I noticed that repeating `make check' with 2.4g would report six or seven
failures, while the first run reported that all sixty checks were successful.

The problem comes from some files and directories left-over from the
previous run of `make check'.  The test suite should normally clean out
the files it creates, so I ought to have forgotten something somewhere.
If you find out, please tell me.  In the meantime, careful cleaning the
files test files in the tests subdirectory of the build structure should
allow you to successfully repeat the test suite.

Note that if some tests fail, then the failing tests will all be repeated,
automatically, once the bulk of the test suite has produced its statistics.
The repetition will also be a bit more verbose.  At the time of the
repetition, the files will _not_ be post-cleaned: this is on purpose, so they
can be examined.  If you have many failed tests, it may be a bit difficult
to sort out all left over files.  The simplest is to pick and repeat one
given test, and then study what it did.  For example, if test 37 failed,
just do `./FAIL-37.sh' to repeat that particular test without post-cleaning.

While debugging, you may edit file `FAIL-37.sh' all your want.  You may
also use `./FAIL-37.sh -n' to inhibit stdout and stderr redirections and
avoid checking stdout and stderr contents, or `./FAIL-37.sh -x' to _also_
get the shell to run commands verbosely.
