Add a test for a type error that this branch improves.

Also add a bunch of infrastructure -- the driver script for this is
massively overkill for just this test but we expect this test to grow
a lot of siblings in the not-too-distant future.

intTests/support/test-and-diff.sh

@@ -0,0 +1,223 @@
+# test-and-diff.sh: run SAW and diff the output against a reference
+# copy.
+#
+# usage: sh ../support/test-and-diff.sh [verb]
+# where verb is:
+#    test (default)	do run-tests, show-diffs, check-diffs
+#    run-tests		run all tests
+#    show-diffs		print all (nonempty) diff outputs
+#    check-diffs	exit 1 if there are any nonempty diff outputs
+#    good		update reference outputs from the current run
+#    clean		delete all results and intermediate output
+#
+# That is, the arguments are akin to make or similar tools.
+#
+# Note that this file is meant to be run/spawned/exec'd, not sourced.
+# It should also be run explicitly with the shell so that the test
+# suite is operable on Windows where #! doesn't work.
+#
+# This script will test all SAW files (*.saw) it finds in its current
+# directory, as follows, assuming T.saw is one such file:
+#    - run "saw T.saw"
+#    - produce T.rawlog
+#    - produce T.log with timestamps stripped out
+#    - diff T.log against T.log.good to produce T.diff
+#
+# These steps are "run-tests". It will always run all tests before
+# checking any of the outputs. If SAW itself fails (exits non-zero)
+# that fact is logged, and if unexpected will cause a diff, but will
+# not cause the test run to fail. This allows having tests that are
+# expected to fail in SAW, which is important for e.g. checking error
+# behavior.
+#
+# Note: don't write anything that relies on the existence of the
+# .rawlog file; one hopes the need for it will eventually go away.
+#
+# Also note: this script assumes that all *.rawlog, *.log, and *.diff
+# files it finds belong to it; they will be deleted at clean time.
+# This prevents old outputs to hang around forever when the list of
+# tests changes.
+#
+# It is fine to use this script when you have only one SAW script to
+# test. The infrastructure for supporting multiple scripts does not
+# cost much.
+#
+# There is, however, an expectation that if there are multiple scripts
+# in one test directory, they are all fast. Having multiple scripts at
+# once loses the ability to run them separately or address them
+# directly from enclosing test infrastructure. We don't want that to
+# incur significant costs or create aggravations.
+#
+# Note: we could, like make, not rerun tests whose output is newer
+# than (all) their inputs. So far I haven't done this. As long as the
+# scripts are all fast it probably doesn't matter.
+#
+# While I've written this script to be careful about spaces, spaces in
+# filenames aren't supported. Don't do that; it just creates headaches
+# for everyone.
+#
+
+# Get the list of tests.
+#
+# Note that in some shells (and depending on settings) asking for
+# *.saw when there aren't any will yield "*.saw" rather than
+# generating an error.
+TESTS=
+for SCRIPT in *.saw; do
+    BASE=${SCRIPT%.saw}
+    TESTS="$TESTS $BASE"
+done
+if [ "$TESTS" = " *.saw" ] || [ "$TESTS" = " " ]; then
+    echo "$0: Found no files matching *.saw" 1>&2
+    exit 1
+fi
+
+# Get the current directory, because we'll need to filter it out of
+# the saw output. Be sure to get the real current directory. Because
+# ksh's silly ideas about symlinks got stuffed into posix, pwd can lie
+# and then what it prints won't be the same as what saw outputs.
+#
+# Don't use the variable PWD for this; it has magic semantics in the
+# shell and, for the same reasons, may also lie.
+#
+# Bomb if the current directory includes a comma (just in case) so we
+# can use comma as a delimiter in the seddery below.
+CURDIR=$(pwd -P || pwd)
+if echo "$CURDIR" | grep -q , >/dev/null; then
+    echo "$0: Your current directory name contains a comma." 1>&2
+    echo "$0: I can't deal with that. Sorry..." 1>&2
+    exit 1
+fi
+
+# shell function for the run-tests op
+run-tests() {
+    for TEST in $TESTS; do
+        # Remove any existing test.log first as a precaution. This
+        # protects against misreading the results if the whole run
