const char* const RESOURCES_ROOT_NAMESPACE = "http://schemas.android.com/apk/res/";
const char* const RESOURCES_ANDROID_NAMESPACE = "http://schemas.android.com/apk/res/android";
+const char* const RESOURCES_AUTO_PACKAGE_NAMESPACE = "http://schemas.android.com/apk/res-auto";
const char* const RESOURCES_ROOT_PRV_NAMESPACE = "http://schemas.android.com/apk/prv/res/";
const char* const XLIFF_XMLNS = "urn:oasis:names:tc:xliff:document:1.2";
}
static const String16 RESOURCES_PREFIX(RESOURCES_ROOT_NAMESPACE);
+static const String16 RESOURCES_PREFIX_AUTO_PACKAGE(RESOURCES_AUTO_PACKAGE_NAMESPACE);
static const String16 RESOURCES_PRV_PREFIX(RESOURCES_ROOT_PRV_NAMESPACE);
+static const String16 RESOURCES_TOOLS_NAMESPACE("http://schemas.android.com/tools");
-String16 getNamespaceResourcePackage(String16 namespaceUri, bool* outIsPublic)
+String16 getNamespaceResourcePackage(String16 appPackage, String16 namespaceUri, bool* outIsPublic)
{
//printf("%s starts with %s?\n", String8(namespaceUri).string(),
// String8(RESOURCES_PREFIX).string());
size_t prefixSize;
bool isPublic = true;
- if (namespaceUri.startsWith(RESOURCES_PREFIX)) {
+ if(namespaceUri.startsWith(RESOURCES_PREFIX_AUTO_PACKAGE)) {
+ NOISY(printf("Using default application package: %s -> %s\n", String8(namespaceUri).string(), String8(appPackage).string()));
+ isPublic = true;
+ return appPackage;
+ } else if (namespaceUri.startsWith(RESOURCES_PREFIX)) {
prefixSize = RESOURCES_PREFIX.size();
} else if (namespaceUri.startsWith(RESOURCES_PRV_PREFIX)) {
isPublic = false;
return String16(namespaceUri, namespaceUri.size()-prefixSize, prefixSize);
}
+status_t hasSubstitutionErrors(const char* fileName,
+ ResXMLTree* inXml,
+ String16 str16)
+{
+ const char16_t* str = str16.string();
+ const char16_t* p = str;
+ const char16_t* end = str + str16.size();
+
+ bool nonpositional = false;
+ int argCount = 0;
+
+ while (p < end) {
+ /*
+ * Look for the start of a Java-style substitution sequence.
+ */
+ if (*p == '%' && p + 1 < end) {
+ p++;
+
+ // A literal percent sign represented by %%
+ if (*p == '%') {
+ p++;
+ continue;
+ }
+
+ argCount++;
+
+ if (*p >= '0' && *p <= '9') {
+ do {
+ p++;
+ } while (*p >= '0' && *p <= '9');
+ if (*p != '$') {
+ // This must be a size specification instead of position.
+ nonpositional = true;
+ }
+ } else if (*p == '<') {
+ // Reusing last argument; bad idea since it can be re-arranged.
+ nonpositional = true;
+ p++;
+
+ // Optionally '$' can be specified at the end.
+ if (p < end && *p == '$') {
+ p++;
+ }
+ } else {
+ nonpositional = true;
+ }
+
+ // Ignore flags and widths
+ while (p < end && (*p == '-' ||
+ *p == '#' ||
+ *p == '+' ||
+ *p == ' ' ||
+ *p == ',' ||
+ *p == '(' ||
+ (*p >= '0' && *p <= '9'))) {
+ p++;
+ }
+
+ /*
+ * This is a shortcut to detect strings that are going to Time.format()
+ * instead of String.format()
+ *
+ * Comparison of String.format() and Time.format() args:
+ *
+ * String: ABC E GH ST X abcdefgh nost x
+ * Time: DEFGHKMS W Za d hkm s w yz
+ *
+ * Therefore we know it's definitely Time if we have:
+ * DFKMWZkmwyz
+ */
+ if (p < end) {
+ switch (*p) {
+ case 'D':
+ case 'F':
+ case 'K':
+ case 'M':
+ case 'W':
+ case 'Z':
+ case 'k':
+ case 'm':
+ case 'w':
+ case 'y':
+ case 'z':
+ return NO_ERROR;
+ }
+ }
+ }
+
+ p++;
+ }
+
+ /*
+ * If we have more than one substitution in this string and any of them
+ * are not in positional form, give the user an error.
