‣ I was convinced by several that more magic is never the solution
• fix a comment: function cat already had precedence
• change cat loader to look for existence, FPATH included, before
ditching the builtin; note that in manpage
plus plugs an ancient memory leak (hist_execute afree’s its arg now);
also partial revert of commitid 78014291f06497b3 as COMPLEX_HISTORY handles
multi-line commands correctly now (r1.4, 2005-05-23)
in commitid 1004D8283F068C41C3C was bogus; it fixed Jb_boin’s issue
but izabers’s 「var=foo; echo "${var/*/x}"」 was broken; in fact we
only want to not do the looping for // if the pattern matches much.
Also, fix a spelling mistake in the manpage and change some wording
to also work with associative arrays (in the future; no change).
• support NSIG_MAX from http://austingroupbugs.net/view.php?id=741
and make a TODO for later to use sysconf(_SC_NSIG) at runtime
• bounds-check signals (e.g. no smaller than 1, but smaller than NSIG)
• align trap errorlevel with other shells
• make trap match what’s in POSIX and fixup the manpage
• refactor some code related to signals
• hide from kill builtin both EXIT and ERR pseudo-signals
• ord() new, From: Daniel Richard G. <skunk@iSKUNK.ORG>
‣ used in some places
• (c - '0') → ksh_numdig(c) # may take *x++ argument
• (c - 'A') → ksh_numuc(c) # may NOT take *x+= argument
‣ idem for ksh_numlc(c) and 'a'
‣ these need changing for EBCDIC
‣ add testsuite for this
• use macros more, they exist already often
• use digits_lc[foo] instead of ('0' + foo), especially for letters
• caught another ksh_eq case…
• also caught a maybe-UB overflow check, but we don’t have TIME_T_MAX ☹
GNU C11 (Debian 20150413-1) version 5.0.1 20150413 (prerelease) [gcc-5-branch revision 222050] (x86_64-linux-gnu)
including ASan (testsuite) and Valgrind (short)
for mksh but not lksh; bump to R51-alpha.
While here, tweak build scripts a bit, fixup MirBSD-specific Makefile
things, remove part of a comment that’s uninteresting.
‣ : → \:
‣ alias → \alias
⇒ except in some internally used cases, where we use \builtin alias
‣ command . → \command .
• protect Korn Shell builtins from aliases and functions, e.g:
‣ typeset → \builtin typeset
⇒ also unravels the “local” alias used
‣ print → \builtin print
• protect internally-used things from aliases
‣ “let]” is not a valid function name
‣ “set” is POSIX so we don’t expect anyone to override it in a function
• use “command -v” instead of “whence -p” (“which”) in most
places; thanks izabera from #ed on IRC for pointing out
that “command -v” is POSIX – except, “whence -p” a̲l̲w̲a̲y̲s̲ looks
for an executable and shows its full pathname; “command -v”
also resolves to aliases, functions and builtins, so only use
it where it makes any sense (both never output to stderr)
• make most of dot.mkshrc work in the face of such aliases
‣ “ulimit -c” is used; this is not POSIX, and not portable;
maybe we should make ulimit accept-and-ignore the most
common limits even if the OS doesn’t use them?
• update list of builtin aliases in the manpage
handle unknown bases as ksh93 does: larger downgrade to 10
(although our max will stay 36, as ksh93 doesn’t have upper/lowecase)
and smaller downgrade for typeset -i, but not for arithmetics
• all: bump version to R50-current; add more comments; whitespace
• all: remove all mkssert(); we’ll do full re-runs of scan-build and,
hopefully, Coverity Scan/Prevent
• check.t: fix a testcase (sed could exit false, but we don’t care)
• eval.c: fix tilde_ok data type (only unsigned may shl constantly)
• exec.c: fix shebang buf array accesses to always go via uint8_t *
another showing deeper problems (probably LP#1381993 “non-list contexts” or
the IFS_WS/IFS_IWS story, perhaps *all* IFS_WS (not just ternaries) really
should be IFS_IWS instead?)
• permit interrupts during a write(2) loop in the cat builtin, too,
not just in the read(2) loop – fixes inability to kill a clogged
output cat
• kill the cat when smores finish
TODO: revisit this ⓐ in more depth, ⓑ for other functions, such as
“hd”, and ⓒ test on AOSP as well
calls in Build.sh, we need HOSTCC for that… which we should do, using BER
or something encoded for integers, and pregenerated hashtables as planned)
also, bump to R50 beta, due to today’s language changes
use errorf() while nameref states were being changed (by almost completely
eliminating the global variable) and the readonly first array variable
bypass (typo/refactoro); also, whitespace, one int → bool, and add a
comment wrt. the parser rewrite talked about with igli during a fever ;)
‣ not like oksh did, but using mksh’s built-in features
• handle suggested __pure additions
• revert cid 1004F7F096867C83CF0
‣ always use our wcwidth code
‣ only use our strlcpy code if none found
• fix a couple of gcc-snapshot and clang/scan-build warnings
• mksh R49~rc1
one set of CTRL, UNCTRL, and new ISCTRL macros) C0 and DEL handling; the
optimisation only works for 7-bit ASCII, so those places 8-bit must pass
intact have an added check
also, while here, remove an editor oops (‘;’), oksh rcsid sync (they did
accept I was right wrt. set -e), int → bool, and code merge/cleanup
Add a proper suspend builtin that saves/restores the tty and pgrp
as needed instead of an alias that just sends SIGSTOP. Login shells
may be suspended if they are not running in an orphan process group.
