For version 0.16.*. > Sorry forgot to flip this back: perror and strerror are back to using the /lib/liberror.txt file for all 8086 Elks modes. Only i386 has the file linked. Elksemu now traps accesses to this file and substitutes a fake. > Added vararg macros to new preprocessor. > Builtin CPP removed for __BCC__ compile, bcc-cc1 fit in 16bit but crashes. Thirty two bit works fine, gcc or bcc. (almost any 32bit.) > Added __BCC_VERSION__ macro, defined by bcc.c contains hex version no. > Bugfix for offsets generated by #asm when using -O option to bcc-cc1. Extra compile option ASM_USES_CALLEE_REGS. > Extra checking and bugfix for mixing char values with long shifts. > WARNING to distributions: I've altered the install scripts and paths you will want to check them. Altered paths for bcc.c and normal installs, defaults should now be good for distributions where bcc is a cross compiler. If you set the PREFIX to / it'll install to suggested Native paths. As86_encap moved to LIBDIR. > I've altered 'perror' and 'strerror' to be normal when compiled for everything except libc_f.a. Only with that library will it look for the liberror.txt file and it now looks in "/lib/liberror.txt" only. This means there's no need to install it anywhere except ELKS itself. > Change -Mg option to use uclibc as glibc2 seems to have stopped working. > ar.c switch to using strerror. > Move elksemu, it's used like a shared library but is best treated like an emulator so it's now installed in DISTBIN. Also on Linux-386 it's compiled using the system compiler. > Use the POSIX -R argument to cp rather than the normal standard -r because some new systems have decided to stop supporting their '-r' option. > Oops, the constant error strings in the assembler were char pointers not char arrays, small waste of space. > Allow for Linux-2.6 32bit dev_t. > More changes so it compiles better on 'other' machines. Removed some GNU-Make'isms in the top header file. > It seems that CYGWin's problem with _P is inherited from BSD. It also seems most of my problems come from BSD recently, haven't they realised that they down own the 'Unix Standard' anymore and haven't since SYSV got popular. > Some changes for cygwin, it isn't quite as nasty as compiling for minix but it doesn't miss by much! > Compiling for the 'tcc' complier, nice and easy. > Hitting bcc.c again, bcc-cpp now the default. > Some bugfixes for DEC Alpha -- 64 bit longs! Constant folding still done with 64bit longs though. > The fopen function (and friends) is now a real function not a macro. > Ctype.h updated to ansi. > Libraries appear to compile properly with -O and -ansi now. > Copt's hash string table adjusted; should now have fewer pathological cases. > Order of copt's rule processing reversed; it now works from the top of a rule file to the bottom. Also comment lines may be include in the match lines. !BCC_EOS added to bcc-cc1 for 'end of statment'. > bcc-cc1's asm("") now works in declare mode too. > New independed cpp added, has both K&R and Ansi modes. Changes made to the embedded cpp too to try and fix some bad bits. > Conditional assembler added around push and pop of 'si' and 'di' if the optimiser (and as86's optimiser) are to be run. > Signed keyword added to bcc-cc1 for 'signed char' also works with other int types but is usually just a noiseword. > As86 macro syntax altered, you now don't have to include any brackets. > Gcc warnings in ar86 cleaned up. > Mk_dist now makes incremental patch files too. > Update magic file. > Various updates to the bios libc, the vt52 and ansi emulations are now seperated (and can be both included if needed) and don't have to be linked. Conio functions work properly in bios and msdos. > Added -v (version) > as86 and ld86 now delete their binary outputs if there were errors. > Added more 'set' commands generated by '#asm' lines. The new ones begin with '.' rather than '_' an are based off the 'bp' register rather than the 'sp' register. > Various unused, test and old document files removed. > as86 modified to use normal malloc routines - no longer has limited input file size. > as6809 repaired, appears to work. > as86 error display adjusted, some errors demoted to warnings. > Complete re-write of bcc.c driver program. Should now be easier to modify, does have more versatility so that things that were formerly compile time options are all run time. > #asm and #endasm now interact with statment processing inside functions. If statments and case labels no longer need special layout however #asm must be placed between statments now. > The __heap_top virtual constant added, allows the -H option to be used with the -d option in