The Greg testing framework

for Greg Version 1.4

February 2001

Richard Frith-Macdonald <rfm@gnu.org>

Copying

See the file `COPYING.LIB'.

New in this release

New in 1.4

Updated to save/restore expect settings at start of a tool and end of a tool so that they can be readily set up for the duration of a tool test.
The default action on reading an end of file from a child is now to make the test FAIL rather than be unresolved.
The default action on having a timeout when reading from a child is now to make the test FAIL rather than be unresolved.
Added hooks for pre and post testcase procedures.

New in 1.3

Add greg-directory to return the absolute path to the directory in which the executing test script is located.

New in 1.2

Permit the files in the greg-files variable to use the `.scm' extension or not.
Added greg-paths variable to permit direct running of named files rather than using the DejaGNU-like tools directory structure.
Alter the `greg' script to set the greg-paths variable to contain any files listed on the command-line.
Added greg-expect-pass and greg-expect-fail macros.

New in 1.1

Changed from using GNU Library General Public License to the newer GNU Lesser General Public License.
Changed from using a compiled module (deprecated in guile) to using a dynamic library for pty support.

New in 1.0

Nothing really - a few very minor fixes - no bug reports, which is why I decided to risk bumping the version number to 1.0 - presumably people will now find hundreds of bugs...
Updated to work with Guile and 1.3.4

New in 0.7

Updated to work with Guile 1.3.2 and 1.3.3
Updated to work on GNU/Linux systems with pts rather than ptys
Improved configuration code

New in 0.6

Converted makefiles to work without GNUstep
Improved configuration code
Improved documentation somewhat
Updated to work with guile version 1.3

New in 0.5

greg-testcase now returns a useful result.
Fixed bug in creating child processes on some systems.

New in 0.4

Allow an empty string as the program name for creating a child process - in which case, we fork without execing to another program. Modified Greg to allow (quit) to exit from the test system.

New in 0.3

Improved installation - now checks for location of guile shell.
Fixed a couple of minor errors in the documentation.

New in 0.2

Ported to Unixware 2.1.3 - should work for any SysV4.2 based unix.

General news

This release of Greg provides a test framework much like that of DejaGNU, but also provides Guile modules to permit `embedded' testing of applications that use Guile as a scripting language and libraries that are directly accessible to Guile.

This release has been tested with Guile-1.3.4 and 1.3.3, it will certainly not run on versions earlier than 1.3.1

Apologies to Guile/Scheme/Lisp programmers out there - I came to this from Objective-C programming and taught myself during the four weeks that I wrote GNUstep-Guile and Greg - so my code is probably really ugly - but it does seem to work.

The code to run a child process in a pseudo-terminal works for GNU/Linux and SysV4.2 systems - it probably needs work to make it more portable - Please email bugfixes and comments to <bug-gnustep@gnu.org> or directly to me <rfm@gnu.org>

Greg in brief

Greg is a framework for testing other programs and libraries. Its purpose is to provide a single front end for all tests and to be a small, simple framework for writing tests. Greg leverages off the Guile language to provide all the power (and more) of other test frameworks with greater simplicity and ease of use.

The simplicity of the Greg framework makes it easy to write tests for any program, but it was specifically written for use with GNUstep-Guile to permit direct testing of the GNUstep libraries without the necessity to run a separate driver program.

The core functionality of Greg is a Guile module which can be loaded into any software with an embedded Guile interpreter. Any program which uses Guile as it's scripting language can therefore use Greg to test itself directly!

For testing external programs, Greg provides a compiled module that may be dynamically linked into Guile to permit you to run an application as a child process on a pseudo-terminal. In conjunction with the standard Guile `expect' module, this lets you test external programs.

Also provided is `greg' - a Guile script to invoke the Greg test framework in much the same way that runtest is used in DejaGNU.

All tests have the same output format (enforced by the greg-testcase procedure). Greg's output is designed to be both readable and readily parsed by other software, so that it can be used as input to customised testing processes.

Greg provides most of the functionality of DejaGNU but is rather simpler. It omits specific support for cross-platform/remote testing since this is really rather trivial to add where required and tends to vary from site to site so much that an attempt at a generic solution is pretty pointless. What Greg does do, is provide hooks to let you easily introduce site specific code for handling those sorts of situations.

The current version of Greg can normally be found on GNU ftp sites, with documentation online at http://www.gnu.org/software/greg/gregdoc.html

or, for the bleeding edge - availably by anonymous cvs as part of the GNUstep-Guile package in the GNUstep project -

CVSROOT=":pserver:anoncvs@cvs.net-community.com:/gnustep"
export CVSROOT
cvs login (password is `anoncvs')
cvs -z3 checkout guile

How to run a Greg testsuite

To run tests from an existing collection, try running

make check

If the check target exists, it usually saves you some trouble--for instance, it can set up any auxiliary programs or other files needed by the tests.

