Bug 1597903 - Improve error reporting on diffoscope tasks. r=froydnj

Somehow parse the diff output to print out slightly more unique error messages.

With this change, the error for bug 1597901 becomes:
TEST-UNEXPECTED-FAIL | firefox/libmozavutil.so differs. See the diff.html or diff.txt artifact

And the error for bug 1601150 becomes:
TEST-UNEXPECTED-FAIL | firefox/libxul.so differs. See the diff.html or diff.txt artifact

Which is not great, but better than the status quo.

Differential Revision: https://phabricator.services.mozilla.com/D55770

--HG--
extra : moz-landing-system : lando

taskcluster/docker/diffoscope/get_and_diffoscope

@@ -51,8 +51,6 @@ esac
 POST=true
 
 fail() {
-	set +x
-	echo "TEST-UNEXPECTED-FAIL | Builds differ. See the diff.html or diff.txt artifact"
 	exit 1
 }
 
@@ -79,9 +77,51 @@ if [ -n "$PRE_DIFF" ]; then
 	eval $PRE_DIFF
 fi
 
-diffoscope \
+if diffoscope \
 	--html diff.html \
 	--text diff.txt \
 	--progress \
 	$DIFFOSCOPE_ARGS \
-	a b || $POST
+	a b
+then
+	# Ok
+	:
+else
+	# The builds differ, let's try to output a useful error.
+
+	# We "parse" the diff output, so we look at the lines that contain a "tee", like:
+	# ├── firefox
+	# │ ├── libxul.so
+	# │ │ ├── readelf --wide --notes {}
+	# We ignore lines like the last one, to only report file names. And we ignore
+	# lines for directories such as the first one, but still look at them to report
+	# full paths.
+	python3 <<-EOF
+		TEE = '├──'
+		paths = set()
+		path = []
+		with open("diff.txt") as fh:
+		    for l in fh:
+		        if TEE not in l:
+		            continue
+		        fields = l.split()
+		        # We rely on the number of │ to figure out at what level the file
+		        # name applies.
+		        if fields[-2:-1] == [TEE]:
+		            path[len(fields) - 2:] = [fields[-1]]
+		        else:
+		            # Align path length to match the number of │
+		            path.append(None)
+		        path_ = [p for p in path if p]
+		        full_path = '/'.join(path_)
+		        parent_path = '/'.join(path_[:1])
+		        if parent_path in paths:
+		            paths.remove(parent_path)
+		        if full_path:
+		            paths.add(full_path)
+
+		for p in sorted(paths):
+		    print('TEST-UNEXPECTED-FAIL | {} differs. See the diff.html or diff.txt artifact'.format(p))
+	EOF
+	$POST
+fi