TODO: I am seriously considering following Chet and changing
the way this works, by explicitly dropping privs unless the
shell is run with -p. Every other shell does it like mksh,
except Heirloom sh, which on the other hand doesn’t know any
explicit set -p or set +p (though it doesn’t know set +foo
for any foo either).
┌──┤ QUESTION: Do we need the ability to do this:
│ tg@blau:~ $ ./suidmksh -p -c 'whoami; set +p; whoami'
│ root
│ tg
If not, I’m seriously considering to drop set ±p as well,
only parse -p on the command line, with +p being the default,
and dropping FPRIVILEGED.
Thanks to RT for noticing and jilles for initial follow-up
discussion, as well as Chet Ramey for doing the sane/secure
thing instead of following Debian.
Steffen Daode Nurpmeso stumbled upon it and gave very detailed
instructions on how to reproduce it (thanks!); fix that
also only call x_bs0 if xcp < xep because *xep is undefined
prompt display routines; make Emacs and Vi share
code, permitting reducing of duplication and code
removal as well as more consistent behaviour; put
some common code into shared helper functions, too
• New x_adjust() logic (Emacs mode): when determining
what portion of a line to render use a much more
sophisticated algorithm and try to fill up ⅔ of the
total screen width (with line and prompt both) also
as wished from Steffen Daode Nurpmeso
Whenever the SIGEXIT trap was set (to anything, really)
syntax errors and interruptions were not ignored any more
in an interactive shell (where they should be, unless
set -e is used); fix that.
tbd: traps should probably only be marked as pending
and run for LLEAVE/E_NONE
merged:
• new regression tests
• check.pl (tests/th) better tmpfile handling
• exec.c 1.50: POSIX specifies that for an AND/OR list,
only the last command's exit status matters for "set -e"
• ksh.1 1.147: document the above
• eval.c 1.39: “Make $(< /nonexistent) have the same behaviour
as $(cat /nonexistent) wrt. errors (do not unwind and do not
treat this as fatal if set -e is used).”
‣ additionally make shf_open() return errno and actually show
the error message from the system
• regression-39 test: remove the “maybe” marker
‣ but decide on correct POSIX behaviour
already been fixed in mksh:
• check.pl (tests/th) exit 1 if tests fail
not merged:
• main.c 1.52: run traps in unwind() before exiting;
I’m pretty sure this is already working as-should in mksh
• eval.c 1.38: “Commands executed via `foo` or $( bar ) should
not inherit "set -e" status.” As discussed in IRC, this is
just plainly WRONG.
• sig_t detection was a bit insane, it is a function-pointer type after all
• fix uninitialised variable in c_select which led to mistakenly accepting
invalid (nōn-numeric) input and acting, randomly, upon it
• keep SIGCHLD blocked in child after forking longer, for job list manip
• block SIGCHLD ifdef DEBUG_LEAKS to not run job foo during/after afreeall
• fix annoying ISO C90 vs. C99 (un)signed constant warning
command; setting one still unsets the other at first)
• Change subst_exstat to be conformant unless -o sh is set and -o posix isn’t
• In lksh, make subst_exstat (newly) conformant if -o posix
• New MKSH_BINSHPOSIX to accompany MKSH_BINSHREDUCED
• Sync lksh manpage precisely
• make parsing numbers with leading digit-zero as octal independent of
mksh/lksh and dependent on set -o posix; adjust manpages to match
• warn about these changes and why mksh uses 32-bit consistent arithmetics
and point people to lksh for host-long undefined-behaviour arithmetics
• point out, explicitly, that it is *legal* for the operating environment
to make 'print $((2147483647 + 1))' (on a 32-bit system; adjust for a
64-bit system) to run 'rm -rf ~ /' instead
• correct order of built-in commands; use POSIX special versus “all others”
plus “keeps assignments” as distinction, no longer play POSIX regular vs.
others game; sync manpage
• fix LP#1156707: map (( internally to “let]” which is no valid function
name and so can’t be overridden but is unlikely to be used otherwhere
and not strictly permitted (by POSIX) anyway
• we do not need -Wno-overflow any more, either
• bump to R45
is larger than the positive range of the latter (implementation-defined), so
avoid them in all explicit cases and rearrange stuff and check for it
(I’m gonna have to revise lots more code…)
tg