Introduced hash valued location markers in the spec

tools/addlatexhash.dart

+// Copyright (c) 2014, the Dart project authors.  Please see the AUTHORS file
+// for details. All rights reserved. Use of this source code is governed by a
+// BSD-style license that can be found in the LICENSE file.
+//
+// ----------------------------------------------------------------------
+// This is a very specialized tool which was created in order to support
+// adding hash values used as location markers in the LaTeX source of the
+// language specification.  It is intended to take its input file as the
+// first argument and the output file name as the second argument. From
+// docs/language a typical usage would be as follows:
+//
+//   dart ../../tools/addlatexhash.dart dartLangSpec.tex tmp.tex
+//
+// This will yield a normalized variant tmp.tex of the language
+// specification with hash values filled in.  For more details, please
+// check the language specification source itself.
+//
+// NB: This utility assumes UN*X style line endings, \n, in the LaTeX
+// source file receieved as input; it will not work with other styles.
+//
+// TODO: The current version does not fill in hash values, it only
+// standardizes the LaTeX source by removing comments and normalizing
+// white space.
+
+import 'dart:io';
+import 'dart:convert';
+import '../pkg/crypto/lib/crypto.dart';
+
+// ----------------------------------------------------------------------
+// Normalization of the text, i.e., removal or normalization
+// of elements that do not affect the output from latex
+
+final commentAllRe = new RegExp("^%");

[Lasse Reichstein Nielsen, 2014/10/15 09:13:17]

Using a RegExp for this is overkill, just do string.startsWith("%").

Two-letter acronyms can be fully upper-case in names, so maybe "commentAllRE"
(but that's a matter of taste).

[eernst, 2014/10/15 13:19:27]

Looks more meaningful with "RE".  Done.

+final commentRe = new RegExp("[^\\\\]%[^\\n]*");

[Lasse Reichstein Nielsen, 2014/10/15 09:13:17]

I recommend using raw strings for RegExp sources:

final commentRE = new RegExp(r"[^\\]%[^\n]"); 

Much more readable! :)

[eernst, 2014/10/15 13:19:27]

Indeed; added the final "*", done.

[Lasse Reichstein Nielsen, 2014/10/15 14:09:57]

Also just noticed that as a regexp, "[^n]" is equivalent to ".", so it can get
even smaller: r"[^\\]%.*"
Yey!

[eernst, 2014/10/15 14:26:41]

Cool!  Done + Added a comment, just in case someone else will ever forget that .
does not match \n.  ;)

+final whitespaceAllRe = new RegExp("^\\s+\$");
+final whitespaceLeadingRe = new RegExp("^\\s+[^\\n]");
+final whitespaceRe = new RegExp("[ \\t][ \\t]+");

[Lasse Reichstein Nielsen, 2014/10/15 09:13:17]

Shorter regexp possible:
  final whitespaceRE = new RegExp(r"[ \t]{2,}");

[eernst, 2014/10/15 13:19:27]

Done.

+
+// normalization steps
+
+cutMatch(line, match, {startOffset: 0, endOffset: 0, glue: ""}) {
+  if (match == null) return line;
+  var start = match.start + startOffset;
+  var end = match.end + endOffset;
+  var len = line.length;
+  if (start < 0) start = 0;
+  if (end > len) end = len;
+  return line.substring(0, start) + glue + line.substring(end);
+}
+
+cutRegexp(line, re, {startOffset: 0, endOffset: 0, glue: ""}) {
+  return cutMatch(line, re.firstMatch(line),
+                  startOffset: startOffset,
+                  endOffset: endOffset,
+                  glue: glue);
+}
+
+cutFromMatch(line, match, {offset: 0, glue: ""}) {
+  if (match == null) return line;
+  return line.substring(0, match.start + offset) + glue;
+}
+
+cutFromRegexp(line, re, {offset: 0, glue: ""}) {
+  return cutFromMatch(line, re.firstMatch(line), offset: offset, glue: glue);
+}
+
+isWsOnly(line) => whitespaceAllRe.firstMatch(line) != null;
+isCommentOnly(line) => commentAllRe.firstMatch(line) != null;

[Lasse Reichstein Nielsen, 2014/10/15 09:13:17]

This would be the non-regexp version: => line.startsWith("%");

Avoiding unnecessary RegExps is a pet peeve.

[eernst, 2014/10/15 13:19:27]

Arg null is treated also as an error by firstMatch, so: done.

