Copyright (c) 1992 SAS Institute, Inc, Cary, NC All Rights Reserved This document describes bug fixes and enhancements made in the 6.1 release of the SAS/C Development System for AmigaDOS. Our philosophy in reporting fixed bugs is that users have a need and a right to know what has changed in the latest release. This means publishing a list of fixed bugs with each update. Please don't be alarmed by the number of items on the list; many of the "bugs" reported here are not very serious, and others are very obscure. Some of them have only been reported once, others have been reported many times. Thanks to all the alert programmers out there who have brought problems to our attention and provided the detailed information necessary for us to reproduce and fix the problem. If this update does not fix a problem that you have been having, please contact our Technical Support Division for help as soon as possible. All reported CXERRs, Enforcer hits, and cases of bad code are fixed. If you get one of these, please contact technical support immediately. It will help if you have a compilable test case. In addition to the specific bugs listed, several code generation bugs were fixed in the peephole optimizer and the code generator (sc2). No bugs were found or fixed in the global optimizer, but some sc2 or peepholer bugs only showed up when the global optimizer was run. Since the code generation bugs did not affect most people, and they are not easy to describe or reproduce, we will not describe them individually here. We suggest that you recompile all your code if it is behaving strangely. A note about the release numbering system is probably in order. With version 6.0, we have adopted Commodore's shared library numbering system for the entire product. This means that the number to the right of the decimal in the version number should be considered a number in its own right, not merely as a fraction. For example, our first release was 6.0; this release is 6.1; the logical progression will be 6.2, 6.3, ... 6.9, 6.10, 6.11, ... 6.99, 6.100, 6.101 and so forth. This is a change from version 5, in which the number to the right of the decimal was considered a fraction, with possible leading zeros. We reserve the right to skip some numbers to keep track of internal releases, but the number to the right of the decimal will always increase with a new release. Also note that not all executables and shared libraries have been modified. Those that have not been modified keep their old version number of 6.0. This is normal. A list of modified files appears at the end of this document. Users of Comeau C++ should be aware that Comeau Systems has produced a new version, version 3.0b, that has integrated support for SAS/C V6.x. Version 6.1 works well with Comeau C++. Contact Comeau Systems for information on obtaining the 3.0b patch, which is a free upgrade for registered Comeau C++ V3.0 customers. ***************************** IMPORTANT ************************************ Many customers are experiencing problems that could easily be avoided by referring to the correct sections of the manual. Specifically 1) There are two sections in the manual covering GST's. The first is in the User's Guide Volume 2 in the Utilities section. The second starts on page 21 in the Library Reference Manual. Both sections contain information needed to use GSTs. 2) A guide to converting programs from version 5 to version 6 can be found in Appendix 5 of the User's Guide Volume 1. 3) With a floppy disk install, no icons will be copied to the created disks. This means that you will have to work from the CLI only, or that you will have to delete something off of the disks in order to make room for the icons. This is not mentioned in the manual and should be. For more information on workarounds for this, refer to Problem 5 under the Miscellaneous Problems button on the main help screen for Common Problems. ============================= Library and Header Changes ============================= The tmpnam() function used the first three characters of the command as invoked in constructing the name of the temporary file. This did not work if you invoked a command via an explicit path that contains a ':' or '/' in the first three characters. This was the cause of several strange problems in the SC and SLINK commands as well. In __main(), the return value for the Open() which opens the stdio window was never checked. Also, __main() now checks the __stdiowin[] array and does not add the program name if __stdiowin[] does not end in a "/". Some system tagcall routines did not have tagcall pragmas, most notably EasyRequest() and BuildEasyRequest(). did not include the definitions for REG_FP0 - REG_FP7. These definitions are used by getreg() and putreg(). scanf() routine failed to scan a floating point number with a leading '+' character. fread() did not set EOF flag properly. (It did return the correct return code, however.) forkl() did not correctly open "*" for stdin and