* Built-in Tests

'make check' will use the Python script run-tests to run the built-in
tests which all in the tests/ directory.  These test the server
against a separate though not very independent client implementation.

** Interface To Tests

Each of these scripts is an input to the sftpclient program.  They are
invoked with both the client and the server sitting in the same
(initially empty) directory; if your test requires separate
directories then use !mkdir and cd/lcd.

Any line in the script starting with a "#" is a regular expression.
The script is expected to produce output matching the regexp at this
point.  For example, mkdir3456 starts off as follows:

  mkdir defperms
  ls -ld defperms
  #drwxr-[xs]r-x +[\?\d] +\S+ +\S+ +\d+ +[a-zA-Z]+ +\d+ +\d+:\d+ defperms

Here we create a directory with default permissions and then list.
The regexp matches the expected ls output.

** Per-Protocol Tests

The filename of the scripts is significant.  It should contain the
protocol version numbers it applies to.  For instance mkdir3456
applies to all four known protocol versions.  Similarly ls3, ls45 and
ls6 provide similar tests against different protocol versions.

ls3, ls45 and ls6 are testing stat and attributes in detail so it is
worth producing separate scripts for each of them.  mkdir3456 however
compares more about the mkdir and rmdir operations so there is not
much point in having separate versions for each protocol; therefore
its regexps will match the output produced in all the different
versions.

The other reason for having protocol-specific tests is of course to
test facilities that only exist in a subset of the protocols.

** Hit-List

Things that still particularly want testing:
   * read-write opens
   * serialization logic
   * encoding translation
   * reverse symlink logic
   * non-regular files (in lots of ways)
   * invalid UIDs/GIDs and owner/group names
   * run tests under valgrind

Untested but lower priority:
   * "anonymous sftp" code
   * the client (if I decide to promote it to a proper client rather
     than a test tool)

* Interoperability Tests

** Programmable Clients

It is possible in principle to do something similar to the above for
command line clients such as lftp, psftp and the OpenSSH client,
though I have not done so yet.  Similarly it should be possible to
write mechanical tests for libraries such as Paramiko or even to
repurpose their own tests, where they exist.

** Graphical Clients

Graphical clients such as WinSCP are less convenient to test this way;
instead it is necessary to build up a library of activities which
tests as many protocol features as possible in the minimum number of
user actions.  This hasn't been done yet l-)