donn@sdchema.UUCP (Donn Seeley) (12/16/83)
First I want to thank the people who responded to my f77 bug reports with more helpful suggestions, especially Peter Gross, Ralph Campbell and Bob Corbett. As a result I have lots of useful information on f77 bugs and I expect to make use of it to fix them later (when I get back from vacation in January). Next, a warning about bug (5) on my list, allowing REAL variables to be put in register. Bob Corbett said that he could recall a previous attempt to do this with f77 (& C), and that it resulted in problems when certain expressions were computed. Specifically the code generator occasionally caused part of a DOUBLE PRECISION temporary to be trashed when evaluating expressions containing both REALs and DOUBLE PRECISIONs. I have tried hard to get this to happen in test programs under my version of the compiler, and I have been unsuccessful. In fact the code looks suspiciously like the compiler does a 'global substitute' of register names for memory references when register REALs are enabled. Still, people who implement my suggested fix should consider themselves warned. If anyone does try my fix and finds this bug, please send me mail describing it so I can have a chance to make a better fix. Ralph Campbell and Bob Corbett have a better fix for my bug (1) -- I will post this if I get the chance before I leave on vacation, otherwise it will wait till January... It may interest readers on the net to know that there is a reason for the unpolished state of the f77 compiler in the 4.2 distribution. I have been told by people both at and recently away from Berkeley that there has been some severe instability in the CSRG and related groups preparing VAX software. One example of this is that the Berkeley f77 project was (alas) cancelled abruptly in May of this year. The only person currently doing any work on f77 at Berkeley is the staff programmer who oversees the 4bsd-bugs service, who does his repairs in his spare time. The cancellation is doubly unfortunate because the project had planned major optimizer improvements for the compiler but had not completed implementing them. Work on a particularly sticky piece of code had been postponed until this summer; with the cancellation of the project this work was apparently abandoned. I was told that the powers that be at Berkeley are not dropping Fortran for philosophical reasons; they simply have better ways to use their present staff and students. So it is possible that new personnel may eventually be allowed to work on Fortran at Berkeley. My impression is, however, that it would be a mistake for you to hold your breath. There is a possibility that recent changes in the f77 group at BTL (or whatever that organization is called this week) will bring about a better compiler from them, so stay tuned. That's about all the good news I have. Trying to forget Fortran until January, Donn Seeley UCSD Chemistry Dept. RRCF ucbvax!sdcsvax!sdchema!donn 32 52' 30"N 117 14' 25"W (619) 452-4016 sdcsvax!sdchema!donn@noscvax