+
+justEol(line) {

[Lasse Reichstein Nielsen, 2014/10/15 09:13:17]

I'd prefer return types, and parameter types, in general.
So:
  String justEol(String line) {

[eernst, 2014/10/15 13:19:27]

I decided to use a typeless style in this particular case because I spent so
much time working on types, so that's the "missing case".  But I get the
impression that there is a "type everything" or at least "type the interface
parts" rule in this context?  Haven't made this change yet.

+  if (line.length == 0) return line;
+  return line[line.length-1] == "\n" ? "\n" : "";

[Lasse Reichstein Nielsen, 2014/10/15 09:13:17]

return line.endsWith("\n") ? "\n" : "";

[eernst, 2014/10/15 13:19:27]

Done.

+}
+
+stripComment(line) {
+  // NB: it is tempting to remove everything from the '%' and out,
+  // including the final newline, if any, but this does not work.
+  // The problem is that TeX will do exactly this, but then it will
+  // add back a character that depends on its state (S, M, or N),
+  // and it is tricky to maintain a similar state that matches the
+  // state of TeX faithfully.  Hence, we remove the content of
+  // comments but do not remove the comments themselves, we just
+  // leave the '%' at the end of the line and let TeX manage its
+  // states in a way that does not differ from the file from before
+  // stripComment
+  if (isCommentOnly(line)) return "%\n";
+  return cutRegexp(line, commentRe, startOffset: 2);

[Lasse Reichstein Nielsen, 2014/10/15 09:13:17]

Doesn't this loose the trailing '\n'?

[eernst, 2014/10/15 13:19:27]

No, commentRE avoids matching a trailing '\n', so an '\n' at the end of the line
will be preserved, and absence of '\n' at the end will also be preserved.

+}
+
+// Reduce a wsOnly line to its eol, remove leading ws
+// entirely, and reduce multiple ws chars to one
+normalizeWhitespace(line) {

[Lasse Reichstein Nielsen, 2014/10/15 09:13:17]

To remove leading WS (including empty lines), try:
  var trimLine = line.trimLeft();
  if (trimLine.isEmpty) return justEol(line);
  line = trimLine;

[eernst, 2014/10/15 13:19:27]

Done.

+  if (isWsOnly(line)) return justEol(line);
+  line = cutRegexp(line, whitespaceLeadingRe, endOffset: -1);
+  var match;
+  while ((match = whitespaceRe.firstMatch(line)) != null) {
+    line = cutMatch(line, match, glue: " ");
+  }

[Lasse Reichstein Nielsen, 2014/10/15 09:13:17]

Maybe just:
  line = line.replaceAll(whitespaceRe, " ");

[eernst, 2014/10/15 13:19:27]

Entire method much nicer now!  Done.

+  return line;
+}
+
+// Reduce sequences of >1 wsOnly lines to 1, and sequences of >1 
+// commentOnly lines to 1; moreover, treat commentOnly lines as
+// wsOnly when occurring in wsOnly line blocks
+multilineNormalize(lines) {
+  var afterBlankLines = false; // does 'line' succeed >0 empty lines?
+  var afterCommentLines = false; // .. succeed >0 commentOnly lines?
+  var newLines = new List();
+  for (var line in lines) {
+    if (afterBlankLines && afterCommentLines) {
+      // can never happen
+      throw "Bug, please report to eernst@";
+    } else if (afterBlankLines && !afterCommentLines) {
+      // at least one line before 'line' is wsOnly
+      if (!isWsOnly(line)) {
+        // blank line block ended
+        afterCommentLines = isCommentOnly(line);
+        // special case: it seems to be safe to remove commentOnly lines
+        // after wsOnly lines, so the TeX state must be predictably right;
+        // next line will then be afterCommentLines and be dropped, so
+        // we drop the entire comment block---which is very useful; we can
+        // also consider this comment line to be an empty line, such that
+        // subsequent empty lines can be considered to be in a block of
+        // empty lines; note that almost all variants of this will break..
+        if (afterCommentLines) {
+          // _current_ 'line' a commentOnly here
+          afterBlankLines = true;
+          afterCommentLines = false;
+          // and do not add 'line'
+        } else {
+          // after blanks, but current 'line' is neither blank nor comment
+          afterBlankLines = false;
+          newLines.add(line);
+        }
+      } else {
+        // blank line block continues, do not add 'line'
+      }
+    } else if (!afterBlankLines && afterCommentLines) {
+      // at least one line before 'line' is commentOnly
+      if (!isCommentOnly(line)) {
+        // comment line block ended
+        afterBlankLines = isWsOnly(line);
+        afterCommentLines = false;
+        newLines.add(line);
+      } else {
+        // comment line block continues, do not add 'line'
+      }
+    } else {
+      assert(!afterBlankLines && !afterCommentLines);
+      // no wsOnly or commentOnly lines preceed 'line'
+      if (isWsOnly(line)) afterBlankLines = true;
+      if (isCommentOnly(line)) afterCommentLines = true;

[Lasse Reichstein Nielsen, 2014/10/15 09:13:17]

Maybe:

if (isCommentOnly(line)) {
  afterCOmmentLines = true;
} else {
  if (isWsOnly(line)) afterblankLines = true;
  newLines.add(line);
}

(I'm just a simple programmer, seeing assignments to a variable followed by a
condition on that variable confuses and frightens me).

