On 20 May 2016 at 17:24, Liam Proven <[email protected]> wrote: > > On 18 May 2016 at 21:40, Fred Cisin <[email protected]> wrote: > > But, "Moore's Law" held that it wouldn't be much longer. > > Just one doubling of the speed of the Lisa's hardware would have been enough > > to silence the speed complaints. > > A general point, really. > > One of Microsoft's strokes of brilliance was selectively exploiting > this. I think maybe it learned it from the 80286 OS/2 1.x débacle. > > NT 3.1 was brilliant if a bit bulky and unoptimised. Fair enough, it > was a v1.0 OS. It was way way WAY too heavy for the average 1993 PC, > but power users played, partly 'cos it fixed serious problems with > Windows 3.1. > > [[make it work, ship version, then make it fast, then ship new version]]
I don't know if that was a specific market ploy based on Moore's Law, or just the generally accepted practice of getting an initial version with the API working any which way, then refactoring to improve performance/correctness in later versions. I seem to recall a comment that when select() was introduced in BSD4.2 it was implemented by an in kernel loop of polling each filedescriptor so they could ship something with the API they wanted, and then in 4.3 the implementation was refactored to... how should we say... be much more performant :)