Alternatively, if you are in the top-level source directory of an existing testsuite (ie. there are subdirectories containing files with a `.scm' extension), you can get the `greg' script to test all the tools in the directory by typing -

greg

Finally, if you just want to run the tests that are in a specific file (or files), you can get the `greg' script to run them simply by listing the files on the command line.

greg a-file-to-run another-file-to-run

or, for verbose output -

greg --verbose a-file-to-run another-file-to-run

If you have a test suite that is intended to be used for `embedded' testing - You need to start the application to be tested, gain access to it's Guile command line (or other guile interface) and enter the commands -

(use-modules (ice-9 greg))
(greg-test-all)

A trivial example of a testcase

Each Greg test is a Guile script; the tests vary widely in complexity, depending on the nature of the tool and the feature tested.

;
; GNUstep-guile interface library test
;
; Create an object using the NSString [stringWithCString:] method and
; check that the resulting object is of the correct class.
;
(greg-testcase "The `stringWithCString:' method creates an NSString object" #t
(lambda ()
  (define obj ([] "NSString" stringWithCString: "Hello world"))
  (gstep-bool ([] obj isKindOfClass: ([] "NSString" class)))
))

Though brief, this example is a complete test. It illustrates some of the main features of Greg test scripts:

The test case does not deal with accessing the library - the Guile interpreter in use is linked with the library. If you were testing an external application which needed to be started up as a child process running on a pseudo-terminal, the process startup would have been performed in a separate `begin.grg' script.
You use the Guile programming language to write the script and, in this case, the builtin macros and procedures (such as gstep-bool and [] provided by the GNUstep-Guile library.
Test cases use the procedure greg-testcase to record the test outcome. You pass this procedure a string (`assertion') describing the testcase, a boolean specifying the expected outcome, and a `thunk' (parameterless procedure) that performs the actual test.
The `thunk' is normally a `lambda expression' - you can look these up in the Guile/Scheme documentation - or just follow the example.
As much code as possible is placed inside the `thunk' passed to greg-testcase so that, in the event of an error, the greg-testcase procedure can trap it and report an unresolved test.
The code - (define obj ([] "NSString" stringWithCString: "Hello world")) could have appeared outside the testcase, but that would have been less safe.

Here is the same example in a slightly different form - using the greg-expect-pass macro -

;
; GNUstep-guile interface library test
;
; Create an object using the NSString [stringWithCString:] method and
; check that the resulting object is of the correct class.
;
(greg-expect-pass "The `stringWithCString:' method creates an NSString object"
  (define obj ([] "NSString" stringWithCString: "Hello world"))
  (gstep-bool ([] obj isKindOfClass: ([] "NSString" class)))
)

Why Greg does what it does

Greg was written to support regression testing for the GNUstep libraries. It was inspired by an earlier test framework (by Ovidiu Predescu) that used DejaGNU along with a `driver' program (to make the calls to the library) and a suite of TcL scripts to control the driver.

There were three main problems (inherent in the nature of DejaGNU) with that approach -

Writing/Debugging tests was a pain - errors in the TcL code of a testcase could all too easily interfere with the TcL routines that interfaced to the driver program.
Solution - use another language where errors in testcases are localised.
Writing/Debugging tests was a pain - you needed to attach gdb to the driver program that was running as a child of the DejaGNU process in order to debug errors in the driver/library under test.
Solution - `embedded testing' make the test framework and the code under test part of the process, so it's easy to run the whole thing under gdb.
The driver program and TcL scripts required quite a bit of support when adding new tests.
Solution - `embedded testing' we needed to remove the intermediary layers and test as directly as possible.

So, something different was required, a test framework in a safer, simpler language that made it easy to create thin interfaces to libraries, so simplifying the task of producing testcases.

Of course, the good points of DejaGNU needed to be retained - clear output, Posix compliance, the ability to test separate programs as well as libraries.

A couple of additional goals seemed worthwhile -

Make the test framework flexible so that applications could easily use it for self-testing.
Give the framework a similar structure to, and a user interface like that of DejaGNU, so that people with a background in testing could easily adopt it.

Hopefully, Greg meets all its goals.

A POSIX conforming test framework

This section is copied almost directly from the DejaGNU documentation with minor modifications -

Greg is believed to conform to the POSIX standard for test frameworks.

POSIX standard 1003.3 defines what a testing framework needs to provide, in order to permit the creation of POSIX conformance test suites. This standard is primarily oriented to running POSIX conformance tests, but its requirements also support testing of features not related to POSIX conformance.