[eernst, 2014/10/15 13:19:27]

Actually that was because of the following pattern, which is used in all the
cases here:

  detect state (conditions of if-tree)
  // in each case
  update state variables
  act according to situation (which case, possibly new state)

Changed the code to make this pattern a bit more obvious.

+      if (!afterCommentLines) newLines.add(line);
+      // else skipping commentOnly line after nonWs, nonComment text
+    }
+  }
+  return newLines;
+}
+
+// select the elements in the pipeline
+
+normalize(line) => normalizeWhitespace(stripComment(line));
+
+sispNormalize(line) => stripComment(line);
+
+// ----------------------------------------------------------------------
+// Managing fragments with significant spacing
+
+final dartCodeBeginRe = new RegExp("^\\s*\\\\begin{dartCode}");

[Lasse Reichstein Nielsen, 2014/10/15 09:13:17]

Need to escape '{' and '}' in RegExp:

  new RegExp(r"^\s*\\begin\{dartCode\}")

and ditto below.

[eernst, 2014/10/15 13:19:27]

OK.  Note that we get no exceptions for this one, though:

  var re = new RegExp("some{other}thing");
  if (re.firstMatch(" some{other}thing ") == null) {
    throw "Braces must be escaped in REs to mean braces!";
  }

so the braces stand for themselves without escaping in that case, presumably
because the repeat count syntax failed to parse.  Changed it anyway, because
that's probably a temporary quirk of RegExp.

Create an issue on that?  If so, where?

[Lasse Reichstein Nielsen, 2014/10/15 14:09:57]

RegExp in browsers have traditionally been very forgiving (as in: If Netscape
allowed something, we better keep allowing it, whether it's in the spec or not).
Most RegExp syntax errors are silently error-corrected by treating the invalid
character as a literal. There's a good chance that PCRE (currently used by the
VM) has done the same.

I don't think we'll get dart2js to not use V8's RegExp, so it'll probably keep
working there. I'd file a bug for the VM, and try to keep it from getting rolled
into the "VM RegExps are just wrong" bug somehow. Maybe wait until the new
RegExp implementation lands - it will likely use the same irregexp parser used
by V8, so it will have the same forgivances. I'd love to weed those out once and
for all. No need to be backwards compatible with non-standard JS code in new
language!

[eernst, 2014/10/15 14:26:41]

What's the smart way to install a wakeup call to act when that new RegExp
implementation arrives?  We cannot create an issue now, supposedly.

+final dartCodeEndRe = new RegExp ("^\\s*\\\\end{dartCode}");
+
+sispIs(line, targetRe) {
+  return targetRe.firstMatch(line) != null;
+}
+
+sispIsDartBegin(line) => sispIs(line, dartCodeBeginRe);
+sispIsDartEnd(line) => sispIs(line, dartCodeEndRe);
+
+// ----------------------------------------------------------------------
+// main
+
+main ([args]) {
+  if (args.length != 2) {
+    print("Usage: addlatexhash.dart <input-file> <output-file>");
+    throw "Received ${args.length} arguments, expected two";
+  }
+
+  var inputFile = new File(args[0]);
+  var outputFile = new File(args[1]);
+  assert(inputFile.existsSync());
+
+  var lines = inputFile.readAsLinesSync();
+  // single-line normalization
+  var inDartCode = false;
+  var newLines = new List();
+
+  for (var line in lines) {
+    if (sispIsDartBegin(line)) {
+      inDartCode = true;
+    } else if (sispIsDartEnd(line)) {
+      inDartCode = false;
+    }
+    if (inDartCode) {
+      newLines.add(sispNormalize(line + "\n"));
+    } else {
+      newLines.add(normalize(line + "\n"));
+    }
+  }
+
+  // multi-line normalization
+  newLines = multilineNormalize(newLines);
+
+  // output result
+  outputFile.writeAsStringSync(newLines.join());
+}