From: uhclem@nemesis.UUCP Date: 24 Jan 94 21:26 CST Newsgroups: comp.sys.tandy Subject: 6000 Compiler can handle long names Message-ID: <-30918@nemesis> Path: mnemosyne.cs.du.edu!tali.hsc.colorado.edu!csn!magnus.acs.ohio-state.edu!math.ohio-state.edu!howland.reston.ans.net!cs.utexas.edu!uunet!seas.smu.edu!utacfd.uta.edu!trsvax!trsvax!nemesis!uhclem Nf-ID: #N:nemesis:-30918:000:3468 Nf-From: nemesis.UUCP!uhclem Jan 24 21:26:00 1994 Lines: 91 There is a myth that the 6000 'C' compiler cannot handle long variable names. As of the 3.2 XENIX Development System Upgrade this is false. Quoting from the printed man page: (Dated 6-Dec-87) " -+ Increase the significant length of preprocessor symbols from 8 to 32 chafacters; and increase the length of external symbols from 7 to 31. This flag should be used on all or none of a set of source files that will be linked together. " Internally we called these "flexnames". They work fine. The 3.2 compiler also adds (or documents) the following options over the 3.0 (28-Jun-85) man page: " -Os Causes the optimizer to output a summary of its optimizing activity to stderr. -Oa The optimizer will attempt to convert automatic variables into automatic register variables, if they were not already declared as register variables. The optimizer assigns automatic variables to unused registers based on the number of times they are referenced in a function. "An" registers will also be used with certain types of variables. (Automatic variables whose address is passed to other functions are not converted.) This option should not be used on modules that utilize setjmp(). There is the potential that if automatic variables are part of the environment passed to setjmp(), one or more of these variables may be converted into register variables, thus causing incorrect results. -t0 replaces the compiler phase with a program called c68 from the current directory. (Older man page called this -t1, which is a typo.) -t1 replace the object code optimizer phase with a program called c68o from the current directory. (Older man page called this -t2, which was also a typo.) -Bppathname Replaces the preprocessing phase with the program pathname. For example: -Bp/lib/cpp.test causes /lib/cpp.test to be used instead of /lib/cpp for the preprocessing phase. (/lib/cpp.test must be supplied by the user.) -B0pathname Replaces the compilation phase with the program pathname. -B1pathname Replaces the optimization phase with the program pathname. -Bapathname Replaces the assembly phase with the program pathname. -Blpathname Replaces the linking phase with the program pathname. Notes: The -Oa option does not always have sufficient information to perform its conversions. It is not supposed to change anything if a change is dubious. If unusual behaviour is exhibited, recompile the program or individual module without the -Oa option. End quote I use -Oas all the time and there it has only produced faulty code in one or two cases. The 3.2 XENIX kernel was compiled this way, as were many of the utilities and libraries. It is interesting to compile something reasonably small with CC -S, look at the resulting code, then compile with CC -S -O or CC -S -Oa and see what happens in there. This compiler doesn't know about void, nor does it understand ANSI prototypes. But it does handle the long symbol names. A typedef will take care of void, and there are utilities to make K&R-style function declarations. On the other hand, the 3.2 compiler will run on a 1 Meg non-paged system where GCC has lots of trouble. Frank Durda IV |"How do I know? Oh I did some ...utacfd!nemesis!uhclem (nearest internet) | work on the optimizer and ...letni!rwsys!nemesis!uhclem | created -a, plus lots of other ...decvax!microsoft!trsvax!nemesis!uhclem | optimizations, that's how."