the linker. > Conio routines added, merged with bios_getc/bios_putc routines which now no longer exist under those names. > monitor.out now copes with very old zimages, like the memtest86 program, and has 'Retry' prompt for disk errors. > msdos.s and tarboot.s loading of non-a.out executables adjusted for dos BIN like files. > Standalone library startup adjusted to allow execution of a '-d' executable from DOS, if __argr.x.cflag is set this is running under DOS. > Compile for DOS again, disable cpm86 headers under DOS. > Minor bugfix for as86 binary output re sizeof(int)/K&R problem. For version 0.16.0. > As major bugs seem to be absent 0.16.0 is escaping. > Seperated the multi-pass optimisation from the jump extension. We now have the -j back to it's old meaning and -O added to help shorten instructions. > Fix for nested conditionals in the assembler. > Fix for listing to unusual locations; assembler tried to list non-existant lines. > Removed lots of warning from the ar86 code. (__STDC__ flags too) > Stop trying to strip scripts on install. > Remove non-working versions of malloc and stdio from libc directories. > Lots of changes in the 'bootblocks' example code. > Updated 'make distribution' to be all in the distribution. > bin86 Makefile adjusted so that the as86 script isn't installed with -s. For version 0.15.0. > Added support for archives to nm86/objdump86 and size86. > ar86 added. Used for Linux because RedHat is broken. > Bugfixes on utmpent and friends. > A bit more in the manpages on how to use the assembler etc. > Bug fixes in stdio, fp->mode cached too much, buffer type mixup. > Altered initial make variables, you only need to modify 'PREFIX=' to allow installs into /usr/local. > The BIOS library has a simple MSDOS fs file reader now, based on the code from monitor.out. > Added bin86-0.... source distribution archive. > Fixed bug in -t option to as/bcc (for moving the text segment) > Lines that have a unexpected character are now listed by the assembler not just ignored. > Added assembly externals in as86_encap generated code. > I appear to have bcc generating a consistant set of code fragments for floating point operations. No floating libs though ... > Lots of bootblocks changes including Added lsys.com to install the dosfs boot sector under dos. monitor.out now (finally!) loads zImage files. Monitor.out now works with a TAR "filesystem" on a floppy. Added two compile time options (-DTARFLOPPY and -DDOSFLOPPY) for smaller executables to only boot linux-386 from a floppy. Makeboot can install the mbr. Removed the ELKS specific code from the minixfs loader, added a helper program to load and relocate ELKS. One config file for booting a [b]zImage. DOSFS boot can boot any named file. > Fixed/avoided some problems with the optimiser breaking hand optimised code. > I've made some alterations to the header files in an effort to merge the ELKS kernel files with fewer problems. > Oops, Claudio says that qsort went the wrong way, hmm. > Added some more library code for MSDOS programs, there's still more to go. > Removed the 'libc-8086' distribution tgz file. For version 0.14.0. > Objdump86 utility added, also includes nm86 and size86 that work on as86's object files and elks executables. > Added ability to put ELKSSRC=/././. on initial make command line. > Some bug fixes and buffering improvements for stdio2. > Changed output of linker -m option to be more verbose (useful). > Linker errors are now directed to STDERR on unix/elks. > Alterations to elksemu and the C library to be closer to elks, 0.13.1+ should not need a patch to work with 0.0.60+ ELKS. > Link symbol dump now flags local variables differently to exported. > Fixes for libc/misc/crypt.c, they do change the values of the encrypted password so beware. > as86_encap now uses a static array and #defines. > Added -t flag to as86 and bcc to move the text segment for this object into another one. > Moved segments 4..14 so that instead of being attached above the data segment they now sit between the text and data segments in the final a.out, this is where they'd be needed for medium model. > Added -x option to bcc to prevent it linking in crt0.o. > Added various improvemnts to the optimiser, I don't think it'll generate illegal code now for "-O", "-O1", or "-Oi". The last inlines the functions __get_es(), __seg_es(), __poke_es(), __peek_es() etc. > Code generator improvement for integer multiply and shifts. (Including making it far less keen to do multiple shifts instead of a multiply. > Code generator fix to remove incorrect (but accepted by the assembler) addressing modes like: mov ax,#_ar[bx] > Been messing with boot block stuff again, still can't get it to load zImages (bzImages ok tho). Minix.c can't load an ELKS kernel either, it's looking like there isn't actually enough room :-( > Fix to elksemu for breakages caused by glibc. > Other fixes to elksemu for execve(), dup(), readdir() etc functions. For version 0.13.0. > Removed too eager 'typeconv.c' warning. > Well, I just broken everything! To be exact if Dev86 sees a directory called ${ELKSSRC:-/usr/src/elks}/include it'll assume that is the equlivent of libc/kinclude and use that in preference. I've sent off a patch so that elks-0.0.