On Jan 25, 2022, at 11:03:23, Ed Jaffe wrote: >>> >> Fixing it should take precedence over documenting it. Through a half-century >> IBM has documented too many things that should have been fixed. > > Your assertion suggests the existing functionality broken. It's not. > > There is nothing wrong with 31-bit programs providing 18-word save areas. > Most do. > > And there are valid, documented save area formats called programs use to save > the high halves (and even the ARs) when invoked by "old school" callers. > > The key is proper documentation to understand what the called program's > environment looks like. That was arguably "broken" and will be corrected. > The defect is in the training burden for entry programmers and secular bloat of that documentation; a form of obsolesce.
64-bit should be the default and 31-bit or 24-bit quaint antiquities, akin to Roman numerals. When I write a C program I rarely worry about 64-bit versus older environments. Feature test macros in header files do the worrying for me. (Well, "rarely". Once on a 64-bit big-endian system I built a FOSS findutils, apparently targeted for 32-bit filesystem. File sizes appeared wrong; I accommodated. The problem would have been hidden on little-endian. I bet it's been fixed since.) -- gil