The POSIX documentation refers to assertions. An assertion is a description of behavior. For example, if a standard says "The sun shall shine", a corresponding assertion might be "The sun is shining." A test based on this assertion would pass or fail depending on whether it is daytime or nighttime. It is important to note that the standard being tested is never 1003.3; the standard being tested is some other standard, for which the assertions were written.

As there is no test suite to test testing frameworks for POSIX 1003.3 conformance, verifying conformance to this standard is done by repeatedly reading the standard and experimenting. One of the main things 1003.3 does specify is the set of allowed output messages, and their definitions. Four messages are supported for a required feature of POSIX conforming systems, and a fifth for a conditional feature. Greg supports the use of all five output messages; in this sense a test suite that uses exactly these messages can be considered POSIX conforming. These definitions specify the output of a test case:

PASS

A test has succeeded. That is, it demonstrated that the assertion is true.

UPASS

POSIX 1003.3 does not incorporate the notion of unexpected passes, so for strict POSIX, PASS, instead of UPASS, is returned for test cases which were not expected to pass but did. This means that PASS is in some sense more ambiguous than if UPASS is also used.

FAIL

A test has produced the bug it was intended to capture. That is, it has demonstrated that the assertion is false. The FAIL message is based on the test case only. Other messages are used to indicate a failure of the framework.

XFAIL

POSIX 1003.3 does not incorporate the notion of expected failures, so for strict POSIX, FAIL, instead of XFAIL, is returned for test cases which were expected to fail and did not. This means that FAIL is in some sense more ambiguous than if XFAIL is also used.

UNRESOLVED

A test produced indeterminate results. Usually, this means the test executed in an unexpected fashion; this outcome requires that a human being go over results, to determine if the test should have passed or failed. This message is also used for any test that requires human intervention because it is beyond the abilities of the testing framework. Any unresolved test should resolved to PASS or FAIL before a test run can be considered finished. Note that for POSIX, each assertion must produce a test result code. If the test isn't actually run, it must produce UNRESOLVED rather than just leaving that test out of the output. With Greg this is not a problem - any unexpected termination of a greg-testcase procedure will produce UNRESOLVED. Here are some of the ways a test may wind up UNRESOLVED:

A test's execution is interrupted.
A test does not produce a clear result. This is usually because there was a Guile exception thrown while processing the test. This usually requires a human being to examine the output to determine what really happened--and to improve the test case.
A test depends on a previous test, which fails.
The test was set up incorrectly and failed to produce a boolean value.

UNTESTED

A test was not run. This is a placeholder, used when there is no real test case yet.

The only remaining output message left is intended to test features that are specified by the applicable POSIX standard as conditional:

UNSUPPORTED: There is no support for the tested case. This may mean that a conditional feature of an operating system, or of a compiler, is not implemented.

Greg uses the same output procedures to produce these messages for all test suites, and these procedures are already known to conform to POSIX 1003.3. For a Greg test suite to conform to POSIX 1003.3, you must set the variable greg-posix to be true (or run the `greg' command with the --posix flag). Doing this will ensure that non-posix extensions are not used.

Installing Greg

Requirements

Greg needs to have Guile installed. It should work with Guile-1.3 or later. You need to have the `guile' program in your path in order for the installation process to determine the proper locations for things.

You can get Guile from any GNU ftp site.

The current version of Greg can normally be found on GNU ftp sites, with documentation online at http://www.gnu.org/software/greg/gregdoc.html

or, for the bleeding edge - availably by anonymous cvs from the GNUstep project (http://www.gnustep.org/) -

CVSROOT=":pserver:anoncvs@cvs.net-community.com:/gnustep"
export CVSROOT
cvs login (password is `anoncvs')
cvs -z3 checkout guile/Greg

Building

To build Greg -

Type ./configure in the main Greg directory to configure for your system.

Once configuration is complete, go into the `Library' subdirectory and type make install to build and install things.

You should end up with -

A dynamic library, which can be dynamically linked into Guile with (if (not (feature? 'greg-pty)) (dynamic-call "scm_init_greg" (dynamic-link "libgreg.so")))

a module defining Guile procedures and variables providing the main test framework, which can be accessed using (use-modules (ice-9 greg)),

and a Guile script that you can use to run tests from the unix command-line (`greg').

You MUST install Greg before you attempt to use it (or run it's self-tests) because the Guile modules making it up must be in place in the standard Guile directories before Greg can work.

Once Greg is installed, you can type make check in the Tests directory to get Greg to test itself.

You can type make in the Documentation directory to build the documentation in info, html and dvi formats.

