Club Amiga de Montreal

home *** CD-ROM | disk | FTP | other *** search

/ Club Amiga de Montreal - CAM / CAM_CD_1.iso / files / 567a.lha / DICE_v2.06.37 / version.doc.pp / version.doc

Wrap

Text File | 1991-11-11 | 21.4 KB | 570 lines

VERSION.DOC DICE VERS DES COMMENT VERS = Release Version DES = bug fixed / feature designator XB - major bug fixed in main compiler (use of routine in question could have caused a crash) B - minor bug fixed in main compiler F - feature added E - enhancement to existing feature O - optimization added ADD - feature added or fixed that is standard C but was not previously implemented or working. XBL - major bug fixed to library BL - minor bug fixed to library FL - feature added to library OL - optimization added to library XBA - major bug fixed in support program (i.e. not DCC or sub programs) BA - minor bug fixed in support program FA - feature added to support program (or new support program) OA - optimization added to support program COMMENT = comment or bug, feature 2.06.37 B arrays are now aligned the same as their element type alignment. Thus, char a[3]; is byte aligned while short a[3]; is word aligned. Previously arrays were always word aligned. (not really a bug, but required to work with Amiga cia structures) F PROFILING ADDED. New -prof option to DCC, LIBMAKE, and FDTOLIB. This tool gives you unmatched profiling capabilities. You may profile just your source, standard C calls, amiga library calls, or any combination. A <programname>.dprof file is automatically generated when your program exits and may be turned into a textual representation with the DPROF program. Profiling routines have a 1 uS granularity under 2.0, 20 mS granularity under 1.3. __noprof qualifier for procedures added -- prevents a procedure from being profiled (used by the system to prevent the profiler routine from getting profiled itself) F C.LIB renamed to CS.LIB to generalize library names. Infix IDs are now generalized as follows: L large data model S small data model R registered args, else stack args (registered release only) P profiled, else not profiled. DCC will now use the -prof and -md/-mD options to determine which c.lib and amiga.lib to use. The more useful combinations are included in the distribution and registered users may make additional versions as needed. Note that you may mix options -- it is perfectly valid to compile source modules with -mD but link small-data. BL bug in fread() fixed relating to an fread() occuring after an ungetc(). E ALL LIBRARIES IN THE SYSTEM REORGANIZED. There are now many more libraries. Registered users now have a means of remaking *ANY* library from scratch by unpacking the library source code and typeing 'dmake' with no arguments for instructions. Generally only the DTMP: assignment and -2.0 in your DCCOPTS is required. Since there are many more libraries possible then I can fit in the distribution, this means may be used to create esoteric libraries that the majority of people will never use. F Fixed two major bugs with constant floating point arithmatic... addition and subtraction would not work correctly in a few cases (it would get the sign wrong), the exponent would get screwed up in some cases (usually after a negation), and I was overflowing the single-precision exponent range for conversions to type float. Ooooo this stuff should have been caught long ago! F Fixed a pointer arithmatic bug, yet another case where a scratch register would get overwritten due to a procedure call. 2.06.36 B register allocation bug fixed, related to floating point routines mainly. Would generate a 'Too many redos' error in some cases (bug did not generate bad code). F PRECOMPILED INCLUDES ADDED, see -H option in DOC/DCC.DOC. 2.06.35 FL under 2.0 (V37), getenv() now checks for local enviroment variables before global ENV: variables. Under 1.3 only global ENV: variables are checked for. BL fixed %U and %W in strftime(), would generate incorrect week-of-year B fixed stringize operator for \n, i.e. quoteme(fubar\n) -> "fubar\n", would previously generate incorrect result (I think) B fixed bugs in integer multiply & divide, some cases would generate fatal errors (all caught by DICE) B fixed bug &= operator, would generate assembly error in conditional: while (x &= (x - 1)) ... F optimizations made to mulitply & divide, handles 16 bit expressions better. F minor optimization added for procedure calls within the same module. If call is within a byte offset then BSR.B will be used, else BSR.W is used. F DLink now checks for illegal small-data references to far objects and generates the appropriate error message. This was added for completeness and also because __far or -mD data objects are no longer A4-relative. The check will catch illegal model mixing. 2.06.34 BL bug in <varargs.h> (old style var-args) fixed, was skipping first argument. Note that all ANSI projects should use <stdarg.h> 2.06.33 B bug in offsetof(type, array_elm) fixed, previously caused error message. 2.06.32 B fixed bug in dlink Introduced in .31 that would cause a machine crash. 