JehanneOS
/
mksh
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity
1,686 Commits 2 Branches 0 Tags 6.4 MiB
Commit Graph

1686 Commits

Author SHA1 Message Date
tg d97691ca61 by the principle of the working tree, we cannot enable aalloc as of now. 
damn, pointers in C are so uselessly complicated… in asm it’s much easier
 2008-11-17 01:14:57 +00:00
tg 90e8e7ea72 rewrite the cookie logic… while this is better than previously, 
we still have issues:

(gdb) r
Starting program: /usr/obj/bin/mksh/mksh -c print\ \$KSH_VERSION
mksh in free(): error: modified (chunk-) pointer

Program received signal SIGABRT, Aborted.
 2008-11-15 11:42:18 +00:00
tg 1ca9790f1d typo 2008-11-15 09:05:29 +00:00
tg 546d5d0def give hints to anew() according to aalloc stats output for testsuite and a 
little interactive use, to reduce the number realloc(3) must be called
 2008-11-15 09:00:19 +00:00
tg 6be05862db meet AALLOC_STATS 2008-11-15 08:52:01 +00:00
tg 6b9cdbbfd6 globalise internal_verrorf() 2008-11-15 08:42:36 +00:00
tg cd716426e4 introduce AALLOC_STATS, which somewhat breaks the encapsulation for need 
of a friendly name
 2008-11-15 08:03:24 +00:00
tg d8bc1cf161 thread-safety in the atexit function, justin case 2008-11-15 07:59:46 +00:00
tg 24a23d52d8 add plugs for thread-safety, requested by Jonathan Schleifer 
example:
#define AALLOC_THREAD_ENTER(ap)	@synchronized(aalloc_guard) {
#define AALLOC_THREAD_LEAVE(ap)	}

notes:
• you must declare and define eventual guard objects yourself
• while there will not be any direct “return” calls inside the
  locked area, AALLOC_ABORT and AALLOC_WARN may be called
• AALLOC_RANDOM does not need to be thread-safe, but malloc
  and free should be; if “ap” is used for locking, all functions
  called must be thread-safe though
 2008-11-15 07:51:53 +00:00
tg 32e1ecf5b3 enable passing of a hint how many pointers we’ll need to anew() 
if 0, the default is used
 2008-11-15 07:35:25 +00:00
tg 03acb8ce51 this was NOT intended to be committed; my sincere apologies 2008-11-13 00:36:07 +00:00
tg 4e975aa081 make silently leaking a define for better integration into other programmes 2008-11-12 23:34:02 +00:00
tg e0853c5908 it is, apparently, not portably guaranteed that realloc can deal with NULL 2008-11-12 19:23:09 +00:00
tg 255c77ff32 use -DAALLOC_TRACK in MirOS-current but not releases, similar to 
-DMKSH_AFREE_DEBUG, which is even similar in functionality

cost 412 code, 32 bss
 2008-11-12 07:38:42 +00:00
tg d099b00b53 leave at least a working tree, with zero-penalty -DAALLOC_NO_COOKIES 
XXX cookies are still broken?

