the recent discussions of regressions, and with people proposing patches that
seemed buggy, makes me think the test framework is not easy to run, or that
comprehensive.  seems like we should be able to configure newlib for native
builds (i.e. non-cross-compiling), and `make check` should "just work", and
we should require people add/improve tests when changing core functions.

anyone have experience with the existing framework ?
-mike