I've just attempted to build minheadless on win32 in
build.minheadless.cmake. It failed, but it took a long time to do so,
while I waited. So I cleaned and repeated to measure just how long I
waited. So this time is a best case; all read files are in cache etc. I'm
running a Windows VM on a top of the line MacBook Pro and yet it takes 3
minutes and 20 seconds to configure and then start to build and fail
because it can't find sqplatfozrmspecifc-win32.h. I have to wait 3 minutes
and 20 seconds before I can find out a missing file. This is *wrong*.
Using make on a build slave is fine, if you insist, but clearly not
necessary (our current builds do not use cmake, and even then on ARM build
slaves they sometimes timeout). But for development this is madness. We
should have a build system which is reactive, which gives feedback too the
developer quickly, not after 3 minutes and 20 seconds on state of the art
I find it absurd that a Smalltalk community, which well understands the
value of incrementally and reactivity, can be satisfied with a VM build
system that takes 3 minutes and 20 seconds before it does anything useful.
It feels like being back at York University in the early '80's, waiting for
the PDP 10. No, that was faster. On reflection, it reminds me of having
to post coding sheets to be typed in by entry clerks through the mail while
I was at school in the '70's. Back to the future indeed.