NB.
You must have the `makeinfo' program installed to build the documentation in info format
You must have the `texi2html' program installed to build the documentation in html format
You must have the `texi2dvi' program installed to build the documentation in dvi format

Problems

Greg is quite simple, so there is not much to go wrong with it. Of course, you must have a working copy of Guile installed, and you need to make sure you ran the configure script to configure Greg for your system, and installed Greg, but after that, most stuff should just work.

The single area where you are most likely to encounter problems is if you are using Greg to test external programs run in a child process using the (greg-child) procedure.

If you install Greg in a non-standard location, it's possible that guile will not be able to find the libgreg.so dynamic library to load it. In this case, you will need to change/set the LD_LIBRARY_PATH environment variable (or whatever other environment variable your operating system uses to list the places in which dynamic libraries may be found) to include the directory in which the library has been installed.
The C code for this procedure is quite system-dependent and has only been tested on a few operating systems. If you have problems with this, look at `greg.c' in the Library directory.

Please attempt to make a patch to fix things on your operating-system and send it to me - <rfm@gnu.org> or to <bug-gnustep@gnu.org>

Using `greg'

The Greg framework is designed to be used in two ways - as an embedded system from within any application which is linked with the Guile library, or stand-alone using the command-line `greg' driver script. For both of these methods of usage the test cases are written the same way and the expected output is the same.

Output

While Greg may produce more verbose output in response to various settings, the basic output of a test run is a series of lines describing the success/failure state of each testcase encountered, followed by a summary of all test cases.

In `normal' mode, only unexpected results are displayed, but in `verbose' output mode, results for all results are displayed.

`greg' flags the outcome of each test case with a line consisting of one of the following codes followed by a colon, a space, and then the testcase description.

PASS: The most desirable outcome: the test succeeded, and was expected to succeed.
UPASS: A pleasant kind of failure: a test was expected to fail, but succeeded. This may indicate progress; inspect the test case to determine whether you should amend it to stop expecting failure.
FAIL: A test failed, although it was expected to succeed. This may indicate regress; inspect the test case and the failing software to locate the bug.
XFAIL: A test failed, but it was expected to fail. This result indicates no change in a known bug. If a test fails because the operating system where the test runs lacks some facility required by the test, the outcome is UNSUPPORTED instead.
UNRESOLVED: Output from a test requires manual inspection; the test suite could not automatically determine the outcome. For example, your tests can report this outcome is when a test does not complete as expected.
UNTESTED: A test case is not yet complete, and in particular cannot yet produce a PASS or FAIL. You can also use this outcome in dummy "tests" that note explicitly the absence of a real test case for a particular property.
UNSUPPORTED: A test depends on a conditionally available feature that does not exist (in the configured testing environment). For example, you can use this outcome to report on a test case that does not work on a particular target because its operating system support does not include a required subroutine.

Files and directories

A Greg test run expects to find files and directories in a certain layout (modeled on that used by DejaGNU) - though it is possible to override this DejaGNU compatibility feature and simply run the tests in a list of files.

The test source directory (normally your current directory) is expected to contain one or more tool directories. Each tool directory should contain one or more test scripts. In fact any file in a tool directory which has a `.scm' extension is assumed to be a Guile test script.

When a normal Greg test run is done, Greg goes through each tool directory in turn and loads each test script in turn.

You may set the Guile variable greg-tools or use the --tool ... command-line option to specify a list of tools directories to use rather than assuming that all subdirectories are tool directories. If you do this, the tools are tested in the order in which they appear in the list rather than the default order (ASCII sorted by name).

You may set the Guile variable greg-files or use the --file ... command-line option to specify a list of file names to use rather than assuming that all `.scm' files in each tool directory are test scripts. If you do this, the files are loaded in the order in which they appear in the list. You may omit the `.scm' extension from filenames and Greg will supply it for you if necessary.

You may set the Guile variable greg-paths to specify a list of test files to be run directly, or simply list the files to be run on the command-line.

Doing this overrides the greg-tools and greg-files variables, and simply runs the files you list in the order you list them.

As a (minor) complication to this simple layout, Greg permits the use of `begin.grg' and `end.grg' scripts in both the main source directory and in each tool directory. These scripts permit you to add any initialisation and cleanup code you want. Typically (for non-embedded testing) you would use a `begin.grg' script to start the application to be tested.

If `begin.grg' exists in the main source directory, it will be loaded before any tools are tested.

If `end.grg' exists in the main source directory, it will be loaded after all the tools are tested.

If `begin.grg' exists in a tool directory, it will be loaded before any test scripts in that directory are loaded.

If `end.grg' exists in a tool directory, it will be loaded after all the test scripts in that directory are loaded.

NB. Even when you use the greg-paths variable to run one or more test files directly, the `begin.grg' and `end.grg' files in your current directory will be loaded.

Embedded usage

Greg is designed primarily for embedded usage - any application that uses Guile as it's scripting language should be able to use Greg to test itself.