2.06.31 B fixed floating point bug relating to array indexing, would clobber A0 in some cases B fixed preprocessor bug relating to the use of hex constants in #if expressions (they wouldn't work) E redid the handling of the ?: routine to allocate its result storage (a temporary register) after running the sub-expressions instead of before, which allows you to nest ?: indefinitely without generating an too-many-redos error. B register allocation bug (would generate fatal error) in &array (redundant & operator) has been fixed. B DCPP now accepts null directives. # with no directive. 2.06.30 F DCPP now evaluates undefined symbols in #if expressions as 0 instead of generating an error message, as per ANSI. B Yet another bug in the # (stringtize) operator found. Fixed. Really really really should work right this time! B bug in switch/case fixed, case <enumid>: would generate an error message. BA I was shipping the wrong mathieeedoub*.library shared libraries, would cause erronous float->double conversion. B another floating point bug found in && and || of the form 'fp && fp' or 'fp || fp', would cause incorrect evaluation and/or assembly error B small bug in fread() would sometimes set error flag when returning EOF, and fseek() would not reset this flag (it does now). BL bug in *scanf(), %c would overwrite the char by one byte (i.e. store into two chars instead of one). fixed. B DCPP didn't handle a quoted /* in directives correctly. F Large programs that have more than 64K of *initialized* data may now use either __far for selected items (preferable) or -mD without causing link errors. Previously dlink would combine far data along with near data making all small-data referenced bss items out of range if more than 64K of initialized data had been declared, __far or not. Now, DC1 generates 'fardata' sections properly when -mD or __far is used (previously only worked with __far), and dlink will not combine any section name beginning with 'far' into the small data model. This is only of major import when compiling incredibly huge programs. Note that having greater than 64K of bss with -mD or __Far has always been supported since library bss is always forced in front of program bss. F DLink now takes approximately 15% less memory. While it still requires memory to hold the combined size of all objects and libraries, incidental memory requirements now reuse much of this memory from unlinked sections instead of allocate new memory. 2.06.29 B another long standing bug found: short a = i * j; where i and j are long's and *also* a memory reference (not in regs), the multiplication short-result optimization was screwed up. B register allocation bug (would produce a fatal error) fixed, registers would not be deallocated in constant conditional expressions such as: 1 ? (a = 4) : b; 2.06.28 B really dumb bug fixed, made retroactive to 2.06.27 (all 2.06.27 dists were redone with this fixed). Had been introduced in 2.06.27 2.06.27 E add -// option to DCC and DCPP, enables C++ style // comments. If not specified, // style comments are ignored. Generally one puts -// in DCCOPTS. E added -no-env option for DCC. If specified, DCC will *not* load options from DCCOPTS. B fixed bug: static int fio(), fio(); (procedure declared twice on same line) would force procedure to be non-static. fixed bug: __geta4 int fio(), bar(); ... __geta4 would apply only to first procedure. B fixed bug w/ labels ... label: case ... would generate error message (case or default following directly after label). Was introduced when I initially fixed labels. E AUTO-AGGREGATE INITIALIZATION ADDED, e.g. main() { int a[] = { 1, 2, 3 }; ... }. static & global initialization has been re-vamped. Note that bit-field aggregate initialization is NOT supported yet. B Fixed bug that has been in there from the beginning. int *a; a[20000] = 1; ... would not multiply the constant by 4 to get the index. ONLY EFFECTED CONSTANT INDEXES WHOS OFFSET WOULD BE LARGER THAN A SIGNED SHORT, which is why it was not caught. B Fixed bug in float point stack allocation... temporary fp variables from the previous procedure would not properly be cleared from the cache, causing erronous stack offsets (that overlap other variables) in the next procedure. E Added ANSI 'U' extension for constants, e.g. 123U. DICE now automatically promotes a constant to unsigned if it does not fit in a signed long, such as 4000000000. Note, however, the int->float conversion is currently always signed due to lack of unsigned->float calls in the amiga's shared fp libraries. B small bug in fread(), would not set the EOF flag when an EOF occurs for an fread() requesting a buffer size larger than the stdio buffer size. B when making a registered-args call (-mr, -mR, -mRR) where a pointer - int subtraction occurs, i.e. foo(ptr - int, int), arguments after the argument containing the ptr - int would be incorrect (registers would get squashed). ONLY applies to registered-args calls. This has been fixed! B bug in preprocessor fixed, #define fubar(x) "a x a" would replace the 'x' in the string, which is wrong. B bug with bitfields has been found and fixed. Essentially, bitfield operations in routines that need to run a retry pass can get munged. This is usually caused by routines that declare lots of local variables. 