cost for aalloc.c: data -= (4, 0, 4, 0); text += (1665, ?, 2115, 2217)
 2008-11-12 07:36:19 +00:00
tg 079e86501a fix the cookie checks 2008-11-12 07:02:47 +00:00
tg 518d37b6bf same fuckup as in commitid 100491A793F1CA8D9CA, different place 2008-11-12 06:58:49 +00:00
tg 1cf58ac17e more silently 2008-11-12 06:44:04 +00:00
tg bfec7328d4 allow mksh to leak silently 2008-11-12 06:42:22 +00:00
tg 28a0e28b1d adelete is supposed to be quiet on valid pointers 2008-11-12 06:38:08 +00:00
tg 3c07bc66f0 fix up the fuckage in afree() 2008-11-12 06:35:27 +00:00
tg c6bcbc7bb0 improve error reporting 2008-11-12 06:14:26 +00:00
tg 364a7d116a mergo 2008-11-12 06:12:56 +00:00
tg 548c86147e more careful about disabling cookies 2008-11-12 06:10:51 +00:00
tg 4d62063eca and fix another set of CPPFLAGS for debugging 2008-11-12 06:05:54 +00:00
tg 42b789fc04 no %t and %z specifiers in shf_vfprintf ☹ so use %p and %l instead 2008-11-12 05:55:43 +00:00
tg c113091e2d forgot to charge for the extra size in the abstraction 2008-11-12 05:46:45 +00:00
tg b08956d6a7 raise AALLOC_INITSZ to be less expensive 2008-11-12 05:46:14 +00:00
tg 18b38ac5c7 if the address of bp changes, all backpointers must be adjusted as well ☹ 
pretty expensive
 2008-11-12 05:45:28 +00:00
tg 7f17e3fc91 temporarily disable cookies for debugging 2008-11-12 05:40:23 +00:00
tg b1673ff1a3 with other combinations too 2008-11-12 05:34:20 +00:00
tg c3e17d8f0b get the debugging methods to build 2008-11-12 05:32:34 +00:00
tg 1211cf79cc • bp->last == bp->endp is allowed 
• do the realloc properly
 2008-11-12 05:27:01 +00:00
tg 4b456b1457 __CRAZY cleanliness: pointer arithmetic 2008-11-12 05:11:05 +00:00
tg 7cbc0b6fa2 this one I could have seen by myself if not for the late time of the day 2008-11-12 05:05:17 +00:00
tg e56b1f628e alloc.c is now not always included any more, aalloc.c is MirOS licenced 2008-11-12 05:00:40 +00:00
tg d15671f6f6 fix first batch of compile warnings, enable aalloc for mirbsd 2008-11-12 04:59:42 +00:00
tg 3c1e46ee4d prepare for aalloc.c, which I have just written myself, as an area-based 
allocator using malloc and free, with mmap malloc and omalloc in mind,
not counterfeiting its security measures such as guard pages, and having
some of our own, e.g. XOR random cookies, optional mprotect, etc.

zero cost (for we have arc4random())
 2008-11-12 04:55:19 +00:00
tg 8d5d720f08 get rid of tempting sizeofN macro 2008-11-12 00:55:32 +00:00
tg 4d3fc2413e switch to a (nmemb,size) API from a (nmemb*size) API 
cost: text += (308, 256, 4, -100)
 2008-11-12 00:54:52 +00:00
tg 246b762af7 isolate all knowledge of the area-based allocator from the rest of the code 
cost: bss -= (0, 0, 16, 16); text += (520, 504, 516, 480)
[ gcc,pcc X full,small ]
 2008-11-12 00:27:57 +00:00
tg c80c28633b change use of “Area *” to “PArea” and “struct Area” to “TArea” 
no change in size (mgcc and pcc, small and full)
 2008-11-11 23:50:31 +00:00
tg 6a4afdde05 add the workaround for Debian #492377 into the main mirtoconf function, 
because Hanno Böck has something similar triggered on Gentoo as well
 2008-11-11 23:20:08 +00:00
tg f4f4f64492 if ‘-DMKSH_ASSUME_UTF8’, default HAVE_SETLOCALE_CTYPE to 0 since it’s 
unused then anyway, to save two followup mirtoconf checks and headers
 2008-11-11 21:52:51 +00:00
tg 6c02c37e52 add new Build.sh option -llvm for bytecode compilation and link-time 
intra-module optimisation
 2008-11-10 20:25:04 +00:00
tg 2a0e181a70 work around even more format string warning stuff 2008-11-10 19:33:08 +00:00
tg 65b1923b61 build warning-free with LLVM+Clang on MirBSD itself 
$ (CCC_LD=mgcc CC=ccc sh Build.sh -r && ./test.sh -v) 2>&1 | tee log
Total failed: 2 (as expected)
Total passed: 278

Just the result is huge, and we could of course build to intermediate
byte code to optimise globally…
 2008-11-09 20:32:18 +00:00
tg c6e2845d63 remove dead nested assignment, result from first llvm+clang scan-build 
run native on mirbsd
 2008-11-09 19:48:02 +00:00
tg 446e019c07 the QNX /bin/ed problem is worse than thought; use a new mechanism, 
because categories in check.t are OR’d:
• no-stderr-ed disables the newish-ed tests (tried using testcase)
• stdout-ed enables the oldish-ed tests (variable, if the above
  testcase succeeds, it’s added, but QNX overrides the variable)
 2008-11-08 17:36:35 +00:00
tg 5206ddcc65 much simplify and shorten default $PS1 (uses even more magick now) 2008-11-07 20:07:10 +00:00