Attempt to test for failure #1226

This commit attempts to modify sharness/test_build_static_site.t so as
to catch the error described in #1226. Unfortunately, the test does not
work as intended: the test passes even when the script is crashing.

The reason is that the exit trap in build_static_site.sh traps the
failure, and as such the script returns with exit code 0. Based on
googling, it should be possible to catch the actual exit code and
re-exit with it during the cleanup trap, and this indeed works for
"regular" exit codes, but not for the `1` that comes from referencing
an unset variable.

See the included script `problem.sh` for a reproduction.

problem.sh

@@ -0,0 +1,31 @@
+#!/bin/bash
+
+set -eu
+
+# Demonstrates the difficulty of reporting a proper exit code when referencing
+# an unset variable and using a cleanup trap.
+#
+# If you switch from failure_without_exit_code to failure_with_exit_code in the
+# main block, you'll find that the trap proxies the error code correctly for a
+# "normal" return code but not for one generated by an unset variable.
+
+main() {
+  trap cleanup EXIT
+  failure_without_exit_code
+}
+
+cleanup() {
+  rv=$?
+  echo "CLEANUP"
+  exit $rv
+}
+
+failure_without_exit_code() {
+    echo ${UNSET_VARIABLE}
+}
+
+failure_with_exit_code() {
+  rm nonexistent_file_1234
+}
+
+main

sharness/test_build_static_site.t

@@ -97,6 +97,11 @@ test_expect_success "should fail with multiple cname values" '
     printf "redacted\n" | test_cmp - important_dir/.wallet.dat
 '
 
+test_expect_success "should not error when SOURCECRED_DIRECTORY is unset" '
+    unset SOURCECRED_DIRECTORY &&
+    run --target build_output/output_sourcecred_directory_unset
+'
+
 #
 # Now, actually generate output in two cases: one with repositories, and
 # one with no repositories. We can only do this if we have a token.