+ */
+ if (argCount > 1 && nonpositional) {
+ SourcePos(String8(fileName), inXml->getLineNumber()).error(
+ "Multiple substitutions specified in non-positional format; "
+ "did you mean to add the formatted=\"false\" attribute?\n");
+ return NOT_ENOUGH_DATA;
+ }
+
+ return NO_ERROR;
+}
+
status_t parseStyledString(Bundle* bundle,
const char* fileName,
ResXMLTree* inXml,
const String16& endTag,
String16* outString,
Vector<StringPool::entry_style_span>* outSpans,
+ bool isFormatted,
bool pseudolocalize)
{
Vector<StringPool::entry_style_span> spanStack;
std::string pseudo = pseudolocalize_string(orig);
curString.append(String16(String8(pseudo.c_str())));
} else {
- curString.append(text);
+ if (isFormatted && hasSubstitutionErrors(fileName, inXml, text) != NO_ERROR) {
+ return UNKNOWN_ERROR;
+ } else {
+ curString.append(text);
+ }
}
} else if (code == ResXMLTree::START_TAG) {
const String16 element16(inXml->getElementName(&len));
}
spanStack.pop();
- if (empty) {
- fprintf(stderr, "%s:%d: WARNING: empty '%s' span found in text '%s'\n",
+ /*
+ * This warning seems to be just an irritation to most people,
+ * since it is typically introduced by translators who then never
+ * see the warning.
+ */
+ if (0 && empty) {
+ fprintf(stderr, "%s:%d: warning: empty '%s' span found in text '%s'\n",
fileName, inXml->getLineNumber(),
String8(spanTag).string(), String8(*outString).string());
printf("=?0x%x", (int)value.data);
} else if (value.dataType == Res_value::TYPE_STRING) {
printf("=\"%s\"",
- String8(block->getAttributeStringValue(i, &len)).string());
+ ResTable::normalizeForOutput(String8(block->getAttributeStringValue(i,
+ &len)).string()).string());
} else {
printf("=(type 0x%x)0x%x", (int)value.dataType, (int)value.data);
}
const char16_t* val = block->getAttributeStringValue(i, &len);
if (val != NULL) {
- printf(" (Raw: \"%s\")", String8(val).string());
+ printf(" (Raw: \"%s\")", ResTable::normalizeForOutput(String8(val).string()).
+ string());
}
printf("\n");
}
namespaces.pop();
} else if (code == ResXMLTree::TEXT) {
size_t len;
- printf("%sC: \"%s\"\n", prefix.string(), String8(block->getText(&len)).string());
+ printf("%sC: \"%s\"\n", prefix.string(),
+ ResTable::normalizeForOutput(String8(block->getText(&len)).string()).string());
}
}
, mFilename(filename)
, mStartLineNumber(0)
, mEndLineNumber(0)
+ , mUTF8(false)
{
if (isNamespace) {
mNamespacePrefix = s1;
XMLNode::XMLNode(const String8& filename)
: mFilename(filename)
{
+ memset(&mCharsValue, 0, sizeof(mCharsValue));
}
XMLNode::type XMLNode::getType() const
return mChildren;
}
+const String8& XMLNode::getFilename() const
+{
+ return mFilename;
+}
+
const Vector<XMLNode::attribute_entry>&
XMLNode::getAttributes() const
{
return mAttributes;
}
+const XMLNode::attribute_entry* XMLNode::getAttribute(const String16& ns,
+ const String16& name) const
+{
+ for (size_t i=0; i<mAttributes.size(); i++) {
+ const attribute_entry& ae(mAttributes.itemAt(i));
+ if (ae.ns == ns && ae.name == name) {
+ return &ae;
+ }
+ }
+
+ return NULL;
+}
+
+XMLNode::attribute_entry* XMLNode::editAttribute(const String16& ns,
+ const String16& name)
+{
+ for (size_t i=0; i<mAttributes.size(); i++) {
+ attribute_entry * ae = &mAttributes.editItemAt(i);
+ if (ae->ns == ns && ae->name == name) {
+ return ae;
+ }
+ }
+
+ return NULL;
+}
+
const String16& XMLNode::getCData() const
{
return mChars;
return mEndLineNumber;
}
+sp<XMLNode> XMLNode::searchElement(const String16& tagNamespace, const String16& tagName)
+{
+ if (getType() == XMLNode::TYPE_ELEMENT
+ && mNamespaceUri == tagNamespace
+ && mElementName == tagName) {
+ return this;
+ }
+
+ for (size_t i=0; i<mChildren.size(); i++) {
+ sp<XMLNode> found = mChildren.itemAt(i)->searchElement(tagNamespace, tagName);
+ if (found != NULL) {
+ return found;
+ }
+ }
+
+ return NULL;
+}
+
+sp<XMLNode> XMLNode::getChildElement(const String16& tagNamespace, const String16& tagName)
+{
+ for (size_t i=0; i<mChildren.size(); i++) {