To this end - Greg provides a Guile module containing definitions of various procedures (used to run tests) and variables (used to modify the behavior of a test run).

Before trying to use any part of Greg, You need to load the Greg module with (use-modules (ice-9 greg))

The main procedure to run Greg tests is (greg-test-run). You can use this to run tests in much the same way as the `greg' script is used to run tests from the command-line. The behavior of this procedure is modified by setting the following top-level variables -

greg-debug The greg-debug variable is a boolean used to determine whether to output extra debug information. The debug information is written to a `.dbg' file. Test scripts may use the greg-dlog procedure to output debug information.
greg-files The greg-files variable is a list of strings used to specify the names of the test scripts to be run. If this list is empty, all the files with the `.scm' extension in a tool subdirectory are loaded. If the list contains names, the named scripts are loaded in the order in which they occur in the list rather than the default order (ASCII sorted by name).
The exact names of scripts in this list will be used to match filenames, and if no match is found a `.scm' extension will be added automatically by Greg and a re-match attempted.
greg-obj-dir The greg-obj-dir variable is a string naming the directory in which Greg expects to find binaries to be run as child processes.
By default, this is the current directory (.).
greg-out-dir The greg-out-dir variable is a string naming the directory in which Greg will generate log and debug output files.
By default, this is the current directory (.).
greg-paths The greg-paths variable is a list of pathname of files to be run. This is used where you don't want to use the normal tools hierarchy and need to test specific files explicitly. The filenames are used exactly as given. If you use this variable, you should set the greg-tools variable to an empty list.
greg-posix The greg-posix variable is a boolean used to determine whether to produce strict posix output, or permit non-posix extensions.
greg-src-dir The greg-src-dir variable is a string naming the directory in which Greg will look for tool subdirectories and where it expects to find the main initialisation and cleanup scripts (`begin.grg' and `end.grg') for a test run.
By default, this is the current directory (.).
greg-tools The greg-tools variable is a list of strings used to specify the names of the tools to be tested (ie the names of the subdirectories of the main source directory). If this list is empty, all the subdirectories of the main source directory (normally .) are assumed to be valid tool directories. If the list contains names, the named subdirectories are used in the order in which they occur in the list.
greg-verbose The greg-verbose variable is a number used to determine the level of detail produced in the log output. At zero (the default), only minimal information is output (unexpected testcase results and a summary of the number of testcases passed). At one, the result of each testcase is reported and a more detailed summary is output. At two, entry and exit of tool directories and test files is reported.

Command-line usage

`greg' is the executable test driver for Greg. This is a Guile script that you can use to run tests from the command line. The command-line options that you can pass to `greg' are listed below.

`greg' returns an exit code of 1 if any test has an unexpected result; otherwise (if all tests pass or fail as expected) it returns 0 as the exit code.

This is the full set of command line options that `greg' recognizes.

greg --tool tool ...
[ --debug ]
[ --file script ... ]
[ --help ]
[ --objdir path ]
[ --outdir path ]
[ --posix ]
[ --srcdir path ]
[ -v | --verbose ]  [ -V | --version ]
[ files to run ]

--tool tool ...: tool specifies what set of tests to run. It provides a list of subdirectories (each corresponding to a tool) in which test scripts can be found. Initialisation code (including executable tool startup) for each tool may be placed in `begin.grg' in the appropriate tool subdirectory. Cleanup code may be placed in `end.grg'. For example, including --tool gcc on the `greg' command line runs tests from the gcc subdirectory. The order in which the tools are tested will be the same as the order in which the tool names occur on the command line.
--file script ...: Specify the names of testsuites to run. By default, `greg' runs all tests for the tool, but you can restrict it to particular testsuites by giving the names of the `.scm' Guile scripts that control them. You do not need to supply the `.scm' file extension - it will be assumed. testsuite may not include path information; use plain filenames. The order in which the test scripts are run will be the same as the order in which the script names occur on the command line.
--debug: Turns on the expect internal debugging output. Debugging output is displayed as part of the `greg' output, and logged to a file called `tests.dbg'. The extra debugging output does not normally appear on standard output.
--help: Prints out a short summary of the `greg' options, then exits (even if you also specify other options).
--objdir path: Use path as the top directory containing any auxiliary compiled test code. This defaults to `.'. Use this option to locate pre-compiled test code. You can normally prepare any auxiliary files needed with make.
--outdir path: Write output logs in directory path. The default is ., the directory where you start `greg'. This option affects only the log and the debug files `tool.log' and `tool.dbg'.
--srcdir path: Use path as the top directory for test scripts to run. `greg' looks in this directory for any subdirectory whose name matches any toolname (specified with --tool).
--verbose
-v: By default, `greg' shows only the output of tests that produce unexpected results; that is, tests with status FAIL (unexpected failure), UPASS (unexpected success), or ERROR (a severe error in the test case itself). Specifying --verbose to see output for tests with status PASS (success, as expected) and XFAIL (failure, as expected). It also causes a more detailed summary to be displayed.
Specifying --verbose more than once causes more detail to be displayed.
--version
-V: Prints out the version numbers of Greg, and Guile, then exits without running any tests.