* should be compatible but meantime there are bound to be lots of problems caused by shuffling of include files. (Later: Actually doesn't seem too bad) > Added '-g' to gcc options, make sure install has '-s' to strip symbols. > Grrrr, missing files in bootblocks directory, I didn't have them all. OK, as part of the dist script I'll have to do a 'make other' ... Some other bits there too ... > Minor alteration to opening files in ld to fix problem when share.exe loaded under dos. > Added Config.dflt for default libc configuration flags. I can now re-introduce Joel's standard I/O functions ... but they haven't been updated in a while ... > Addition to minix.c to use allow BB to determine the track size by trial and error (not intuit from superblock) to allow mis-sized filesystems on a floppy. Also add some error retries to the floppy version. > Squeeze the minix BB a bit more, DOTS is now working again. > Can now do 'use16 8086' in assembler instead of 'use16 86' also 'use16 80386' > Add 'make install-ln', this installs _symbolic_links_ to the dev directory in the normal directories. This means you don't have to do a make install after re-compiling. But it also means the 'install' is dependent on the dev dirs. > Add ability to use overrides on the command line of the initial make, and have them saved, for CC, CFLAGS, LDFLAGS, PREFIX, BINDIR and LIBDIR. Includes minor alteration to bcc.c and as86_encap. > Added in 'objchop' as an option in the ld directory. > The cpu type warnings were worrying some people so I've set the initial value of the '-w' flag to off, use -w- to turn them back on. The -W flag to bcc now does this too. > Added the ELKS_SETUP from Ken Yap to minix.c. > Added __ELKS__, __8086__ and __i386__ #defines for relevent bcc configs. > Clarified some of the copyrights, esp for Bruce's stuff. > Added a few changes to clean up compiles on an AIX/RS6000. It still needs gnu-make installed and is only tested with gcc though. (Because of IBM's recent contribution to increasing the popularity of GCC) For version 0.12.0. > Notice the change in the version numbers, as long as I'm releasing the upto the minute patches I thought I'd better give everyone a way of keeping track. So now the 0.X.0 are the main releases, the 0.X.Y are interim patches. > I've added a new output binary for ld86, it can now do the broken relocatable a.out that bin86 produces. I don't like this but it does let me move ld86 back to /usr/bin without breaking DOSEMU. Note! If you've installed vsn 0.0.11 or earlier you will need to uninstall it before installing this one. > Altered makefiles for GNU-make 3.74, GNU-make no longer generates automatic 'rm' commands so I have to do it manually. > Altered makefiles again, starting to get non GNU makefile working. > Added some cleanups to ifdef.c. (voids etc) > Assembler now understands monadic '-' and '+' operations (for copt). > Improved the type conversion functions, they should now work on most machines. There may still be some problems with sizeof(long) == 8 tho, added some warnings/traps. > Improved the auto-detection and auto-rejection of ld is more likely to compile even if a.out.h is completely incompatible (leaving out native support). > Fix for stdio, there was a problem if the OS only did a partial write. > Alter align() macros to use ssize_t iff both __STDC__ and _POSIX_SOURCE are defined. > Add some changes for compiling under AIX. Assember and BCC appear to work, ld has problems with AIX's 'ar' command - problem: AIX's make and gmake won't like the smell of a standard following ar86 command ... hmmm. > Changes in bootblocks, minix.c can now be compiled for a hard disk (works on dosemu, dunno about the real thing tho), fixes for mbr.s re preboot facility. Addition to minix.c to fix number of sectors in bpb for floppy. For version 0.0.11. > New, most recent, version of unproto inserted. > AAARG! First_arg_in_ax breaks stdarg.h! As a consequence printf wasn't working, now why didn't _you_ notice! OK. printf now uses varargs, that is more likely to work and does for printf & scanf. > Added -Mc