Use HTTP response headers for PNaCl caching instead of bitcode hash.

ppapi/native_client/src/trusted/plugin/nacl_http_response_headers.cc

Index: ppapi/native_client/src/trusted/plugin/nacl_http_response_headers.cc
diff --git a/ppapi/native_client/src/trusted/plugin/nacl_http_response_headers.cc b/ppapi/native_client/src/trusted/plugin/nacl_http_response_headers.cc
new file mode 100644
index 0000000000000000000000000000000000000000..d10347138bc2cfd51bf5729c74d70ead45b99456
--- /dev/null
+++ b/ppapi/native_client/src/trusted/plugin/nacl_http_response_headers.cc
@@ -0,0 +1,105 @@
+// Copyright (c) 2013 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#include "native_client/src/trusted/plugin/nacl_http_response_headers.h"
+
+#include <algorithm>
+#include <sstream>
+
+namespace {
+
+void SplitString(const std::string& str,
+                 char delim,
+                 std::vector<std::string>* elems,
+                 int limit) {

[dmichael (off chromium), 2013/06/14 16:21:43]

Might be worth a TODO to use Tokenize from base/string_util.h when this moves.

(Though that doesn't have the limit, I'm not sure you really need that)

[jvoung (off chromium), 2013/06/14 20:38:03]

Done.

+  std::stringstream ss(str);
+  std::string item;
+  int num = 0;
+  while (std::getline(ss, item, delim)) {
+    ++num;
+    if (limit == 0 || num < limit) {
+      elems->push_back(item);
+    } else {
+      // Loop through the rest, slap back on the delim in case
+      // we ate it earlier. We can't just use "ss >> rest"
+      // because the operator>> also tokenizes.
+      std::string more;
+      while (std::getline(ss, more, delim)) {
+        item.push_back(delim);

[dmichael (off chromium), 2013/06/14 16:21:43]

nit: you may tack on the delimiter even if it wasn't there. That's probably
okay?

[jvoung (off chromium), 2013/06/14 20:38:03]

I'm not sure if that's the case?

(1) item should be some non-empty string when here, from the earlier call to
getline at 19.
(2) this while-loop should only loop if there is more.

Then you do item + delim + more, else item remains the same?

Will split the two functions as you suggest to make easier to be sure =)

[dmichael (off chromium), 2013/06/14 21:09:03]

My bad, I was misunderstanding it. But I like it better now, anyway :)

+        item += more;
+      }
+      elems->push_back(item);
+      return;
+    }
+  }
+}
+
+}  // namespace
+
+namespace plugin {
+
+NaClHttpResponseHeaders::NaClHttpResponseHeaders() {}
+
+NaClHttpResponseHeaders::~NaClHttpResponseHeaders() {}
+
+void NaClHttpResponseHeaders::Parse(std::string headers_str) {
+  // PPAPI response headers are \n delimited. Separate out the lines.
+  std::vector<std::string> lines;
+  SplitString(headers_str, '\n', &lines, 0);
+
+  for (size_t i = 0; i < lines.size(); ++i) {
+    std::vector<std::string> tokens;
+    // Split along the key-value pair separator char.
+    SplitString(lines[i], ':', &tokens, 2);
+    if (tokens.size() != 2)

[dmichael (off chromium), 2013/06/14 16:21:43]

It actually seems to me that rather than make SplitString so general, it might
be better to have a separate algorithm or function that just splits up to the
first occurrence of the delimiter. You'd have two algorithms, but each would be
easier to understand I think. And when you're only looking for the first
occurrence of the delimiter, you might end up doing a lot less work.

[jvoung (off chromium), 2013/06/14 20:38:03]

Sounds good, that will also make it easier to use Tokenize() later, if we do end
up needing this.

+      // Ignore funny header lines that don't have the key-value separator.
+      continue;
+    std::string key = tokens[0];
+    // Also ignore keys that start with white-space (they are invalid).
+    // See: HttpResponseHeadersTest.NormalizeHeadersLeadingWhitespace.
+    if (key[0] == ' ' || key[0] == '\t')
+      continue;
+    // Strip trailing whitespace from the key to normalize.
+    size_t pos = key.find_last_not_of(" \t");
+    if (pos != std::string::npos)
+      key.erase(pos + 1);
+    // Strip leading whitespace from the value to normalize.
+    std::string value = tokens[1];
+    value.erase(0, value.find_first_not_of(" \t"));

[dmichael (off chromium), 2013/06/14 16:21:43]

The above could be replaced with TrimWhitespaceASCII when this moves to
Chromium...

However, it does not deal with all types of UTF-8 whitespace. That would be a
good TODO, perhaps?

[jvoung (off chromium), 2013/06/14 20:38:03]

Added a TODO.  I think http headers are only allowed tabs and space as
whitespace within a line:
http://www.w3.org/Protocols/rfc2616/rfc2616-sec2.html#sec2

       LWS            = [CRLF] 1*( SP | HT )

but perhaps in practice it is more messy?

+    header_entries_.push_back(Entry(key, value));
+  }
+}
+
+std::string NaClHttpResponseHeaders::GetCacheValidators() {
+  std::string result;
+  for (size_t i = 0; i < header_entries_.size(); ++i) {
+    const Entry& entry = header_entries_[i];
+    std::string key = entry.first;
+    std::transform(key.begin(), key.end(), key.begin(), ::tolower);
+    if (key.compare("etag") == 0 || key.compare("last-modified") == 0) {
+      if (!result.empty())

[dmichael (off chromium), 2013/06/14 16:21:43]

(could become LowerCaseEqualsASCII later)

[jvoung (off chromium), 2013/06/14 20:38:03]

Done.

+        result += "&";
+      // Return the original headers, not the tolower ones.
+      result += entry.first + ":" + entry.second;
+    }
+  }
+  return result;
+}
+
+bool NaClHttpResponseHeaders::CacheControlNoStore() {
+  for (size_t i = 0; i < header_entries_.size(); ++i) {
+    const Entry& entry = header_entries_[i];
+    std::string key = entry.first;
+    std::transform(key.begin(), key.end(), key.begin(), ::tolower);
+    if (key.compare("cache-control") == 0) {
+      std::string cc = entry.second;
+      std::transform(cc.begin(), cc.end(), cc.begin(), ::tolower);
+      if (entry.second.find("no-store") != std::string::npos)
+        return true;
+    }
+  }
+  return false;
+}
+
+}  // namespace plugin