Writing tests

getting started

The simplest way to get started is to write a file (say `myTests') and type greg --verbose myTests to run the tests in it.

Your file might contain code like -

; A simple test that basic arithmetic works

(greg-expect-pass "one plus one is two"
  (eq? (+ 1 1) 2)
)

(greg-expect-pass "one minus one is zero"
  (eq? (- 1 1) 0)
)

And would produce output like -

PASS: one plus one is two
PASS: one minus one is zero

# of testcases attempted   2
# of expected passes       2
# of expected failures     0
# of unexpected passes     0
# of unexpected failures   0
# of unresolved testcases  0
# of unsupported testcases 0
# of untested testcases    0

types of testsuite

There are three types of situation where Greg may be used as a test framework -

Embedded testing
In this case, you have an application that uses Guile as it's scripting language - so you have provided an interface between the internals of your program and the Guile interpreter. Your testcases can use that interface to test your program directly, and you run the tests by loading the Greg framework and then doing (greg-test-run) A testsuite for embedded use does not normally need `begin.grg' or `end.grg' since the application to be tested will be running the tests on itself - so it doesn't need to control child processes.
Library testing
When you want to test a library, you need to write a tiny program to start up the Guile command-line interpreter with the library to be tested inked in to the program via a via a set of stubs. The resulting program can then be used to run tests on the directly, just like normal embedded testing. There is a tool called G-Wrap (by Chris Lee), which will generate stubs to import your library functions into Guile. The home page of this tool is at http://www.cs.cmu.edu/~chrislee/g-wrap/ For testing Objective-C libraries, you should use the GNUstep-Guile interface library (documented online at http://www.tiptree.demon.co.uk/gstep/guile/gstep_guile_toc.html) which provides a fairly complete interface between the Guile and Objective-C languages (using the GNU objective-C runtime). You can get hold of this as part of the GNUstep core package (from any GNU ftp site) or, for the bleeding edge - by anonymous cvs as the GNUstep-Guile package in the GNUstep project -
```
CVSROOT=":pserver:anoncvs@cvs.net-community.com:/gnustep"

export CVSROOT

cvs login (password is `anoncvs')

cvs -z3 checkout guile
```
External testing
When you want to test an external program, you use Greg in much the same way that you would use DejaGNU. That is - you run the program to be tested as a separate child process and your test cases consist of Guile code to send commands to the child process and wait for responses from it. The child process is normally started by a `begin.grg' script using the (greg-child) procedure, though it is of course possible to start and stop child processes in each test script.

testsuite file layout

A testsuite for a tool is a directory containing one or more test script files and (optionally) `begin.grg' and `end.grg' files to handle initialisation and cleanup.

Each script file has a `.scm' extension and contains Guile (Scheme) code, but you do not need to know much about the Guile programming language to write most tests.

A script file will contain one or more testcases - each of which constitutes a test of a single well defined feature of the tool that the script is meant to test. A testcase is always written using the greg-testcase procedure, though this procedure could be invoked by a convenience macro.

greg-expect-pass

The greg-expect-pass macro is a shorthand method of writing the most usual sort of testcase - where a fragment of Guile code is run and is expected to return a true result. It passes an assertion and a fragment of Guile code that performs a test to the greg-testcase procedure -

; A simple test that basic arithmetic works

(greg-expect-pass "one plus one is two" (eq? (+ 1 1) 2))

is equivalent to -

(greg-testcase "one plus one is two" #t
  (lambda ()
    (eq? (+ 1 1) 2)
  )
)

greg-expect-fail

The greg-expect-fail macro is a shorthand method of writing a testcase to confirm that a known bug is still present in the code being tested. Once the bug is fixed, it would be altered to be a greg-expect-pass testcase.

; A test that basic arithmetic DOESN'T work!

(greg-expect-fail "one plus one is two" (eq? (+ 1 1) 2))

is equivalent to -

(greg-testcase "one plus one is two" #f
  (lambda ()
    (eq? (+ 1 1) 2)
  )
)

greg-testcase

The greg-testcase procedure takes three arguments -

The assertion
This is a string describing the test to be performed. It should take the form of an assertion that is expected to hold true.
The expected outcome
This is a boolean value - #t if the test is expected to return #t (ie. the assertion is expected to be proved correct), and #f if the test is expected to return #f (ie. the assertion is expected to be proved to be incorrect).
The test itself
This is a thunk (a Guile procedure that takes no arguments) and is normally a lambda expression. The return value of this procedure should be #t if the assertion is proved correct, #f if it is proved incorrect.