for 'caller saves' _without_ first_arg_in_ax. As this uses the normal libc there are likely to be problems with library -> user callbacks. > Added -Ml for 386 'Large linux' compile. > Added -Mn for 'normal' elks compiles so the default for dos can be -Md > Brand new, an optimiser has been added, it's in its early stages yet the 8086 (plain -O) version seems ok. The 386 generator (-O3) is definitly missing some requirements. It's probably not a good idea to use -Ml and -O. (Later: The 8086 can generate bad code it seems, oh well) > Some changes to the makefile, minix make will now _try_ to compile bcc, it doesn't seem to succeed yet though. Also because of this I've gone back over the headers, should be better on non-ansi or non-posix systems now. BTW: minix make has no chance with compiling libc so I've added another zip archive containing just the compiled libcs. > Add the -a-, -j-, -g-, -u- and -w- flags to as86; specifically for -w- to allow easy enabling of assembler warnings while using bcc. > Bugfix for assembler 'ifc' pseudo op. > Bugfix for preprocessor and unterminated last line of included file. > Add minix syscall detector to elksemu, I may make it into an emulator ... > I've split of the 8086 ELKS and i386 Linux libc stuff into seperate directories. I expect Chad will want to take over the 8086 one for a while. Plus some other files moved around to help new users. For version 0.0.10. > Some serious changes to the makefiles, practically a rewrite of them. 'make', 'make all', 'make install', 'make clean' and 'make realclean' do the expected things. 'make distribution' should let you create the partial and binary dirtributions from the full source. > The makefiles now allow multiple libc binaries to co-exist. > The makefiles now allow a _full_ compile without installing anything. > In part because of the makefile changes a lot of files in the library have been moved, this is a large portion of the patch size. > ifdef.c added as part of makefile changes, I'm not quite sure if I like this way of doing things but it's ok for the mo. > MSDOS, in the output of *.bin files from as86 skipped areas have zeros written not semi-random bytes. > MSDOS (+ other machines where sizeof(char*) != sizeof(int)) problem with macros resolved, they appear to work correctly now. > NCC added, this is a different version of bcc with libs and include dirs in the same format as the MSDOS driver. This is used for 'uninstalled' compiling and also if you do: $ ln -s `pwd`/bcc/ncc $HOME/bin/bcc The bcc will use the 'uninstalled' include and library files for when you don't have root access. > -H option added to ld86, this controls the size of the heap or 'chmem' field in the executable. > BOOT-Blocks, makeboot can be compiled as a DOS COM, bootblock for msdos filesystem completed, this will allow lilo, monitor.out or any Linux-86 standalone executable to be executed from an msdos floppy. Makeboot can also install several other types of bootblock. > Monitor.out now includes 'bzimage' command, which will boot a Linux-i386 bzImage with command line and initrd. > 'skip.s' bootblock fixed. > A few more functions now work in 'standalone' mode, malloc() especially. -Rob For version 0.0.9. > The debian stuff has been removed again, Christoph want's to keep contol of the debian specific files. > I've done some changes to the makefiles, it'll now honour CC=XXX on the make command line. > The assembler now works again compiled under bcc, it tried to allocate too much memory. > Major change to the assembler, '-j' not auto-sizes jumps this requires a change to at least 3 passes. Sometimes it'll say it needs another -j because lengthing jumps cause others to fail, two -j's will give it lots of passes :-) > The second large change is an effort to identify when instructions are not right for a specific processor and generate a warning. The use16 and use32 instructions now have an optional argument of a processor type (86,186..686). Also bcc has a '-W' to remove the -w it passes to the assembler. > Altered the way bcc deals with LOCALPREFIX, now under DOS it picks this up based on the path to the executable. > Fixed a bug introduced by the 'cpp in asm' change in 0.0.8, bcc-cc1 wasn't compiling under bcc. > Two new bootblocks in the 'examples' mbr.s is an MBR (no really :-)) msdos.s is unfinished but even this far allows lilo to be used on an MSDOS filesystem floppy. > Changed ld to _check_ the magic on files inside an archive, this cures the problem caused by long names inserted by gnu-ar. Also it notices if an archive has been included in an archive and will link that too. > Changed ld to do a final check that it knows