R56c is a bugfix-only release everyone must upgrade to:

[komh] Remove redundant OS/2-specific code, clean up others
[komh, tg] Fix drive-qualified (absolute and relative) DOS-style path support in realpath functionality, partially other places
[tg] Don’t substitute ${ENV:-~/.mkshrc} result again
[tg] Improve OS/2 $PATH (et al.) handling, drive-relative paths
[tg] Add MKSH_ENVDIR compile-time option for Jehanne and Plan 9
[tg] Limit nesting when parsing malformed code (Debian #878947)
[tg] Update wcwidth data with bugfixed script (still Unicode 10; resulting values are identical to glibc git master for extant chars)
[Dr. Werner Fink] Raise some time limits in the testsuite
[Shamar] Add support for the Jehanne operating system
[komh] Set stdin to text mode before executing child processes on OS/2
[komh] Pass arguments via a resonse file if executing a child fails
[Dr. Werner Fink] Early locale tracking as a compile-time option
[tg] Fix regressions introduced with new fast character classes

Bug: N/A
Test: builds and boots
Change-Id: I44da7a8bb9859a0357bcd59891dd4f2cfc199733

R56b is a bugfix-only release everyone should upgrade to:

[tg] Reference the FAQ webpage
[panpo, Riviera] Fix documentation bug wrt. Esc+Ctrl-L
[tg, Larry Hynes] Fix “0” movement in vi mode
[tg] Replace broken libcs’ offsetof macro with MirBSD’s

R56 is a bugfix release with some experimental fixes:

[tg, Seb] Do not apply alias name restrictions to hash/tilde tracking
[tg] Restore ‘.’, ‘:’ and ‘[’ in alias names (“[[” is still forbidden)
[tg] Fix accidentally defanged $PATHSEP test
[tg] On ^C (INTR and QUIT edchars), shove edit line into history
[iSKUNK, tg] Begin porting to z/OS using EBCDIC encoding, incomplete
[tg] Redo fast character classes code, adding POSIX and other helpers
[tg] bind parses backslash-escaped ‘^’ (and ‘\’) as escaped
[tg] Building with -DMKSH_ASSUME_UTF8=0 no longer causes a known failure in the testsuite
[tg] New test.sh option -U to pass a UTF-8 locale to use in the tests
[tg] re_format(7) BSD: [[ $x = *[[:\<:]]foo[[:\>:]]* ]]
[tg, iSKUNK] Use Config in check.pl only if it exists
[tg] New matching code for bracket expressions, full POSIX (8bit)
[komh] Exclude FAT/HPFS/NTFS-unsafe tests on OS/2 (and Cygwin/MSYS)
[tg] Update to Unicode 10.0.0
[tg, selk] Make readonly idempotent
[tg, multiplexd] When truncating the persistent history, do not change the underlying file, do all operations on the locked one; do not stop using the history at all if it has been truncated
[tg, Jörg] Turn off UTF-8 mode upon turning on POSIX mode
[Martijn Dekker, Geoff Clare, many on the Austin list, tg] In POSIX mode, make the exec builtin force a $PATH search plus execve
[tg] Fix GCC 7, Coverity Scan warnings
[tg, Michal Hlavinka] Track background process PIDs even interactive
[tg] Always expose mksh’s hexdump shell function; speed it up by working on the input in chunks; use character classes to make it EBCDIC safe
[tg] Revamp dot.mkshrc default editor selection mechanism

Bug: N/A
Test: builds and boots
Change-Id: I35d624c4352c1b7c1ee499f9680712a30c64323b

R55 is mostly a feature release with summary bugfixes:

