On 13 November 2014 04:20, Mark Boonie <boo...@us.ibm.com> wrote:
> - No more bunching: Perhaps a reasonable suggestion. Bear in mind, > though, that it would increase the repetitive nature of the document. > Also, the need to ensure that similar instructions were documented > similarly as much as possible would necessitate extensive reorganization > of the "internals" of the document so that similar descriptions were > essentially invoked as subroutines, in order to ensure that a change to, > say, ASI is exactly reflected in AGSI, AGFI, AFI, etc. One of the > hallmarks of PrincOps is a slavish devotion to consistency; do not break > that. > I was reading the RLL instruction today. The 4 paragraphs don't read well because they cover both RLL and RLLG and you select applicable phrases on the fly. While I have not tried, having two separate sections might make it more concise. It gets easier over time, and maybe it's less hard for native speakers. I don't think it affects correctness or completeness if you have more in the context. We don't repeat the rules for ordering of operands with each instruction, why not keep a "if we don't touch it, we don't break it" in the context rather than "bits 0-31 of general registers R1 and R3 remain unchanged" after we already learned that R3 was entirely unchanged. Clearly a lot of the text is by repetition and must have been generated out of something more concise and easier to verify. Changes could be limited to the program that generates the input for Bookie and that way ensure consistency. It would be sad (but not unlikely) when the modernization of the production process eliminated those options... Rob PS It's hard to avoid abuse. A former colleague once complained that an instruction was not documented (yes, I know there are some). But it turned out he never looked at anything but the examples in Appendix A. :-)