2.06.26 B Fixed bug in floating pt compare of the form: x = (f1 == f2); i.e. a compare where the result is a boolean 0/1 when not in a conditional. Was caught by DICE (caused a fatal error). B Fixed bug w/ bitfields (was caught by DICE), would give fatal error if destination was not a register for field extraction. B fixed bugs with stringtize operator in macros, should now work according to the ANSI spec. #define fubar(x) #x BL month was off by one for %m operator in strftime() E PROCEDURES MAY NOW RETURN STRUCTURES. YOU MUST PROTOTYPE ALL SUCH PROCEDURES. E added // comments to DCPP (C++ style comments) 2.06.25 E doc/errors.doc now exists, error handling centralized. E a printer driver and exec device driver example has been added to examples/ ... so now I have a shared library example, dos device driver example, EXEC device driver and printer driver example. 2.06.24 B stack allocation changes in 2.06.23 were broken, broke floating point. (got the stack tracking mixed up w/ variable allocation) B ?: fix in 2.06.23 was broken (sigh), fixed ... 'default:' would not be properly tokenized. E *scanf() now supports %e/E/f/g/G and %[..] 2.06.23 BETA B goto labels are now in their own name space, no more conflicts with int a; ... goto a; ... a: ... F added stack tracking for fp temporaries, meaning that subroutines containing floating point expressions do not take up anywhere near the amount of stack they used to. F allows declarations after switch() { and before first case. F fixed problem with ... ? symbol : ... (symbol was being interpreted as a label instead of a variable within a ?: conditional F fixed problems with conditional returns from fp operations, such as: double a, b; int c = (a < b); that must generate a 0/1 result instead of a branch (as in an if()). Would generate error before. 2.06.22 B fixed problem with GOTO labels, a goto label semantically is not a statement unto itself. The problem is a line like the below: if (i) j = 1; else x: j = 2; Currently j = 2; is completely outside the if in DICE. This is incorrect, 'x: j = 2;' counts as a single statement. E libmake is now documented, it wasn't before! New options have been added to libmake and fdtolib. For libmake, DCC's -D option has been added. For fdtolib, a new -p option is specifiable to set the prefix used on exported symbols (to support creation of dynamic.library compatible object modules) B fixed bug in string constant array initialization, char x[1] = "a"; previously generated an error B fixed infinite ERROR printing loop F revamped constant-address handling (i.e. when you map a complex structure over IO space at a known address, useful when generating ROM'd apps for 68000 based uP's. F dynamic.library & support routines now part of the distribution. 2.06.21 E fixed register reallocation, should now do a better job allocating registers in complex subroutines. Hopefully squashed 'maximum retries exceeded' error for all but the most incredibly complex expressions. B fixed bug in int->float conversion, expressions such as: double x = (double)(a * b); E fixed somestuff that generated an infinite WARN loop. F DCC now creates the destination directory heirarchy if it does not exist. For example, when you DCC com/x.c the object file will be T:com/x.o, with T:com created if necessary. F DYNAMIC FUNCTION/VAR LIBARARY SUPPORT ADDED, see __dynamic explanation in DYNAMIC.DOC . F DCPP now accepts -Ddefine(vars..)=definition, it used to be able to only accept -Ddefine=value 2.06.20 F TimerBase, for timer.device library calls, now has auto-open capability, meaning that you can simply call AddTime(), CmpTime(), SubTime(), and the new 2.0 calls (GetSysTime() & ReadEClock()) without having to open the timer.device and set up TimerBase yourself. B bug in if (ptr & 1) and similar constructs doing a logical and in a compare against a pointer rather than an int produced bad code caught by DAS. B x.o was in c.lib, this is wrong... x.o must be the LAST file in any link and is included automatically by DCC when it runs DLINK. F additional example code in DCC2:examples -- a DOS handler and a shared library. 2.06.19 B bug in dynamic stack code under 1.3 fixed B bug in workbench startup under 1.3 fixed B bug in register allocation fixed (would cause fatal error message 'too many retries' sometimes when procedure args declared as arrays instead of pointers) 2.06.18 F several more libraries have auto-open capability B bug in DLINK fixed -- would GURU on large links due to an error in the allocation routines (relocation info larger than 4K would cause a GURU) 2.06.17 F DCPP now scans DINCLUDE:PD/ for include files. This is where I will be putting .H files for various ports (such as the UNIX regexp port). This was done so as not to clutter the main DINCLUDE: and to differentiate between PD files and non-PD files. UNIX compatible .H files are usually placed in DINCLUDE: or DINCLUDE:SYS/ F preprocessor now handles #if defined(symbol) properly (was previously not implemented) F BITFIELDS now 99% implemented! including signed/unsigned qualifications. The only thing that does not work is ++bitfield and --bitfield. Single-bit bitfields have been optimized. aggregate initialization of bitfields is still wanting. executable actually got smaller when I finally got down and fixed these up. No 020/030 support yet. F startup code modified to allow writes to desc 0, reads from desc 