[komh] Fix OS/2 search_access() and UNC path logic
[tg] Undocument printf(1) to avoid user confusion
[Jean Delvare, tg] Fix printf builtin -R option
[tg] Make ${var@x}, unknown x, fail (thanks izabera)
[tg] ${var=x} must evaluate x in scalar context (10x Martijn Dekker)
[tg] Fixup relation between lksh and mksh, reduce delta
[tg] Improve manpage display; add OS/2 $PATH FAQ
[Jean Delvare] Fix bugs in manpage
[tg] Review tilde expansion, removing “odd use of KEEPASN” and introduce POSIX “declaration utility” concept; wait isn’t one
[tg] Add \builtin utility, declaration utility forwarder
[tg] Make $'\xz' expand to xz, not \0
[tg] Use fixed string pooling (requires the above change in host mksh)
[tg] POSIX declaration commands can have varassign and redirections
[Martijn Dekker] Add typeset -g, replacing homegrown “global”
[Harvey-OS] Disable NOPROSPECTOFWORK, APEX is reportedly fixed now
[tg] Display ulimit -a output with flags; improve Haiku
[tg] Drop old let] hack, use \builtin internally
[tg] Fix padding in Lb64encode in dot.mkshrc
[tg] Move FAQ content to a separate, new FAQ section in the manpage
[tg] Add new standard variable PATHSEP (‘:’, ‘;’ on OS/2)
[Martijn Dekker] Fix LINENO in eval and alias
[komh] Fix “\builtin” on OS/2
[tg] Improve (internal) character classes code for speed
[tg] Fix: the underscore is no drive letter
[tg] No longer hard-disable persistent history support in lksh
[tg] Introduce build flag -T for enabling “textmode” on OS/2 (supporting CR+LF line endings, but incompatible with mksh proper)
[tg] Merge mksh-os2
[tg] Permit changing $OS2_SHELL during a running shell
[tg] Fix multibyte handling in ^R (Emacs search-history)
[tg] Allow “typeset -p arrname[2]” to work
[tg] Make some error messages more consistent
[tg, komh] Disable UTF-8 detection code for OS/2 as unrealistic
[tg, sdaoden] Limit alias name chars to POSIX plus non-leading ‘-’
[tg, Martijn Dekker] Expand aliases at COMSUB parse time
[tg] Make “typeset -f” output alias-resistent
[tg, Martijn Dekker] Permit “eval break” and “eval continue”
[tg] Make -masm=intel safe on i386
[tg] Disambiguate $((…)) vs. $((…)…) in “typeset -f” output
[Jean Delvare] Clarify the effect of exit and return in a subshell
[tg] Simplify compile-time asserts and make them actually compile-time
[tg] Fix ^O in Emacs mode if the line was modified (LP#1675842)
[tg] Address Coverity Scan… stuff… now that it builds again
[Martijn Dekker, tg] Add test -v
[tg] Document set -o posix/sh completely

Bug: N/A
Test: manual
Change-Id: Ifce1d879933a5773e98b4f34f4a9bb86a6bdff3b

otapreopt_script is currently using this. We should probably move it onto
the more portable echo or printf, but for now I'm assuming there's a reason
why print is being used --- the use of -u implies isn't not just a typo for
printf.

(Even if we can switch to echo or printf, I'm tempted to just leave print be
anyway, since unlike the other disabled mksh builtins, it's not shadowing a
better implementation, so it's not causing confusion about why the "same"
command in different contexts behaves differently. The worst it will do is
cause portability problems if we move away from mksh, and in the meantime
print is documented on the mksh man page for users sophisticated enough to
know that we're using mksh, so it's arguably more surprising for it to be
missing.)

Bug: http://b/33761687
Test: manual
Change-Id: I440d759a3c4ef6fa24fd496ea832093e18b7060e

From the release notes...

R54 is a bugfix release with moderate new features:

[tg] Simplify and improve code and manual page
[tg] Try GCC 5’s new -malign-data=abi
[tg] Allow interrupting builtin cat even on fast devices (LP#1616692)
[tg] Update to Unicode 9.0.0
[Andreas Buschka] Correct English spelling
[tg] Handle set -e-related error propagation in || and && constructs correctly
[tg] Initialise memory for RNG even when not targeting Valgrind
[tg] Shrink binary size
[Brian Callahan] Improve support for the contemporary pcc compiler
[tg] Fix side effects with lazy evaluation; spotted by ormaaj
[tg] New flags -c (columnise), -l, -N for the print builtin
[Larry Hynes] Fix English, spelling mistakes, typos in the manpage
[tg, ormaah] Return 128+SIGALRM if read -t times out, like GNU bash
[Martijn Dekker] Install both manpages from Build.sh
[Martijn Dekker] Document case changes are ASCII-only
[Ronald G. Minnich, Elbing Miss, Álvaro Jurado, tg] Begin porting to Harvey-OS and APEX (similar to Plan 9 and APE)
[KO Myung-Hun] More infrastructure for the OS/2 (EMX, KLIBC) port

I've continued to disable the cat, realpath, and sleep builtins,
and added print to the list (now I've realized it's unrelated to
the print on the desktop).