The Guile programming language permits the thunk to return in four ways -

Return the boolean value - #t
In which case, the test has succeeded, the assertion has proved correct, and Greg will record either an expected success (PASS) or an unexpected success (UPASS) depending on the expected outcome passed to greg-testcase.
Return the boolean value - #f
In which case, the test has failed, the assertion has proved incorrect, and Greg will record either an expected failure (XFAIL) or an unexpected failure (FAIL) depending on the expected outcome passed to greg-testcase.
Return a non-boolean value
In which case, the test is unresolved (UNRESOLVED).
Throw an exception
In which case Greg's behavior depends on the exception thrown -
- 'pass
  Greg acts as if the test had returned the boolean #t
- 'fail
  Greg acts as if the test had returned the boolean #f
- 'unsupported
  Greg reports this case as unsupported (UNSUPPORTED)
- 'untested
  Greg reports this case as untested (UNTESTED)
- (quit)
  If the script calls the (quit) primitive - an exception is raised. In this special case, the test is reported as unresolved (UNRESOLVED) and no further tests are executed. The testrun is terminated.
- any other exception
  Greg reports this case as unresolved (UNRESOLVED)

As there are no other ways in which the thunk may be exited, it is impossible for a testcase to produce a result that doesn't fit into the framework (unless your testcase manages either to crash Guile or enter an infinite loop - in which case you won't get any output).

The value returned by the greg-testcase procedure is a boolean - #t if the test resulted in an expected pass, #f otherwise.
You can use this return value to make the execution of subsequent testcases dependent on the success of an earlier testcase.


;
; A testcase to check an instance of numeric addition
;
(greg-testcase "One plus One is two" #t
(lambda ()
  (eq? (+ 1 1 ) 2)
))

;
;  The above testcase will generate output -
;  `PASS: One plus One is two'
;

The system provides hooks for general purpose procedures that are automatically called immediately before and after each testcase is executed. These procedures can be used to perform additional logging or other housekeeping functions. -

greg-case-begin The greg-case-begin variable is a hook for a procedure to be executed immediately before each testcase. It is initially defined to be (lambda () ()) which does nothing except return an unused empty list. You can replace this with a procedure of your choice, and use it to do any pre-test logging etc. When this procedure is called, greg-case-name will contain the name of the current testcase.
greg-case-end The greg-case-end variable is a hook for a procedure to be executed immediately before each testcase. It is initially defined to be (lambda (result) ()) which does nothing except return an unused empty list. You can replace this with a procedure of your choice. When this procedure is called, greg-case-name will contain the name of the current testcase.

While a testcase is executing (or in the greg-case-begin or greg-case-end procedures) there are a number of public procedures that may be used to obtain information about the system -

greg-casename This procedure returns the name of the current testcase. It is valid during execution of the testcase and during the pre and post test procedures.
greg-directory The greg-directory procedure takes no arguments and simply returns a string containing the absolute path name of the directory in which the current testscript is located. This value can be used where a testscript wishes to load data from local files in order to support a test. The string returned by this method includes a trailing slash character, so to refer to a file located in the same directory as the testscript simply by appending the fielname. eg (string-append (greg-directory) "foo")
greg-filename This procedure returns the name of the current test file. It is valid during execution of the testcase and during the pre and post test procedures.
greg-toolname This procedure returns the name of the current test tool. It is valid during execution of the testcase and during the pre and post test procedures.

Multiple testcases

It is normal to have more than one testcase in a file and this produces no problems - the only thing to watch out for is communicating information between testcases -

The scope of variables defined in the thunk in a greg-testcase procedure call is that thunk - the variable will not be visible to the next testcase.

So - to pass information from one testcase to the next it is necessary to define variables that can be seen in each testcase. The way to do this is normally to define these variables at the start of the file and then use the set! procedure within each testcase to set a value for a variable to be passed to the next testcase.


(define arith-ok #f)

;
; A testcase to check an instance of numeric addition
;
(greg-testcase "One plus One is two" #t
(lambda ()
  (if (eq? (+ 1 1 ) 2)
    (begin (set! arith-ok #t) #t)
    #f)
))

;
; A testcase to check arithmetic - only supported if we have addition.
;
(greg-testcase "X multiplied by 2 is X plus X" #t
(lambda ()
  (if arith-ok
    (eq? (+ 1 1) (* 1 2))
    (throw 'unsupported))
))

Of course, if (as above) the only information you want to pass from a testcase is whether the test succeeded or not, you can use the return value from the greg-testcase procedure directly -


(if
  (greg-testcase "One plus One is two" #t
    (lambda ()
      (eq? (+ 1 1 ) 2)
    )
  )
  (greg-testcase "X multiplied by 2 is X plus X" #t
    (lambda ()
      (eq? (+ 1 1) (* 1 2))
    )
  )
  (greg-dlog "Arithmetic operations not supported\n")
)

External tests

When Greg is used to test an external application, you usually want to run that application as a child process on a pseudo-terminal and handle tests sending a sequence of commands to the application and reading anticipated output from the application.

