Port test.sh to platform independent D.

This executes also way faster by not spawning GNU diff.

tests/test.d

@@ -0,0 +1,104 @@
+#!/usr/bin/env rdmd
+/**
+Platform independent port of `test.sh`. Runs the tests in this directory.
+
+Ignores differences in line endings, unless the test uses `--end_of_line`.
+**/
+import std.algorithm, std.array, std.conv, std.file, std.path, std.process;
+import std.stdio, std.string, std.typecons, std.range, std.uni;
+
+version (Windows)
+    enum dfmt = `..\bin\dfmt.exe`;
+else
+    enum dfmt = `../bin/dfmt`;
+
+int main()
+{
+    foreach (braceStyle; ["allman", "otbs", "knr"])
+    {
+        foreach (entry; dirEntries(".", "*.d", SpanMode.shallow).
+                            filter!(e => e.baseName(".d") != thisExePath.baseName(".exe")))
+        {
+            const source = entry.baseName;
+            const outFileName = buildPath(braceStyle, source ~ ".out");
+            const refFileName = buildPath(braceStyle, source ~ ".ref");
+            const argsFile = source.stripExtension ~ ".args";
+            const dfmtCommand = 
+                [dfmt, "--brace_style=" ~ braceStyle] ~ 
+                (argsFile.exists ? readText(argsFile).splitter!isWhite.filter!(a => a.length).array : []) ~
+                [source];
+            writeln(dfmtCommand.join(" "));
+            if (const result = spawnProcess(dfmtCommand, stdin, File(outFileName, "w")).wait)
+                return result;
+
+            // As long as dfmt defaults to LF line endings (issue #552), we'll have to default to ignore
+            // the line endings in our verification with the reference.
+            const keepTerminator = dfmtCommand.any!(a => a.canFind("--end_of_line")).to!(Flag!"keepTerminator");
+            const outText = outFileName.readText;
+            const refText = refFileName.readText;
+            const outLines = outText.splitLines(keepTerminator);
+            const refLines = refText.splitLines(keepTerminator);
+            foreach (i; 0 .. min(refLines.length, outLines.length))
+                if (outLines[i] != refLines[i])
+                {
+                    writeln("Found difference between ", outFileName, " and ", refFileName, " on line ", i + 1, ":");
+                    writefln("out: %(%s%)", [outLines[i]]); // Wrapping in array shows line endings.
+                    writefln("ref: %(%s%)", [refLines[i]]);
+                    return 1;
+                }
+            if (outLines.length < refLines.length)
+            {
+                writeln("Line ", outLines.length + 1, " in ", refFileName, " not found in ", outFileName, ":");
+                writefln("%(%s%)", [refLines[outLines.length]]);
+                return 1;
+            }
+            if (outLines.length > refLines.length)
+            {
+                writeln("Line ", outLines.length + 1, " in ", outFileName, " not present in ", refFileName, ":");
+                writefln("%(%s%)", [outLines[refLines.length]]);
+                return 1;
+            }
+
+            if (outText.endsWithNewline)
+            {
+                if (!refText.endsWithNewline)
+                {
+                    writeln(outFileName, " ends with a newline, but ", refFileName, " does not.");
+                    return 1;
+                }
+            }
+            else
+            {
+                if (refText.endsWithNewline)
+                {
+                    writeln(refFileName, " ends with a newline, but ", outFileName, " does not.");
+                    return 1;
+                }
+            }
+
+        }
+    }
+
+    foreach (entry; dirEntries("expected_failures", "*.d", SpanMode.shallow))
+        if (execute([dfmt, entry]).status == 0)
+        {
+            stderr.writeln("Expected failure on test ", entry, " but passed.");
+            return 1;
+        }
+
+    writeln("All tests succeeded.");
+    return 0;
+}
+
+bool endsWithNewline(string text) pure
+{
+    // Same criteria as https://dlang.org/phobos/std_string.html#.lineSplitter
+    return
+        text.endsWith('\n') ||
+        text.endsWith('\r') ||
+        text.endsWith(lineSep) ||
+        text.endsWith(paraSep) ||
+        text.endsWith('\u0085') ||
+        text.endsWith('\v') ||
+        text.endsWith('\f');
+}