Build Your Own Tests

How to write your own shell scripts for testing


  • Requirements
  • Writing the test scripts
  • How to run the tests and interpret the results

What this is NOT: this is NOT an overview of the YAML testing framework. You can find a description of YAML testing on a separate page.


  • be robust - check that all prerequisites are met, and bail out if not
  • overall goal: all scripts should be portable
  • exit value according to AM standards
  • should not rely on anything outside the own language dir
  • should use variables for configured tools
  • should use both xfst and hfst, depending on what has been configured
  • test only modules that have been built


Check that all prerequisites are met, and bail out if not (exit 77/SKIP)

  • are fst's found?
  • do we find the input data files
  • do we have all tools needed?


Portability means it should:

  • work on all systems (except Windows)
  • work both when you have checked out all of $GTHOME and when you have checked out only $GTCORE and one language
  • work when the language dir (when checking out single languages) is called something else than default
  • work for different flavours of the same tool (e.g. for both awk and gawk)

Exit values

Must be 0 - 255, where some have a special meaning:

everything went ok = PASS
some precondition was not met, we need to SKIP the test
hard error - we can't continue - STOP
everything else
FAIL (usually just 1)

Do not rely on anything outside the own language dir

  • all paths should be relative to the local dir
  • do not reference $GTHOME and similar variables
  • the only variables you can trust are:
    • $srcdir - the directory in which the original test script is located
    • the variables defined and exported by - but ONLY if you process the testing script with (details about this later)

should use variables for configured tools

Most of the tools we need (and in principle all of them) are (should be) declared in or related files (found in $GTLANG/m4/.

Autoconf (the tool that parses has its own machinery to find variants of different tools, to define variables for the identified tools, etc.

By using the variables defined in you can be sure that the tools actually are available, and you can easily add tests for those variables in case the tool is not found.

By following this practice, the system becomes more robust and portable.

Details of how to actually do this is given further down.

It should use both xfst and hfst

... depending on what has been configured.

The new infrastructure treats the Xerox and the Hfst tools on an equal footing, meaning that some have the Xerox tools installed, some have Hfst, and some have both.

The test scripts should check for what has actually been built, and what is available on the system, and use the one or both that is available or configured.

We'll return to the details further down.

Test only modules that have been built


  1. test only spellers if speller building have been turned on

How do we do this?

  1. By using Automake conditionals:

# Add your shell scripts for running tests requiring only a generator: \ \ \


There is a list of all presently defined conditionals here.

Writing the test scripts

  • what to test
  • define variables
  • read in test data if needed
  • test that all tools and data are found
  • make a loop for xfst and hfst if relevant
  • write the real test
  • add the test script to

We will use the test script $GTLANGS/sma/test/src/morphology/ as an example throughout this section.

What to test

You can test anything that is scriptable or programable. The only requirement is that the answer can be captured as a YES or NO, i.e. PASS or FAIL.

Here are some ideas:

  • test that a syntactic analysis is what you expect (compare with expected, FAIL if there is a diff)
  • test whether a given non-word gets a specific suggestion in a speller
  • test that the hyphenation patterns are correct (this could probably be easily done using the YAML testing framework)

What programming languages can I use?

Anything that can return an exit value. Common choices are:

  • shell scripts
  • perl scripts
  • python scripts

... but also C/C++ a.o. are used.

Define variables

Typically you start a shell script by defining variables:

###### Variables: #######

The variable ${srcdir} refers to the source dir of the test script, that is, the directory in which the test script is located.

Here is another variable assignment:

# Get external Mac editor for viewing failed results from configure:

Variables from

If your testing script relies on a lot of external tools, it is a good idea to make sure that the tools are actually installed on the system. This is the job of the file. To make use of this feature, there are a couple of things to remember:

  • the test script filename should end in
  • the testing script must be processed by — this is done by adding two lines as follows to that file:
AC_CONFIG_FILES([test/src/morphology/], \
      [chmod a+x test/src/morphology/])

The first line tells autoconf to process the * file, and produce the actual *.sh file, the second line ensures that the final shell file is executable.

In this processing all variables will be replaced with their actual value as identifed during the configuration phase. Such variables look like @VARIABLE@ in the test script.

Variables from - an example

  • we need to use Xerox' lookup tool as part of the test
  • we use to check for the availability of lookup
  • typically, that will set a corresponding variable LOOKUP in configure
  • you reference this variable in your * file, and when configured, the variable is replaced with the actual value
  • the variable looks like this in the * file: @LOOKUP@

That is, in a hypothetic test file we could write something like:


The corresponding test file will after configuration look something like:


Then we can add tests in the testing script to check whether $LOOKUP is empty, and if it is, the test script can bail out with a SKIP return value ( 77).