Starting a child process

Greg provides the greg-child procedure to start up a child process on a pseudo-terminal. You would usually call this procedure in the `begin.grg' file in your tool directory, but you could call it at the start of each script to get a new child process for each script.

The greg-child procedure expects one argument (the name of the program to be executed) followed by any number of additional arguments which are the arguments to be passed to the child process.
If the program name does not begin with a slash, Greg will look in the directory specified in greg-obj-dir to find it (by default the current directory).
If you want your normal PATH to be searched for the program, you should use -
(greg-child "/bin/sh" "-c" "foo args")
to get the shell to execute program `foo' with arguments args.

The greg-child procedure will automatically close down the I/O channels to any process previously started and wait for that process to die. If the old child process is badly behaved and will not die, this can result in Greg hanging - in this case you may need to explicitly kill the old child by another method before starting the new child process (this is one of the uses of the `end.grg' script).

As a special case, you can use an empty string as the program name - if you do this, another copy of the guile process will be created as a child and the value returned by greg-child in the child process will be a list containing the single number 0 (in the parent it will be a list containing the input port, output port and process id of the child). You can use this information to get the child copy of the process to be the program under test. This is useful for embedded testing where you want to test the I/O capabilities of the program.

NB. The greg-child procedure is implemented on top of the new primitive pty-child. This primitive is used to create a new child process on the end of a pseudo-terminal. Arguments and return values are as for greg-child.

When successfully starting a new program a three item list is returned in the parent - input port to read from the child, output port to write to the child, process id of child.
When returning in a child process because the program name was an empty string a list of one item (zero) is returned.
After a failure to create a child process, an empty list is returned.

Sending to a child process

The greg-send procedure is provided to send instructions to a child process. This procedure takes one or more string arguments and sends them to the child process (if one exists).

Reading from a child process

The greg-recv macro is used to read data from a child process. This procedure actually provides a simple front-end to the expect module. You can use the expect module facilities directly if you want more control than is offered by greg-recv.

The greg-recv macro expects one or more lists as arguments - each list containing a string (a pattern to match) and a result to return on a successful match. The value returned by greg-recv is the result for the pattern actually matched.
If no pattern is matched within the timeout period then an exception is raised, causing the testcase to return a FAIL result (unless you use (set! expect-timeout-proc xxx) to override Gregs timeout handler.
If no pattern is matched before an end-of-file is read, then an exception is raised, causing the testcase to return a FAIL result (unless you use (set! expect-eof-proc xxx) to override Gregs end-of-file handler.

NB. The expect-timeout-proc and expect-eof-proc are saved when a tool is tarted, and restored when it ends ... so if you want to make changes to these procedures in multiple tools, you must do so in the begin.grg and end.grg files for each tool.

In addition to setting up the above expect procedures, Greg also sets the expect-timeout variable to a 15 second timeout, and sets the expect-char-proc to be greg-dlog so that data read from the child process is logged as debug output by default. You can of course override this behavior in begin.grg.

The pattern matching is done with extended regular expressions, usually with input split into lines so that a carat (^) at the start of an expression matches the start of a line, and a dollar ($) matches the end of a line.
This is convenient for testing programs that produce lines of output in an expected format, as you can easily match the start and end of an output line.

If you want to change this behavior to permit multi-line patterns and to have the carat and dollar match the start of input and end of input respectively, then you can use -

(set! expect-strings-compile-flags regexp/extended)
(set! expect-strings-exec-flags 0)

This pattern matching behavior is occasionally useful when you are testing a program that produces output without clearly recognisable individual lines. NB. greg does not save and restore these values, so a change to them effect all tools being tested until you change them back.

A complete external test

;
; Run an interactive shell as a child process
;
(greg-child "/bin/sh" "-i")

;
; Set the shell prompt
;
(greg-send "PS1='% '\n")   

;
; Now test that the shell echoes what we expect.
; If we have a timeout or an eof, we will get a failure result.
;
(greg-testcase "echo 'hello'" #t
  (lambda ()
    (greg-send "echo hello\n")  ; Get it to send us something
    (expect-strings
      ("hello\r\n% " #t)
    )
  )
)