Jeremy C. Reed wrote:
> This is why I think it won't work (easily). On some of my systems, linking
> binaries may take many minutes and sometimes over half an hour. Then there
> may be many lines of output that are completed instantly. I could imagine
> a progress bar quickly getting to 95-percent and then sitting for a long
> time on some steps at the end.

Isn't this what we're all used to? ;-)

For better linearity you would need to time each step of at least one example
build and use those times for calculating the estimates -- this is what I
meant by recording timestamps in my first post. And I don't think there
needed to be a database with build times for each and every processor out
there, maybe just the result of one typical setup that would be "scaled"
for the current run.

Of course we're no longer talking of a simple pipe ending in progress(1)
here... (Which, IMO, should still be good enough in most cases.)

Regards,
Tom