Bug: N/A
Test: manual
Change-Id: Iead4c1b572c6f288b2ab2b884125dcb606f1e356

This makes us match bash in terms of what is/isn't a builtin.

Bug: http://b/20720875
Change-Id: Id90aeb115cf5e5217084bc1c5f382e14033b7d49

They're buggy and/or lacking in functionality. Disable them and use toybox
instead.

Bug: http://b/27064311
Bug: http://b/20720875
Change-Id: Ie57efffd7b98f172eafe9d4bd5ddf723811b6f29

Note that we skipped R52c because it wouldn't build with clang (thanks to
the "Shave 200 bytes off .text by revisiting string pooling" change).

From the release notes:

R53a is a snapshot/feature release:

[lintian] Fix spelling
[tg] Unbreak multi-line command history broken by history flush
[tg] Fix redefining POSIX functions that were Korn functions before
[tg, TNF] Fix bounds checks in Vi editing mode
[tg] Handle combining characters at end of string or output correctly
[tg] Fix ${!#} ${!?} ${!-} (POSIX, prompted by izabera)
[tg] Fix shf.c-internal buffer overread on printing digits
[J�rg] Fix a typo in the testsuite
[arekm] Increase default edit line size (unless MKSH_SMALL)
[tg] Improve description of Emacs mode keybindings, especially ^U
[tg, arekm, jilles] Abort read builtin in case of read(2) errors
[tg, izabera, carstenh] Fix most of the ambiguous corner cases related to ${[pfx]var[op[word]]} (${@:-1} still unsupported)
[carstenh] Contribute some more testsuite coverage
[tg] WDS_TPUTS now emits QCHAR newline reentrant-safe
[tg] Fix var=<< implementation (LP#1380389)
[tg, FreeBSD] Make XSI test(1) extensions behave as if they were POSIX
[tg, izabera] Add $(<<<x) and $(<<EOF…) implementation
[tg] Lower minimum screen size accepted as “sane” from the OS to 4×2
[tg, Torsten Sillke] Simplify tilde-expanded parameters
[tg, Torsten Sillke] Fix default PS1 for substring matches
[tg] Apply defer-builtin-with-arguments logic to realpath builtin
[tg] Rework string pooling (own vs. compiler’s) (LP#1580348)
[tg] Feature: print -A, prints arguments as characters
[tg, izabera] Replace <<< and >>> as ROL and ROR operators with their new ^< and ^> spelling as per this proposal
[tg, slagtc] Clear-to-EOL under tmux to work around its anti-feature
[tg, p120ph37] Remove support for using file descriptors with more than a single digit, in preparation for named file descriptors
[tg] Correct, but simplify (at the potential cost of more tty I/O than strictly necessary, though never redundant and (probably) not more than before when it was miscalculated), line clearing and redrawing
[slagtc, tg] Implement new evaluate-region editing command Esc+Ctrl-E
[tg] Prefer external rename utility over the recovery builtin
[tg] Remove redundant full-line redraws
[tg, Natureshadow] Fix errorlevel of ‘.’ (“dot” special builtin) when the sourced script does not run any commands, for POSIX compliance
[tg] Refactor op tokens and edchars to shave off some more bytes
[tg] Fix some bugs in the manpage and some occasional/minor code bugs
[tg, Brian Callahan] Mark tests requiring new perl as !need-pass
[tg, slagtc] Add $KSH_MATCH and, to make it usable, ${foo@/bar/baz}
[tg, Score_Under] Fix bogus patch from OpenBSD: only NULL the global source in unwind when actually reclaiming its Area
[izabera] Mention in the manpage that integer bases go up to 36
[Natureshadow] Fix /= operator broken during refactoring

R52c is a bugfix-only release:

