Upstream changes:
R50f is a required security and bugfix release:
[tg] Add a patch marker for vendor patch versioning to mksh.1
[tg] SECURITY: make unset HISTFILE actually work
[tg] Document some more issues with the current history code
[tg] Remove some unused code
[tg] RCSID-only sync with OpenBSD, for bogus and irrelevant changes
[tg] Also disable field splitting for alias 'local=\typeset'
[tg] Fix read -n-1 to not be identical to read -N-1
[tg] Several fixes and improvements to lksh(1) and mksh(1) manpages
[tg] More code (int size_t), comment and testsuite fixes
[tg] Make dot.mkshrc more robust (LP#1441853)
[tg] Fix issues with IFS='\' read, found by edualbus
[enh, tg] Fix integer overflows related to file descriptor parsing, found by Pawel Wylecial (LP#1440685); reduce memory usage for I/O redirs
[tg] Document in the manpage how to set ±U according to the current locale settings via LANG/LC_* parameters (cf. Debian #782225)
[igli, tg] Some code cleanup and restructuring
[tg, oksh] Handle number parsing and storing more carefully
R50e is a required bugfix release:
[tg] Add more tests detailing behaviour difference from GNU bash
[tg] Introduce a memory leak for x=<< fixing use of freed memory instead, bug tracked as LP#1380389 still live
[tg] Add x+=<< parallel to x=<<
[tg, ormaaj, jilles] POSIX "command" loses builtin special-ness
[tg] Fix LP#1381965 and LP#1381993 (more field splitting)
[jilles] Update location of FreeBSD testsuite for test(1)
[Martin Natano] Remove dead NULL elements from Emacs keybindings
[tg, Stéphane Chazelas, Geoff Clare] Change several testcases for $*/$@ expansion with/without quotes to expected-fail, with even more to come ☹
[tg] Fix miscalculating required memory for encoding the double-quoted parts of a here document or here string delimiter, leading to a buffer overflow; discovered by zacts from IRC
[RT] Rename a function conflicting with a MacRelix system header
[tg] Use size_t (and ssize_t) consistently, stop using ptrdiff_t; fixes some arithmetics and S/390 bugs
[tg] Remove old workarounds for Clang 3.2 scan-build
[tg] Remove all Clang/Coverity assertions, making room for new checks
[tg] Fix NSIG generation on Debian sid gcc-snapshot
[tg] Make a testcase not fail in a corner case
[tg] Fix issues detected by GCC’s new sanitisers: data type of a value to be shifted constantly must be unsigned (what not, in C…); shebang check array accesses are always unsigned char
[tg] Be even more explicit wrt. POSIX in the manpage
[tg] Fix shebang / file magic decoding
[tg] More int → bool conversion
[tg] Let Build.sh be run by GNU bash 1.12.1 (Slackware 1.01)
[Stéphane Chazelas, tg] Fix here string parsing issue
[tg] Point out more future changes in the manpage
[tg] Call setgid(2), setegid(2), setuid(2) before seteuid(2)
[tg] Fix spurious empty line after ENOENT “whence -v”, found by Ypnose
[tg] Optimise dot.mkshrc and modernise it a bit
[tg] Use MAXPATHLEN from <sys/param.h> for PATH_MAX fallback
[tg] Some code cleanup and warnings fixes
[tg] Add options -a argv0 and -c to exec
[jsg] Prevent use-after-free when hitting multiple errors unwinding
[tg] Fix use of $* and $@ in scalar context: within [[ … ]] and after case (spotted by Stéphane Chazelas) and in here documents (spotted by tg@); fix here document expansion
[tg] Unbreak when $@ shares double quotes with others
[tg] Fix set -x in PS4 expansion infinite loop
079d536444