GiellaLT

GiellaLT provides an infrastructure for rule-based language technology aimed at minority and indigenous languages, and streamlines building anything from keyboards to speech technology. Read more about Why. See also How to get started and our Privacy document.

View GiellaLT on GitHub

How to write your own shell scripts for testing

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.

Requirements

Robustness

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

Portability

Portability means it should:

Exit values

Must be 0 - 127, where some have a special meaning:

Do not rely on anything outside the own language dir

should use variables for configured tools

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

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

By using the variables defined in configure.ac 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

Example:

TESTS=

# Add your shell scripts for running tests requiring only a generator:
if WANT_GENERATION
TESTS+=test-noun-generation.sh \
	   test-verb-generation.sh \
	   test-adj-generation.sh \
	   test-propernoun-generation.sh
endif # WANT_GENERATION

There is a list of all presently defined conditionals here.

Writing the test scripts

We will use the test script $GTLANGS/sma/test/src/morphology/test-noun-generation.sh.in 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:

What programming languages can I use?

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

… but also C/C++ a.o. are used.

Define variables

Typically you start a shell script by defining variables:

###### Variables: #######
sourcefile=${srcdir}/../../../src/morphology/stems/nouns.lexc
generatorfile=./../../../src/generator-gt-norm
resultfile=missingNounLemmas

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:
EXTEDITOR=@SEE@

Variables from configure.ac

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 configure.ac file. To make use of this feature, there are a couple of things to remember:

AC_CONFIG_FILES([test/src/morphology/test-noun-generation.sh], \
      [chmod a+x test/src/morphology/test-noun-generation.sh])

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

In this processing all configure.ac 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 configure.ac - an example

That is, in a hypothetic test file test-lemmas.sh.in we could write something like:

LOOKUP=@LOOKUP@

The corresponding test file test-lemmas.sh will after configuration look something like:

LOOKUP=/usr/local/bin/lookup

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 2
fi

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 hfst, xfst and foma, as long as they are used. First we define a variable fsttype:

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

The strings @CAN_HFST_TRUE@, @CAN_XFST_TRUE@ and @CAN_FOMA_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
fi

Finally, the actual loop looks like:

for f in $fsttype; do
    # Add your test commands here...
done

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-compunds: #######
# 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 # remove backlash!

# 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 Makefile.am

# List here (space separated) all test scripts that should be run
# unconditionally:
TESTS=
if WANT_GENERATION
# Add your shell scripts for running tests requiring only a generator:
TESTS+=test-noun-generation.sh \
	   test-verb-generation.sh \
	   test-adj-generation.sh \
	   test-propernoun-generation.sh
endif # WANT_GENERATION
# List tests that are presently (expected) failures here, ie things that should
# be fixed *later*, but is not critical at the moment:
XFAIL_TESTS=generate-noun-lemmas.sh \
            test-propernoun-generation.sh

Add the test script to configure.ac

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 configure.ac:

AC_CONFIG_FILES([test/src/morphology/test-noun-generation.sh], \
      [chmod a+x test/src/morphology/test-noun-generation.sh])

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

How to run the tests and interpret the results

Basic commands

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 TESTS=a-test-script.sh 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 short report (in red if your console supports colours). The verbose results are saved in a file called test-suite.log in that directory and the full path is printed in the short report:

FAIL: accept-all-lemmas.sh
============================================================================
Testsuite summary for Giella smj 0.2.0
============================================================================
# TOTAL: 2
# PASS:  1
# SKIP:  0
# XFAIL: 0
# FAIL:  1
# XPASS: 0
# ERROR: 0
============================================================================
See tools/spellcheckers/test/fstbased/desktop/hfst/test-suite.log
Please report to feedback@divvun.no
============================================================================

in this case you want to open tools/spellcheckers/test/fstbased/desktop/hfst/test-suite.log with less -R or your favourite editor. To get a summary of all the most recent test results it is possible to use make devtest or find . -name test-suite.log -exec cat \{\} \;.

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: