diff --git a/tests/openpgp/README b/tests/openpgp/README
index 9b384beb1..84faf1cdd 100644
--- a/tests/openpgp/README
+++ b/tests/openpgp/README
@@ -1,161 +1,175 @@
 #                                   Emacs, this is an -*- org -*- file.
 
 * How to run the test suite
-** using the legacy driver
-On POSIX you can just use
+From your build directory, run
 
-  $ make -C tests/openpgp check
+  obj $ make -C tests/openpgp check
 
-or
+to run all tests or
 
-  $ make -C tests/openpgp check XTESTS="setup.scm your-test.scm finish.scm"
+  obj $ make -C tests/openpgp check XTESTS=your-test.scm
 
-as before.
-** using the Scheme driver
+to run a specific test (or any number of tests separated by spaces).
+
+If you want to debug a test, add verbose=1 to see messages printed by
+spawned programs to their standard error stream, verbose=2 to see what
+programs are executed, or verbose=3 to see even more program output
+and exit codes.
+
+** Passing options to the test driver
+
+You can set TESTFLAGS to pass flags to 'run-tests.scm'.  For example,
+to speed up the test suite when bisecting, do
+
+  obj $ make -C tests/openpgp check TESTFLAGS=--parallel
+
+See below for the arguments supported by the driver.
+
+** Calling the test driver directly
 This is a bit tricky because one needs to manually set some
 environment variables.  We should make that easier.  See discussion
 below.  From your build directory, do:
 
-  obj $ srcdir=<path to>/tests/openpgp \
+  obj $ TMP=/tmp srcdir=<path to>/tests/openpgp \
         GPGSCM_PATH=<path to>/tests/gpgscm:<path to>/tests/openpgp \
         $(pwd)/tests/gpgscm/gpgscm [gpgscm args] \
         run-tests.scm [test suite runner args]
 
 *** Arguments supported by the test suite runner
 The test suite runner supports four modes of operation,
 {sequential,parallel}x{isolated,shared}.  You can select the mode of
 operation using a combination of the flags --parallel, --sequential,
 --shared, and --isolated.
 
 By default the tests are run in sequential order, each one in a clean
 environment.
 
 You can specify the tests to run as positional arguments relative to
 srcdir (e.g. just 'version.scm').  By default all tests listed in
 run-tests.scm are executed.  Note that you do not have to specify
 setup.scm and finish.scm, they are executed implicitly.
 
 The test suite runner can be executed in any location that the current
 user can write to.  It will create temporary files and directories,
 but will in general clean up all of them.
 *** Discussion of the various environment variables
 **** srcdir
 Must be set to the source of the openpgp test suite.  Used to locate
 data files.
 **** GPGSCM_PATH
 Used to locate the Scheme library as well as code used by the test
 suite.
 **** BIN_PREFIX
 The test suite does not hardcode any paths to tools.  If set it is
 used to locate the tools to test, otherwise the test suite assumes to
 be run from the build directory.
 **** MKTDATA and GPG_PRESET_PASSPHRASE
 These two tools are not installed by 'make install', hence we need to
 explicitly override their position.  In fact, the location of any tool
 used by the test suite can be overridden this way.  See defs.scm.
 **** argv[0]
 run-tests.scm depends on being able to re-exec gpgscm.  It uses
 argv[0] for that.  Therefore you must use an absolute path to invoke
 gpgscm.
 * How to write tests
 gpgscm provides a number of functions to aid you in writing tests, as
 well as bindings to process management abstractions provided by GnuPG.
 For the Scheme environment provided by TinySCHEME, see the TinySCHEME
 manual that is included in tests/gpgscm/Manual.txt.
 
 For a quick start, please have a look at various tests that are
 already implemented, e.g. 'encrypt.scm'.
 ** The test framework
 The functions info, error, and skip display their first argument and
 flush the output buffers.  error and skip will also terminate the
 process, signaling that the test failed or should be skipped.
 