NB! Sometimes the variable is not empty when the tool is not found, but could contain strings like false or no instead. Check the actual value if the test for the tool doesn't fall out as expected.

test that all tools and data are found

We need to test that the data sources used in the test are actually found:

# Check that the source file exists:
if [ ! -f "$sourcefile" ]; then
	echo Source file not found: $sourcefile
	exit 1

Here we use the variable we defined, and if it does not exist, we exit with an error.

make a loop for xfst and hfst if relevant

When doing morphological tests, we want to test both xfst and hfst. First we define a variable fsttype:

# Use autotools mechanisms to only run the configured fst types in the tests:
@CAN_HFST_TRUE@fsttype="$fsttype hfst"
@CAN_XFST_TRUE@fsttype="$fsttype xfst"

The strings @CAN_HFST_TRUE@ and @CAN_XFST_TRUE@ come from autoconf, and will tell us what they say.

The we check that the variable is not empty:

# Exit if both hfst and xerox have been shut off:
if test -z "$fsttype"; then
    echo "All transducer types have been shut off at configure time."
    echo "Nothing to test. Skipping."
    exit 77

Finally, the actual loop looks like:

for f in $fsttype; do

read in test data if needed

###### Extraction: #######
# extract non-compounding lemmas:
grep ";" $sourcefile | grep -v "^\!" \
	| egrep -v '(CmpN/Only|\+Gen\+|\+Der\+| R )' | sed 's/% /€/g' \
	| sed 's/%:/¢/g' | tr ":+" " " \
	| cut -d " " -f1 | tr -d "#" | tr "€" " " | tr "¢" ":" \
	| sort -u | grep -v '^$' > nouns.txt

# extract compounding lemmas:
grep ";" $sourcefile | grep -v "^\!" \
	| grep ' R '| tr ":+" " " | cut -d " " -f1 | tr -d "#" \
	| sort -u > Rnouns.txt

Write the real test

This is an excerpt from the sma test file mentioned earlier, and should only serve as an example:

###### Test non-comopunds: #######
		# generate nouns in Singular, extract the resulting generated lemma,
		# store it:
		sed 's/$/+N+Sg+Nom/' nouns.txt | $lookuptool $generatorfile.$f \
			| cut -f2 | fgrep -v "+N+Sg" | grep -v "^$" | sort -u \
			> analnouns.$f.txt 
		# Generate nouns, extract those that do not generate in singular,
		# generate the rest in plural:
		sed 's/$/+N+Sg+Nom/' nouns.txt | $lookuptool $generatorfile.$f \
			| cut -f2 | grep "N+" | cut -d "+" -f1 | sed 's/$/+N+Pl+Nom/' \
			| $lookuptool $generatorfile.$f | cut -f2 \
			| grep -v "^$" >> analnouns.$f.txt 

The full test script file can be found here.

Add the test script to

# List here (space separated) all test scripts that should be run
# unconditionally:

# Add your shell scripts for running tests requiring only a generator: \ \ \


# List tests that are presently (expected) failures here, ie things that should
# be fixed *later*, but is not critical at the moment: \

Add the test script to

If we have written an *.in file - as in this example - we need to process it with configure to replace @VARIABLE@ style variables with their configure values. To do that, you need to add two lines like the following to

AC_CONFIG_FILES([test/src/morphology/], \
      [chmod a+x test/src/morphology/])

With these two lines, configure will be able to produce the shell script that we added to above.

How to run the tests and interpret the results

  • basic commands
  • what happens upon FAILs
  • what outcomes can there be?

Basic commands

  • make check - runs all defined tests
  • make check - will run only the test script

To run a subset of tests, cd into the subdir containing the subset of tests you want to run, and do make check there. Only the tests in that directory and its subdirectories will be run.

Single tests and out-of-source building

( NB! Advanced topic - skip if not relevant)

When using out-of-source builds (aka VPATH builds), running single tests like above will not work, due to the way Automake treats the TESTS variable when there are subdirs with their own tests. To make it work, you need to restrict make to only run in the local directory where you have the test script you want to run:

cd to/dir/with/test/script/in/build/tree/
make check SUBDIRS=.

Setting the SUBDIRS variable to just a period (meaning "this directory") forces make to ignore the subdirs, and the single test works as intended.

NOTE: this is only relevant if you have out-of-source builds, and want to run a single test script. If you want to run all test scripts in your working directory and below (i.e. make check), there is no need to do anything extra - everything works as expected.

What happens when something fails

The tests are run on a per directory basis, which means that all tests in a directory will be run, and then make will give a report.

If some of the tests FAILed, then that is an error in the view of make, and make stops. This is a property of make and the Automake system. You can override this behavior with option -i, --ignore-errors. The problem with using -i is of course that you risk ignoring errors, since the error message can easily scroll out of view before make is done.

What outcomes can there be?

Testing within the Automake framework can have five outcomes:

everything is ok
some condition in the test was NOT met
some condition in the test was NOT met, but we are aware of the issue, and will handle it later => testing will CONTINUE despite the FAIL
everything is ok but we didn't know - we expected a FAIL, but got a PASS (an uneXpected PASS) => testing will STOP baecause of this, to ensure that the developer notices the new state of affairs
some precondition was not met, and the test was not performed.