1, etc... at the file descriptor level but not the stdio level. This is to support some UNIX programs which do this. BL bug in [f]putc() fixed -- [f]putc() is supposed to return the put character, but did not do so in the case of a new line (putc), or at all (fputc). F NEW LIBRARY ROUTINESE: stat(), fstat(), getenv(), opendir(), readdir(), closedir(), rewinddir() (see MAN/UNIX.DOC) F chdir() now restores the original directory on exit F Dynamic Stacking added (-gs option). Somewhat primitive, but will get much better. This allows subroutines to be as deeply nested as required, allocating and deallocating stack frames when low stack conditions exist. F longjmp()/setjmp() now properly handle dynamic stacks B fixed another bug, this one having to do with the index address mode (An,Dn.x) combined with a binary operator. B fixed bug in FDTOLIB that screwed up register ordering. 2.06.16 B fixed EXG bug in DAS, EXG As,Ad would generate Ds,Dd.. screwed up registered args (-mRR). B fixed bug in *= and /= combined w/ array indexing, things like temp[i] /= 23; would use the scratch regs for storage across the divide call. 2.06.15 F BASIC REGISTERED ARGUMENTS ADDED. See -mr, -mR, and -mRR options F prototype optimization. If a prototype says a procedure takes a short, then DICE pushes a long (as usual). However, DICE will no longer EXT the object before pushing it (upper word is garbage). This works because the procedure, having declared the argument a short, will only access the lower 16 bits. B fixed token pasting bug in dcpp, a ## b ## c now works properly. F first registered distribution that includes the 2.0 includes, restructuring of 1.3 includes (now called 'amiga13' instead of 'amiga'), and restructuring of disks to make more space. 2.05.14 BL fgets() would overflow the buffer by one character when input lines were larger than the buffer size. B DCC would complain about output (-o) files ending in .h or .c when they do not actually end in .h or .c .. 2.05.13 B DCC was supplying only a 2K stack to NON-RESIDENTED sub programs. Oops. F New floating point support for MATHIEEESING*.LIBRARY ... single precision IEEE library. This is now the default since it is more compatible with the double precision IEEE library for conversions between single & double precision. Use the -ffp option to DCC to use the old motorola fast floating point single precision library. Note that the new ieee single precision library may not exist on every computer. 2.05.12 B another bug with 'const static' items not properly aligning the routine they were declared in (when declared within a procedure definition). Fixed. B pfmt has a local 64 byte buffer which used to be static. This is illegal for the ROM.LIB version of sprintf() & vsprintf() and so I have made this an auto instead. This fix does not effect Amiga C programming though it does mean that [v]sprintf() may now be called out of context or multi-threaded without mishap. F qsort(), rand(), and srand() have been written. B volatile was not being handled proper, 'volatile char *ptr' is a pointer to a volatile char, NOT a volatile pointer, for example. 2.05.11 F new library, ROM.LIB, contains only those routines from C.LIB that are completely self contained (reference no static or global data, or other routines. Use only autos if at all). This library is included instead of C.LIB when the -rom option to DCC is used (with the -rom option, amigas*.lib is not included either, and it is up to the programmer to specify his own startup module) F char ary[0] and empty structs return errors, but a direct 'void' declaration is allowed. i.e. 'void x;', useful in some cases. F Specifying the 'register' keyword now adds some 'weight' to the variable, perhaps giving it more of a chance of becomming a register. B DCPP was improperly handling token pasting (##), this has been fixed. B static const objects declared within a procedure not word-aligned could cause the procedure code to not be word-aligned, thus causing a software error. F CONST object handling has been changed. Before 2.05.11 const objects were not put into a CODE hunk unless the -ms or -mS options are given to DCC. 2.05.11 and beyond put const qualified objects in a code hunk regardless. F C.O Modified. Any program run in the background with RUN will not attempt to Open("*",..) to get the stderr channel but instead will use the stdout descriptor if it exists. This allows programs to be detached from a CLI (i.e. that you can run a program and then close the CLI window without mishap) F -020, -030, -881, -882 options formally added (currently -020/-030 are the same and -881/-882 are the same) -020: uses EXTB.L (no major support has been added yet) B DSEARCH had a bug... when multiple files are scanned it would miss lines in the second file and beyond! F oops, const items in the library were not placed in the code section. This effects <ctype.h> ... you must recompile any source files that use <ctype.h> macros since the access method is now pc-relative instead of data-relative. The fix is more pertainent to ROMd applications that use <ctype.h> ... the lookup tables are no longer copied into RAM for such applications. 2.05.10 F __config and __shared type qualifiers added B --var; generated an internal error