 (for-each-p msg proc list) will display msg, and call proc with each
 element of list while displaying the progress appropriately.
 for-each-p' is similar, but accepts another callback before the 'list'
 argument to format each item.  for-each-p can be safely nested, and
 the inner progress indicator will be abbreviated using '.'.
 ** Temporary files
 (lettmp <bindings> <body>) will create and delete temporary files that
 you can use in <body>.  (with-temporary-working-directory <body>) will
 create a temporary director, change to that, and clean it up after
 executing <body>).
 
 make-temporary-file will create a temporary file.  You can optionally
 provide an argument to that function that will serve as tag so you can
 distinguish the files for debugging.  remove-temporary-file will
 delete a file created using make-temporary-file.
 
 ** Monadic transformer and pipe support
 Tests often perform sequential transformations on files, or connect
 processes using pipes.  To aid you in this, the test framework
 provides two monadic data structures.
 
 (Currently, the implementation mashes the 'bind' operation together
 with the application of the monad.  Also, there is no 'return'
 operation.  I guess all of that could be implemented on top of
 call/cc, but it isn't at the moment.)
 *** pipe
 The pipe monad constructs pipe lines.  It consists of a function
 pipe:do that binds the functions together and manages the execution of
 the child processes, a family of functions that act as sources, a
 function to spawn processes, and a family of functions acting as
 sinks.
 
 Sources are pipe:open, pipe:defer, pipe:echo.  To spawn a process use
 pipe:spawn, or the convenience function pipe:gpg.  To sink the data
 use pipe:splice, or pipe:write-to.
 
 Example:
 
   (pipe:do
     (pipe:echo "3\n1\n2\n")
     (pipe:spawn '("/usr/bin/sort"))
     (pipe:write-to "sorted" (logior O_WRONLY O_CREAT) #o600))
 
 Caveats: Due to the single-threaded nature of gpgscm you cannot use
 both a source and sink that is implemented in Scheme.  pipe:defer and
 pipe:echo are executing in gpgscm, and so does pipe:splice.
 *** tr
 The transformer monad describes sequential file transformations.
 
 There is one source function, tr:open.  To describe a transformation
 using some process, use tr:spawn, tr:gpg, or tr:pipe-do.  There are
 several sinks, although sink is not quite the right term, because the
 data is not consumed, and hence one can use them at any position.  The
 "sinks" are tr:write-to, tr:call-with-content, tr:assert-identity, and
 tr:assert-weak-identity.
 
 A somewhat contrived example demonstrating many functions is:
 
   (tr:do
     (tr:pipe-do
       (pipe:echo "3\n1\n2\n")
       (pipe:spawn '("/usr/bin/sort")))
     (tr:write-to "reference")
     (tr:call-with-content
      (lambda (c)
        (echo "currently, c contains" (string-length c) "bytes")))
     (tr:spawn "" '("/usr/bin/gcc" -x c "-E" -o **out** **in**))
     (tr:pipe-do
       (pipe:spawn '("/bin/grep" -v "#")))
     (tr:assert-identity "reference"))
 
 Caveats: As a convenience, gpgscm allows one to specify command line
 arguments as Scheme symbols.  Scheme symbols, however, are
 case-insensitive, and get converted to lower case.  Therefore, the -E
 argument must be given as a string in the example above.  Similarly,
 you need to quote numerical values.
 ** Process management
 If you just need to execute a single command, there is (call-with-fds
 cmdline infd outfd errfd) which executes cmdline with the given file
 descriptors bound to it, and waits for its completion returning the
 status code.  There is (call cmdline) which is similar, but calls the
 command with a closed stdin, connecting stdout and stderr to stderr if
 gpgscm is executed with --verbose.  (call-check cmdline) raises an
 exception if the command does not return 0.
 
 (call-popen cmdline input) calls a command, writes input to its stdin,
 and returns any output from stdout, or raises an exception containing
 stderr on failure.
 * Sample messages