+        # gets killed before a new test.log gets produced.
+        rm -f $TEST.log
+
+        # run the test
+        # (do not fail if saw does, instead log it)
+        echo "$SAW $TEST.saw"
+        $SAW $TEST.saw > $TEST.rawlog 2>&1 || echo FAILED >> $TEST.rawlog
+
+        # Prune the timestamps from the log since they'll never match.
+        # We also need to shorten pathnames that contain the current
+        # directory, because saw feels the need to expand pathnames
+        # (see also saw-script #2082).
+        sed < $TEST.rawlog > $TEST.log '
+            /^\[[0-9][0-9]:[0-9][0-9]:[0-9][0-9]\.[0-9][0-9][0-9]\] /{
+                s/^..............//
+            }
+            s,'"$CURDIR"'/,,
+        '
+
+        # Check the output against the expected version.
+        # Note: because we (intentionally) aren't using set -e, we
+        # don't need to failure-protect this with || true.
+        # Send any errors from diff to the output so they get seen.
+        diff -u $TEST.log.good $TEST.log > $TEST.log.diff 2>&1
+    done
+}
+
+# shell function for the show-diffs op
+show-diffs() {
+    # The easy way to do this is just "cat *.diff". While (as
+    # documented) above we assume all *.diff files belong to us, don't
+    # do things that way, because there isn't an analogous form for
+    # check-diffs. Thus if a stray nonempty *.diff file appeared it
+    # would show up in the show-diffs output but not cause check-diffs
+    # to fail, and that would be quite confusing.
+    #
+    # As long as the number of tests here isn't really excessive the
+    # shell iteration and resulting multiple cat processes won't cost
+    # enough to care about.
+    for TEST in $TESTS; do
+        cat $TEST.log.diff
+    done
+}
+
+# shell function for the check-diffs op
+check-diffs() {
+    for TEST in $TESTS; do
+        if [ -s $TEST.log.diff ]; then
+            cat 1>&2 <<EOF
+
+Unexpected test diffs.
+If the new outputs are correct, update the reference outputs, but
+please don't do so without thinking.
+EOF
+            exit 1
+        fi
+    done
+}
+
+# shell function for the good op
+good() {
+    # This check might be annoying; sometimes you want to update one
+    # output at a time or similar. For now assume it's probably a good
+    # idea to have it though.
+    for TEST in $TESTS; do
+        if ! [ -f $TEST.log ]; then
+            echo "$0: No test output for $TEST.saw" 1>&2
+            echo "$0: Cannot update reference outputs" 1>&2
+            exit 1
+        fi
+    done
+
+    # now actually do it
+    if ! [ -f $TEST.log.good ] || \
+       ! diff -q $TEST.log.good $TEST.log >/dev/null; then
+	    echo "cp $TEST.log $TEST.log.good"
+	    cp $TEST.log $TEST.log.good
+     fi
+}
+
+# shell function for the clean op
+clean() {
+    echo "rm -f *.rawlog *.log *.diff"
+    rm -f *.rawlog *.log *.diff
+}
+
+# shell function for the test op
+test() {
+    run-tests
+    show-diffs
+    check-diffs
+}
+
+# run the requested operations
+if [ $# = 0 ]; then
+    test
+else
+    for VERB in "$@"; do
+        case "$VERB" in
+            test)
+                test
+            ;;
+            run-tests)
+                run-tests
+            ;;
+            show-diffs|show) # allow "show" as short form
+                show-diffs
+            ;;
+            check-diffs|check) # allow "check" as short form
+                check-diffs
+            ;;
+            good)
+                good
+            ;;
+            clean)
+                clean
+            ;;
+            *)
+                echo "$0: unknown action $VERB" 1>&2
+                exit 1
+            ;;
+        esac
+    done
+fi
+
+# done
+exit 0

intTests/test_type_errors/.gitignore

@@ -0,0 +1,3 @@
+*.rawlog
+*.log
+*.diff

intTests/test_type_errors/err001.log.good

@@ -0,0 +1,10 @@
+ Loading file "err001.saw"
+ err001.saw:1:23-1:24: Type mismatch.
+    Mismatch of type constructors. Expected: String but got Int
+
+    Expected: String
+    Found:    Int
+
+    within "foo" (err001.saw:1:5-1:8)
+
+FAILED

intTests/test_type_errors/err001.saw

@@ -0,0 +1 @@
+let foo () : String = 3;

intTests/test_type_errors/test.sh

@@ -0,0 +1 @@
+exec ${TEST_SHELL:-bash} ../support/test-and-diff.sh "$@"