Testing APE

Author: Kaarel Kaljurand

Version: 2019-08-09

Introduction

This document explains how to test APE including the various DRS translators that APE contains.

All the test scripts in this directory depend on a larger lexicon than included with APE. In order to download the required lexicon, first run

swipl -f none -g ensure_clex -t halt -s downloader.pl

APE Regression test

To regression test APE, just run:

bash rtest.bash

or if you want to download the latest regression testset first then run:

swipl -f none -g download_acetexts -t halt -s downloader.pl

A file with a timestamp containing the regression test results along with APE's output is stored into the testruns-directory.

In order to get a digest of the testrun, grep the file for regression tester messages, e.g.:

cat testruns/rtest_050601-1310.txt | grep "^0"

or cat an already filtered file:

cat tmp/now.txt

In order to explore the erronous DRSes, open the file and search for the strings

• 0000 (Correctly generated an empty DRS)
• ---- (Correctly generated a DRS which is not empty)
• `0->#` (Failed to generate an empty DRS)
• `#->0` (Incorrectly generated an empty DRS)
• #### (DRSs which are not empty do not match.)

To get a listing of all the regressions:

cat tmp/now.txt | grep "\[.*#.*\]"

Note: the regression testing is only tested with SWI-Prolog. It is assumed that SWI-Prolog is called swipl, if it is not the case then modify rtest.bash accordingly or set a symbolic link, e.g.:

ln -s `which pl` swipl

Runtime of the complete test run on different machines:

• Mac OS X G4: ~25 seconds (~3000 testcases)
• Intel i3, 2011-06-20: ~11 seconds (3613 test cases, excluding bug reports)
• Intel i7, 2016-04-02: ~9 seconds (3733 test cases, excluding bug reports, including loading the large lexicon from a flat file)

Testing Drace Core and Drace NP

Running the test:

echo "[test_drace]. test_drace(core)." | swipl > testruns/drace_test_results.txt
echo "[test_drace]. test_drace(np)." | swipl > testruns/dracenp_test_results.txt

Getting an overview of the problems by filtering out certain testcases that point to pseudo-problems.

cat testruns/drace_test_results.txt | grep "FAIL" | grep -v ":" | grep -v "ach of" | wc

Running all the tests

time bash test_everything.sh > out.txt 2> err.txt

This takes about

• 55 seconds on an i3 Linux laptop with SWI-Prolog v6.
• 45 seconds on an i7 Linux laptop with SWI-Prolog v7.