[tg] Shave 200 bytes off .text by revisiting string pooling
[tg, J�rg] Fix manpage for ditroff on Schillix
[tg, wbx] Use sed 1q instead of unportable head(1)
[tg] Implement underrun debugging tool for area-based memory allocator
[tg] Fix history underrun when first interactive command is entered
[tg, bef0rd] Do not misinterpret “${0/}” as “${0//”, fixes segfault
[tg, Stéphane Chazelas] Fix display problems with special parameters
[tg, Stéphane Chazelas] Catch attempt to trim $* and $@ with ?, fixes segfault (Todd Miller did this in 2004 for ${x[*]} already, so just sync)
[Martijn Dekker] Fix “command -p” with -Vv to behave as POSIX requires
[tg, jilles, Oleg Bulatov] Fix recusive parser with active heredocs
[tg] Flush even syntax-failing or interrupted commands to history
[tg, fmunozs] Fix invalid memory access for “'\0'” in arithmetics
[tg] Explicitly reserve SIGEXIT and SIGERR for ksh
[tg, izabera] Catch missing here documents at EOF even under “set -n”
[kre, tg] Document Austin#1015 handling (not considered a violation)
[tg, fmunozs] Fix buffer overread for empty nameref targets
[tg] Fix warnings pointed out by latest Debian gcc-snapshot
[tg, Martijn Dekker] Document upcoming set +o changes
[Martijn Dekker] Expand testsuite for command/whence

Change-Id: Ife475b25acae5a85277757cfe9f14c6575eac0cd

From the release notes:

R52b is a strongly recommended bugfix-only release:

[tg] Recognise ksh93 compiled scripts and LZIP compressed files as binary (i.e. to not run as mksh plaintext script)
[tg] Document that we will implement locale tracking later
[tg] Add EEXIST to failback strerror(3)
[jilles] Make set -C; :>foo race-free
[tg] Don’t use unset in portable build script
[tg] Plug warning on GNU/kFreeBSD, GNU/Hurd
[tg] Document read -a resets the integer base
[J�rg] Fix manpage: time is not a builtin but a reserved word
[J�rg, tg] Make exit (and return) eat -1
[tg] parse “$( (( … ) … ) … )” correctly (LP#1532621), Jan Palus
[tg] reduce memory footprint by free(3)ing more aggressively
[tg] fix buffer overrun (LP#1533394), bugreport by izabera
[tg] correctly handle nested ADELIM parsing (LP#1453827), Teckids
[tg] permit “read -A/-a arr[idx]” as long as only one element is read; fix corruption of array indicēs with this construct (LP#1533396), izabera
[tg] Sanitise OS-provided signal number in even more places
[tg] As requested by J�rg, be clear manpage advice is for mksh
[tg] Revert (as it was a regression) POSIX bugfix from R52/2005 related to accent gravis-style command substitution until POSIX decides either way
[tg] Handle export et al. after command (Austin#351)
[tg] Catch EPIPE in built-in cat and return as SIGPIPE (LP#1532621)
[tg] Fix errno in print/echo builtin; optimise that and unbksl
[tg] Update documentation, point out POSIX violation (Austin#1015)

R52 is a strongly recommended bugfix release:

[_0bitcount] Move moving external link from mksh(1) to the #ksh channel homepage linked therein
[tg] Make setenv “set -u”-safe and fix when invoked with no args
[tg] Make “typeset -f” output reentrant if name is a reserved word
[oksh] Zero-pad seconds in “time” output to align columns
[tg] Check signals and errorlevels from OS to be within bounds
[komh, tg] Quote and document ‘;’ as PATH separator in some places
[oksh, tg] Simplify code to call afree() even if arg is NULL
[tg] Fix tree-printing and reentrancy of multiple here documents
[tg] Work around LP#1030581 by permitting exactly one space after
[tg, oksh] Code quality work, cleanups
[tg] New code for here documents/strings with several bugfixes
[tg] Stop using issetugid(2) for ±p checks, wrong tool for the job
[tg] Reintroduce some -o posix changes lost in 2005, plus fixes
[tg] Make “source” into a built-in command
[tg] Drop “stop” alias, lksh(1) functionality to auto-unalias
[tg] Fix \u0000 ignored in $'…' and print
[tg] Improve portability of Build.sh
[Jilles Tjoelker] Improve portability of testsuite
[tg] Fix tilde expansion for some substitutions (izabera, Chet, Geoff)
[tg] Improve reparsing of ((…) |…) as ( (…) |…)
[Martijn Dekker] Fix test(1) not returning evaluation errors
[tg] Fix ${*:+x} constructs (carstenh)
[tg] Make (( … )) into a compound command (ormaaj)
[tg] Repair a few parameter substitution expansion mistakes

Change-Id: I3d875b5c648c60cc6203c964f11d8cbd42648d9c

Bug: http://b/22413333
Change-Id: Id86d5bf463af540b84d03a7b8fe092d23702f414