stdout. If the UTILLIB option was used and an auto-open library failed to open, the __autoopenfail() function crashed the machine when invoked. Autoopen libraries were opened twice and closed once because the autoopen functions were compiled with PARMS=BOTH. was not protected from multiple #includes. m68881.h did not include definitions for fgetexp() and flog2(). did not contain definitions for the S_xxx bits. (They were previously defined in ; now they are in both files.) The bits returned by stat() in the st_mode field were incorrect. The 4 bits corresponding to the "rwed" attributes were OPPOSITE what they should be. All other bits were correct. did not include all files, notably . __builtin_memcmp() generated incorrect code when a variable containing zero for the length to be compared was passed to it. In , LayersBase was declared struct LayersBase *; it is now declared struct Library *. Closing a file descriptor twice caused a crash. While this is arguably an illegal operation, most systems accept it silently. ===================== CPR Debugger Changes ===================== CPR didn't pass argv[0] correctly when invoked from WorkBench. CPR didn't get the address right for the first data item in the FAR or CHIP sections. The -device, -unit and -speed parameters were not implemented on cross CPR. Some programs required abnormally long times to load under CPR. This load time has been reduced. Watch and register windows are no longer covered up by the source window if they are opened via command-line options. Hitting the ESC key in the CPR WMSG window caused the window to go away. Additional WMSG commands after this crashed CPR. Highlighting was messed up if a window was resized while something was highlighted. The ECHO command didn't work reliably from an externally- invoked AREXX script. The CPR screen stayed open if a visitor window was present, but CPR exited anyway. The Calls window sometimes trashed its own bottom border. CPR was doing incorrect pointer arithmetic for multidimensional arrays. ============================== Parsing and Diagnostic Changes ============================== The __chip keyword produced an Error 82 for arrays declared that were larger than 64k-1 bytes. This restriction has been removed. No diagnostic was produced for assigning a regargs function to a stdargs function pointer, or vice-versa. However, the warning "stdargs used for indirect function call" was produced inappropriately for calls through function pointers declared __stdargs. The inappropriate warning has been removed; the assignment now generates Warning 225. Error 15 was generated inappropriately for a call via a function pointer defined via a typedef. Invoking a preprocessor macro with no parameters that is followed immediately by a parameter list containing the preprocessor macro name once again followed by a parameter list failed to expand the second invocation of the macro. (This is an extremely obscure case, so don't worry if you didn't understand the explanation.) If an undefined member of a structure was accessed, that structure was no longer considered equivalent to other structures with identical member lists. This generated warning cascades. National characters above 0x80 were not permitted in variable names. When using the GenProto option for a file with more than one period in the name, (ex. a.b.c), the generated proto file was placed into a .h file with only the first part if the original name (ex. a.h). The ERROR=ALL option to the SC command turned on all messages, including those normally suppressed. It should have just promoted the warnings that were already enabled to errors. The -ca option on the SC5 command incorrectly enabled Warning 148. Error 67 was generated if the __inline keyword was placed on the definition of a function returning a pointer to a structure. C++-style comments were not recognized if the ANSI option is on. While this is arguably correct, since C++ comments are non-ANSI, many users requested that they be made to work. If the GST has only in it, then you #include in the .c file, IntuitionBase was considered to be an undefined structure tag. Generating a GST when the file being compiled erroneously tried to access a structure that had not been defined resulted in a crash. Generating a GST with static data defined in a header file now generates an error message. This is invalid; data cannot be defined in a header file, and there is no way to make a static data item into a simple declaration, as is done with external data. ==================================== Code Generation/Optimization Changes ==================================== The return value of a function was lost when returning from a call in which the stack extension code allocated additional stack. Auto variables sometimes got trashed when using longjmp() to jump up the stack. When SHORTINT is selected, all numbers between 32k and 64k were translated improperly as negative numbers between -1 and -32k. Initialization of