Paul Eggert wrote: > > It doesn't work because it takes only 32 samples of current_timespec (). > > In order to work reliably it would need 10 million samples. > > Thanks for the fix. My puzzlement was more why one needs 10 million > samples. Perhaps it's possible to alter the test somewhat so that 32 > samples would be enough, if we'd take them more cleverly.
Possibly. But even then, we would probably need to wait ca. 1 sec until we get the "right" samples. And waiting such a long time is not OK in a runtime function (as opposed to a configure test). Bruno