what sort of a.out.h file it's got, if this fails '-N' is disabled. > Ld changed so for 16 bit exes the heap is limited to 32k. > getcwd and execvp added, sash should now work. Also added to the 386-lib except execvp won't run scripts without '#!' > Nasty bug with syscalls of > 3 args in 386 code fixed, none of them would have been working correctly. For version 0.0.8. > Christoph Lameter has provided a patch for installation using the debian packaging, I've probably broken it :-) > Doselks has been added in, it's not that big really. > Triggered by the debian stuff the makefiles can now create a pure binary installation tree, nothing needed in /usr/src. If kinclude is configured out the include directory will reference /usr/src/linuxmt > Some good changes to the linker today, it will generate a Linux-386 (That's right, growed up linux) executable. It's still very fragile but it does work. The '-r' option now works, but unlike bin86-0.3 it creates a _Linux_ OMAGIC file. This means that bcc -G works, the names are mangled by ld86 such that if it was compiled with ELF the OMAGIC will successfully link in an ELF environment. Beware older versions of GCC-binutils can segfault in this situation. > Bit of sneakyness for the assembler, org will accept negative values ... > Made a few changes to idcc, I think I'll translate it to C soon. > Fixed a couple of problems with the preprocessor WRT hashes not at the start of a line and quotes in ifdef'd out code. > Missed a bet in the kernel patch and module, the old one wouldn't execute impure executables it will now. > Fixed matching bug in elksemu, A piece of code was commented out for some reason ... oops. > Some of the error reporting in the linker has improved, if it can't open libc it now says so! > Some 386 specific stuff has been added to libc, mainly the syscall generator, startup stuff and BCC's 386 floating point code. > Change in malloc, Linux-386 limits +ve brk() changes to 32Mb, the malloc will now allow for this ... Phew! grab.c will grab about 3Gb of virtual memory, with 1Gb chunks, when compiled in 386 mode! One oddity, glibc (quite an old one) hangs when trying to free if grab.c is compiled static. > Hunted down the most important 16 bit dependancies in dis88, it seems to generate a reasonable output file now. Plus a couple of fiddles in the disassembler, personally I prefer Hex and I've added a couple of markers for 386 code (use opr32, use adr32, seg fs/gs) and the -f flag to force disassembly of any file. > The i386 compile uncovered a bug in alloca, In fact the move to 386 has been amazingly painless, practically everything so far has worked first time or nearly so ... Wow! > Added minimal dirent to i386-lib, elksemu now compiles under BCC-386 AND WORKS! > Trying to compile bash, it does some hackish stuff with the preprocessor and uncovered a couple of problems, fixed. Now I need more float stuff. > Added for bcc-cc1, code between #asm and #endasm _is_ scanned for #defined words and functions, it didn't effect the libc at all. Of course we now need some way to put multiple instructions in one #define, I've nominated '^' as the statement terminator, I didn't have many choices :-( > Added '-t' for bcc-cc1 when bcc gets -S -Rob For version 0.0.7. I've been making quite a few changes to the assembler: The 'include' or 'get' command now works. The '-b' and '-s' options are now quite useful for bootblocks or encapsulated code. The standard input is now a valid source for the assemble (but only on 'big' machines). Various other functions now work, macros, 'jmp near', 'jmpi qword $100000,$1000', and other bug fixes. As86_encap script added. Some of the 256 char line limits have been removed for non-bcc compiles. The linker will now fail more gracefully for reversed 'org's and complain about too large segments. Bcc-cc1 will now compile under itself, with some limitations, and appears to generate correct output. There are problems tho in that I think the heap<->stack saftey margin is too short and I've disconnected all code for floating point features. The components will all compile and seem to work quite well using MSC 5.1 under dos and will probably compile using other dos compilers. BCC cannot compile itself under dos because the libs to fake the unix syscalls under dos are currently too large. A number of fixes have been added that were apperently in another version of the tools to aid operation on other machines, there's now a better chance of these programs working on 'big-endian' machines. Various bits and pieces have been added to the libs, including some time functions (but these need a _small_ way of dealing with "DST"). -Rob