automatic arrays and structures didn't work if the initializers did not contain as many elements as the array or struct and the STRMERGE option was used. The DISASM option caused a crash if you specified a directory name as the output filename. The assembler code produced by the DISASM option included MULSL.L and DIVSL.L instructions instead of MULS.L and DIVS.L. The DISASM option produced bad assembly code if you had calls to __builtin_emit() in your code. An incorrect function pointer was generated for a forward reference to a static function if the ABSFUNCPOINTER option was specified. With the global optimizer active, calls to tagcall functions made via #pragma tagcall sometimes did not clean up the stack properly. Functions declared __inline sometimes did not generate actual callable instances of themselves when they should have. This occurred if you did not run the global optimizer; if the function called itself recursively; or if the function was called from outside the current module. The functions would generate an unresolved symbol at link time. FPrintf and other tagcall functions that take an OPTIONAL tagcall list generated an error message from phase 2. #pragma tagcall functions now silently accept empty tagcall lists. Implicit references to substructures generated incorrect offsets if the substructure was not the first item inside the main structure. Adjacent bitfields declared explicitly "short" or "long" were not merged properly. This caused the structures containing them to be larger than they needed to be. Bitfields declared "int" or "char" were not affected. Some variables were listed with the wrong type in the compiler-generated cross-reference. This occurred after a typedef definition. ================= SE Editor Changes ================= SE was not pure (as reported by WShell). SE returned immediately when asked to compile the file in the editor buffer (F4 key) if the flag was enabled under AmigaDOS 2.0 that allows the "*" character to be a wild card (RNF_WILDSTAR). Turning on the "Use TAB Character" option and the Autoindent option resulted in garbage characters being added to the file when you hit RETURN after entering an open curly brace. SE did not turn the mouse pointer back on during time-consuming operations (like SEARCH/REPLACE). SE sometimes opened a window smaller than its menus required when switching from interlaced to non-interlaced mode. It was impossible to configure the screen colors and save them. You could cycle the colors while in SE, but when you exited, the colors weren't saved. A bad version of the editor macro "findsym.se" was provided. SE set no pointer during compilation. It now sets a busy pointer. SE did not correctly add the backup file extension when saving backup files. SE did not accept national characters that were entered through a REXX script. ============================= SCMSG Message Browser Changes ============================= SCMSG menu colors were funny-looking on an 8-color workbench screen. SCMSG didn't protect itself against the editor exiting while it was processing a message to the editor. SCMSG did not open on the default public screen. It now does so; it also accepts a SCREEN command-line parameter and an "opts screen " AREXX command (see below). SCMSG redrew its message lines even if they had not changed when receiving new messages. This caused unnecessary flicker. SCMSG didn't set the AUTOEDIT option even if the options file said to. It also reset the GOTOFILE and GOTOLINE options to the SE defaults even if the user had erased those values. SCMSG now accepts the following escape sequences in the REXX commands that it sends to the programmed editor. These are in addition to the %f, %l and so forth that are documented in the manual: %# - message number %m - message text %e - alternate filename %k - alternate line number Note also that the %c escape sequence gives the message Class (ERROR/ WARNING) rather than the column number, as documented in the manual. This was an error in the manaul. SCMSG now accepts the following new or modified REXX commands. All old commands still work. "build [options]" The "build" REXX command now passes any options specified on to the SMAKE utility. "hide" [option] The "hide" REXX command takes "rexxonly", "norexxonly", "autoedit", and "noautoedit" as options. The options control when the window will reappear. The "wait" option listed in the manual is the default, so there is no need to specify it. "number" synonym for "errnum" "newbld " tells SCMSG to clear all messages for the specified compilation unit. Sent by the compiler when a new compile is begun to delete all old messages. It also forces SCMSG to invoke the editor on the next new message if the AUTOEDIT option is on. "newmsg "" "" 0 "" 0 " adds a message to SCMSG's list. is the compilation unit the message is associated with (the .c filename), and indicate the filename and line number of the message, is one of "Error", "Warning" or "Info", is a positive error