.org.


TODO list for the uClibc 1.0.0 release:
-------------------------------------------------

    *) Hide uClibc internal symbols from our exported namespace.
    *) Explicity add some sortof CONFIG_LINUX_2_2_ABI, CONFIG_LINUX_2_4_ABI
	and CONFIG_LINUX_2_6_ABI type options, rather than having the abi
	depend on the selected set of kernel headers.  This will likely also
	require that we provide a set of kernel headers (probably a derivitive
	of Mazur's linux-libc-headers) that can supplies the 2.2, 2.4, and 2.6
	kernel abi.
    *) Documentation updates:
	    *) Write a uClibc HOWTO document
	    *) Update README document
	    *) Update INSTALL document
	    *) Update docs/Glibc_vs_uClibc_Differences.txt document
		and fully document all differences between the feature
		set of uClibc and glibc.
	    *) Update docs/uClibc_vs_SuSv3.txt document
	    *) Update docs/threads.txt document
	    *) Write man pages for ldd and ldconfig utility binaries
    *) Implement some mechanism (perhaps encoded in the .so name,
	perhaps using an abi tag magically embedded into each object)
	for flagging config options that break the ABI.  Options
	such as UCLIBC_HAS_SOFT_FLOAT, UCLIBC_HAS_THREADS, UCLIBC_HAS_LOCALE,
	and perhaps others (finalize list) produce a lib with a differing
	ABI.  Make it so apps cannot use an ABI mis-matched uClibc.
	This is most easily done using symbol versioning...
    *) Implement the long double versions of math funcs
	using wrappers on top of the double versions (size / precision
	trade off where size clearly wins).
    *) Make all small objects (>~50 bytes) into either inlines or
	into a static library
    *) Cleanup / rewrite sysconf.c.  It should get some information
	from ldso (such as HZ).  Other stuff it currently just makes
	up, which is obviously wrong.  Also bits/uClibc_clk_tck.h
	needs to be updated at the same time to get proper HZ values.
    *)  It would nice if valgrind wouldn't complain about the atexit() malloc'd
	memory for destructors, which happens since the dynamic linker calls
	atexit(), which calls malloc() prior to valgrind starting, so valgrind
	complains because it didn't see that memory allocated.
    *) poll emulation using select() for old 2.0.x uClinux kernels
	in libc/sysdeps/linux/common/poll.c fails some python self-tests.
	Of course, modern systems using the actuall poll() syscall work fine.
    *) Rework the build system Makefile to eliminate recursive make.
    *) Build both pic and non-pic objects where appropriate, so that
	static libs need not pay the pic size penalty.
    *) Cleanup/scrub all the Makefile copyright junk
    *) Fix dlopen, for both static and dynamic cases, and make it
	fully comply with SuSv3
    *) From the the ELF spec "...All shared object initializations happen
	before the executable file gains control.  ...  Before the initialization
	code for any object A is called, the initialization code for any other
	objects that object A depends on are called.  For these purposes, an object
	A depends on another object B, if B appears in Ads list of needed objects
	(recorded in the DT_NEEDED entries of the dynamic structure). The order of
	initialization for circular dependencies is undefined."  uClibc's shared
	lib loader should be fixed to run ctors in the specified order.


TODO list for AFTER the uClibc 1.0.0 release:
-------------------------------------------------
    *) Add support for Linux 2.6.x NPTL pthreads, futexes, etc
    *) Add support for Linux 2.6.x fast vsyscalls
    *) Enable pristine source tree builds
    *) Fix regex so it isn't so stinking big
    *) Fix glob so it isn't so stinking big
    *) run 'nm -D --size-sort -t d libuClibc-0.9.26.so' and work on the
	biggest things (i.e. stuff at the end of the list) to make
	them smaller.
    <more wishlist items here>




-----------------------------------------------------------------------------
Manuel's todo:

  1) Little things that need fixing:
  ----------------------------------
  a) Fix bug in *printf: outdigit precison bug
  b) Check that gnu/bsd extension members tm_gmtoff and tm_zone in struct tm
     are respected where they should be.
  c) Implement the obstack printf funcs for glibc compat.
  d) Implement glibc 'a' flag for scanf string conversions.
  e) Allow use of the older non-table-based ctype functions when using
     stub locale support. (smaller)

  2) Additional str{f|p}time issues.
  ----------------------------------
  a) Spacing issue wrt strptime.
  b) Support locale specific alternate digits.  (data is in place)
  c) Support locale era in year designations.   (data is in place)
  d) Deal with mb format string issues in strftime.
  e) Implement wcsftime.

  3) Other locale issues (my implementation):
  -------------------------------------------
  a) Do a little more clean up of ctype and wctype.
  b) Rework of the locale data organization to make using locales reasonable
     when staticly linking.  (mmap)
  c) Rewrite the locale data generation tools to process the text specifications
     rather than relying on glibc.
  d) Adapt regex lib to use my collation data and add the necessary collating
     item tables to support SUSv3 required features.
  e) transliteration of unsupported wchars in 8-bit locales (like glibc).
  f) Support ISO/IEC 14652 draft locale extensions (LC_PAPER, etc).
  g) Implement strfrom.
  h) Shift-state codeset locale support?

  4) Misc:
  --------
  a) Port uClibc to other OSs (including elks), or even bare metal (libgloss).
  b) Write a space-efficient gettext substitute, to avoid storing large amounts
     of redundant data.