The function AC_ARG_ENABLE provided by autoconf is a very bare-bones
function, which needs a lot of boilerplate code to do the right thing.
We have duplicated this code all over the place, often with bugs or
omissions of functionality.
This patch introduces a new UTIL_ARG_ENABLE which sets up a
best-practices framework for handling AC_ARG_ENABLE, and allows users to
just call this and be sure we do the right thing. The code in
UTIL_ARG_ENABLE is definitely non-trivial, but I've tried hard to make
it as documented and clear as possible. In any case, I think the
drawbacks from having a complex implementation like this is more than
compensated for by the added simplicity for all the places it can be
trivially used.
I have also tried to keep the difference introduced in the JVM features
patch between "available" and "enabled", so it's clear if a feature is
not enabled because it is not possible to enable, or just because it is
not the default.
I have also made some usability fixes to UTIL_DEFUN_NAMED. The testing I
added could certainly have been more comprehensive, but at least it's a
start. The real good testing would be to be able to run configure with
different arguments, but that needs another kind of testing framework
than what we got so far. I also have expanded the set of assert
statements available for testing.
This cleanup highlights the major differences in how we handle this kind
of flags, for instance if we set the flag value to "true" or "yes" in
spec.gmk. I have opted not to touch any of that in this patch, but I
feel a follow-up cleaning might be warranted.
This change should not have any impact on normal use cases, but the
added checks that command line options are correct might catch incorrect
syntax that was previously ignored or handled incorrectly.
Bug: https://bugs.openjdk.java.net/browse/JDK-8240820
WebRev:
http://cr.openjdk.java.net/~ihse/JDK-8240820-introduce-UTIL_ARG_ENABLE/webrev.01
/Magnus