number, and is the message text. The 0 "" 0 in the middle must appear exactly as written. If the message text contains the words 'See line file "name"', the is taken to be the alternate line number and the is taken to be the alternate file name. "opts [option]" where [option] is one of the following: load - Load options from the specified file save - Save options to the specified file portname - Set the name of the editor's AREXX port to the specified value screen - Close the SCMSG window and reopen it on the specified public screen. The default value for the GOTOLINE option, intended for use with the SE editor, is now LL "%l\n" DM %m instead of just LL "%l\n" This displays the message text in the message line of SE after taking you to the line of the message. ===================== Other Utility Changes ===================== SMAKE didn't read the tooltypes array as documented; it also reported that A: is a "bad drive specifier" when it was used in a target. Closing the SMAKE window after SLINK started caused a system crash. SMAKE did not use the SYS_UserShell tag when calling the System() routine under AmigaDOS 2.0. It now does. If multiple targets were passed to SMAKE on the command line and the first was up to date, the subsequent targets were not built. ASM didn't deal with packed decimal mode constants correctly. OML gave a "phase error" if you attempted to place object files in the library that contained chip data. LCTOSC and DUMPBOJ did not set the return code correctly. LCTOSC incorrectly translated '-Ln' to 'LINK DEBUGSTRIP' instead of 'LINK STRIPDEBUG' HyperGST could not locate header files obtained from the local project if invoked from WorkBench. SLINK did not always set the execute bit on the generated executables. LPROF/LSTAT were giving strange results when used on programs with multiple code hunks. This is a very long-standing bug dating back to version 5.0. Scsetup did not correctly process the '?' parameter when invoked from the Shell. It printed its command template, then started working on the current directory as if no parameters were entered. SCOPTS only saved to ENV: when asked to "Save Default". It now saves to ENVARC: as well. =============================== Documentation/Help File Changes =============================== The sc.guide hypertext file incorrectly listed 255 as the default value for the IDENTIFIERLENGTH option, instead of the correct value of 31. 255 is the maximum value. Some new "Common Problems" were added to the Common Problems document. The scmsg.guide hypertext file was modified to add Error 145, which is new to this patch release, and to update or add descriptions for Warning 7, Error 82, Warning 184, Warning 198, and Error 200. In addition, the Global Optimizer warnings which were left out of the documentation were added to the Compiler Errors and Warnings section of scmsg.guide. The warnings added are warnings 301, 302, 303, 304, 305 and Note 306. Additional information was added to the ANSI compiler option description in sc.guide. The ANSI option has effects other than simply turning on warning messages. These effects are now described. Three data item names and one function were added to sc_lib.guide. All four additions are listed separately in the main Library Functions help screen under the button "Newly Documented Functions and Data Names". Additional corrections to the documentation can be found in the "Changes to the Documentation" help screen (sc_changes.guide) under the button "Version 6.1 (Error Correction Patch) Changes". The "compile.ced" macro intended for use with CygnusEd had a bug in its use of the "jumpto" command. "jumpto" expects both a line number and a column number, and only the line number was provided. CygnusEd users may want to copy the new version of the macro into their REXX:CED drawer. The TurboText macros referenced explicit directory paths, but the directory was renamed in the final version; therefore, they did not work correctly as installed. The TurboText directory has been renamed. Please reinstall these macros if you had problems with them in 6.0. ====================== List of Modified Files ====================== The following files were modified during the 6.1 patch installation process: Drawer Files SC: read.me SC:C smake, slink, se, scopts, scmsg, cpr, cprx, cprk, scompact, omd, dumpobj, hypergst, asm, oml, lstat, lprof, sc5, tb, lctosc SC:LIBS sc1.library, sc2.library, schi.library, scpeep.library, sekeymap.library, scdebug.library, sclist.library SC:ENV se.dat SC:LIB all .lib files and startup modules SC:INCLUDE fcntl.h, stat.h, dos.h, m68881.h, mffp.h, mieeedoub.h, stdarg.h, sys/dir.h, proto/all.h, proto/layers.h, pragmas/intuition_pragmas.h SC:REXX findsym.se, showcli.cpr, showprocess.cpr SC:HELP sc_prob.guide, sc_lib.guide, sc_change.guide, scmsg.guide, sc_util.guide, sc.guide, cpr.guide SC:SOURCE autoopenfail.c, _main.c, _oserr.c, intuitlib.c, _cxferr.c, c.a, cback.a SC:EXTRAS TTX/instructions, CED/rexx/ced/Compile.ced