move survey results into tests

.gitignore

@@ -0,0 +1,5 @@
+tests/StringSplitTest.class
+tests/java.csv
+tests/perl.csv
+tests/python.csv
+tests/js-strsplit.csv

Makefile

@@ -22,23 +22,43 @@ NPM		 = npm
 #
 # Files
 #
-JS_FILES	:= $(shell find lib tests survey -name '*.js')
+JS_FILES	:= $(shell find lib tests -name '*.js')
 JSL_CONF_NODE	 = tools/jsl.node.conf
 JSL_FILES_NODE   = $(JS_FILES)
 JSSTYLE_FILES	 = $(JS_FILES)
 
+TEST_FILES       = java.csv perl.csv python.csv js-strsplit.csv
+TEST_OUTPUTS     = $(TEST_FILES:%=tests/%)
+CLEAN_FILES	+= $(TEST_OUTPUTS) tests/StringSplitTest.class
+
 #
 # Repo-specific targets
 #
 .PHONY: all
-all:
+all: $(TEST_OUTPUTS)
 	$(NPM) install
 
-test:
+.PHONY: test
+test: $(TEST_OUTPUTS)
 	tests/tst.strsplit.sh
 	tests/tst.strpatterns.js
 	@echo All tests passed.
 
+tests/java.csv: tests/testcases.csv tests/StringSplitTest.class
+	java -cp tests StringSplitTest < $< > $@
+
+tests/StringSplitTest.class: tests/StringSplitTest.java
+	javac $^
+
+tests/js-strsplit.csv: tests/testcases.csv tests/strsplit.js
+	tests/strsplit.js < $< > $@
+
+tests/perl.csv: tests/testcases.csv tests/strsplit.pl
+	tests/strsplit.pl < $< > $@
+
+tests/python.csv: tests/testcases.csv tests/strsplit.py
+	tests/strsplit.py < $< > $@
+
 DISTCLEAN_FILES += node_modules
 
 include ./Makefile.targ

README.md

@@ -71,3 +71,43 @@ and here's strsplit:
     [ 'alpha', 'bravo', 'charlie delta' ]
 
 This is the behavior implemented by `split` in Perl, Java, and Python.
+
+## Background: survey of "split" in Java, Perl, and Python
+
+The tests directory contains test cases and test programs in Java, Perl, and
+Python for figuring out what these language's string split function does.
+Specifically, this is:
+
+* Java: String.split.
+* Perl: split.
+* Python: re.split.  While the "split" method on strings may be more common, it
+  does not handle regular expressions, while the Java and Perl counterparts do.
+
+For comparison, there's also a test case for this implementation of "strsplit".
+in JavaScript.
+
+The test cases here test both a simple string as a splitter (a space) and a
+simple regular expression (`\s+`, indicating some non-zero number of whitespace
+characters), as well as various values of the optional "limit" parameter.
+
+In summary, in all of the cases tried, the Java and Perl implementations are
+identical.  The Python implementation differs in a few ways:
+
+* The "limit" argument is off-by-one relative to the Java and Perl APIs.  It
+  represents the maximum number of splits to be made, rather than the maximum
+  number of returned fields.
+* -1 for "limit" is not special, and seems to mean that at most -1 splits will
+  be made, meaning the string is not split at all.  In Java and Perl, -1 means
+  there is no limit to the number of returned fields.
+* Java and Perl strip trailing empty fields when "limit" is 0.  Python never
+  strips trailing empty fields.
+
+JavaScript has a "split" method, but it behaves substantially different than all
+of these implementations when "limit" is specified.  This implementation of
+"strsplit" for JavaScript mirrors the Java and Perl implementations, as the
+differences in Python do not seem substantial or better.
+
+The remaining use case that would be nice to address is splitting fields the way
+awk(1) and bash(1) do, which is to strip leading whitespace.  Python's *string*
+split also does this, but only if you specify None as the pattern.  strsplit
+doesn't support this; just trim the string first if you want that behavior.

tests/tst.strsplit.sh

@@ -2,18 +2,18 @@
 
 #
 # The main test suite for strsplit is to run the body of test cases in
-# ../survey/ and compare the output to that of Java and Perl, whose
-# implementations we intend to mirror exactly.  errexit will cause this script
-# to exit with failure if any of these operations fail.
+# testcases.csv and compare the output to that of Java and Perl, whose
+# implementations we intend to mirror exactly.  All of these outputs have been
+# generated automatically by "make test".  errexit will cause this script to
+# exit with failure if any of these operations fail.
 #
 set -o errexit
 
-surveydir=$(dirname $0)/../survey
+cd $(dirname $0)
 
 set -o xtrace
-make -C $surveydir perl.csv java.csv js-strsplit.csv
-diff $surveydir/js-strsplit.csv $surveydir/perl.csv > /dev/null
-diff $surveydir/js-strsplit.csv $surveydir/java.csv > /dev/null
+diff js-strsplit.csv perl.csv > /dev/null
+diff js-strsplit.csv java.csv > /dev/null
 set +o xtrace
 
 echo "Test PASSED"