+ sp<XMLNode> child = mChildren.itemAt(i);
+ if (child->getType() == XMLNode::TYPE_ELEMENT
+ && child->mNamespaceUri == tagNamespace
+ && child->mElementName == tagName) {
+ return child;
+ }
+ }
+
+ return NULL;
+}
+
status_t XMLNode::addChild(const sp<XMLNode>& child)
{
if (getType() == TYPE_CDATA) {
return NO_ERROR;
}
+status_t XMLNode::insertChildAt(const sp<XMLNode>& child, size_t index)
+{
+ if (getType() == TYPE_CDATA) {
+ SourcePos(mFilename, child->getStartLineNumber()).error("Child to CDATA node.");
+ return UNKNOWN_ERROR;
+ }
+ //printf("Adding child %p to parent %p\n", child.get(), this);
+ mChildren.insertAt(child, index);
+ return NO_ERROR;
+}
+
status_t XMLNode::addAttribute(const String16& ns, const String16& name,
const String16& value)
{
SourcePos(mFilename, getStartLineNumber()).error("Child to CDATA node.");
return UNKNOWN_ERROR;
}
- attribute_entry e;
- e.index = mNextAttributeIndex++;
- e.ns = ns;
- e.name = name;
- e.string = value;
- mAttributes.add(e);
- mAttributeOrder.add(e.index, mAttributes.size()-1);
+
+ if (ns != RESOURCES_TOOLS_NAMESPACE) {
+ attribute_entry e;
+ e.index = mNextAttributeIndex++;
+ e.ns = ns;
+ e.name = name;
+ e.string = value;
+ mAttributes.add(e);
+ mAttributeOrder.add(e.index, mAttributes.size()-1);
+ }
return NO_ERROR;
}
const attribute_entry& e = mAttributes.itemAt(i);
if (e.ns.size() <= 0) continue;
bool nsIsPublic;
- String16 pkg(getNamespaceResourcePackage(e.ns, &nsIsPublic));
+ String16 pkg(getNamespaceResourcePackage(String16(assets->getPackage()), e.ns, &nsIsPublic));
NOISY(printf("Elem %s %s=\"%s\": namespace(%s) %s ===> %s\n",
String8(getElementName()).string(),
String8(e.name).string(),
status_t XMLNode::flatten(const sp<AaptFile>& dest,
bool stripComments, bool stripRawValues) const
{
- StringPool strings;
+ StringPool strings(mUTF8);
Vector<uint32_t> resids;
// First collect just the strings for attribute names that have a
collect_attr_strings(dest, outResIds, true);
int i;
- if (mNamespacePrefix.size() > 0) {
- dest->add(mNamespacePrefix, true);
- }
- if (mNamespaceUri.size() > 0) {
- dest->add(mNamespaceUri, true);
+ if (RESOURCES_TOOLS_NAMESPACE != mNamespaceUri) {
+ if (mNamespacePrefix.size() > 0) {
+ dest->add(mNamespacePrefix, true);
+ }
+ if (mNamespaceUri.size() > 0) {
+ dest->add(mNamespaceUri, true);
+ }
}
if (mElementName.size() > 0) {
dest->add(mElementName, true);
const void* extData = NULL;
size_t extSize = 0;
ResXMLTree_attribute attr;
+ bool writeCurrentNode = true;
const size_t NA = mAttributes.size();
const size_t NC = mChildren.size();
const String16 style16("style");
const type type = getType();
-
+
memset(&node, 0, sizeof(node));
memset(&attr, 0, sizeof(attr));
node.header.headerSize = htods(sizeof(node));
}
}
} else if (type == TYPE_NAMESPACE) {
- node.header.type = htods(RES_XML_START_NAMESPACE_TYPE);
- extData = &namespaceExt;
- extSize = sizeof(namespaceExt);
- memset(&namespaceExt, 0, sizeof(namespaceExt));
- if (mNamespacePrefix.size() > 0) {
- namespaceExt.prefix.index = htodl(strings.offsetForString(mNamespacePrefix));
+ if (mNamespaceUri == RESOURCES_TOOLS_NAMESPACE) {
+ writeCurrentNode = false;
} else {
- namespaceExt.prefix.index = htodl((uint32_t)-1);
+ node.header.type = htods(RES_XML_START_NAMESPACE_TYPE);
+ extData = &namespaceExt;
+ extSize = sizeof(namespaceExt);
+ memset(&namespaceExt, 0, sizeof(namespaceExt));
+ if (mNamespacePrefix.size() > 0) {
+ namespaceExt.prefix.index = htodl(strings.offsetForString(mNamespacePrefix));
+ } else {
+ namespaceExt.prefix.index = htodl((uint32_t)-1);
+ }
+ namespaceExt.prefix.index = htodl(strings.offsetForString(mNamespacePrefix));
+ namespaceExt.uri.index = htodl(strings.offsetForString(mNamespaceUri));
}
- namespaceExt.prefix.index = htodl(strings.offsetForString(mNamespacePrefix));
- namespaceExt.uri.index = htodl(strings.offsetForString(mNamespaceUri));
LOG_ALWAYS_FATAL_IF(NA != 0, "Namespace nodes can't have attributes!");
} else if (type == TYPE_CDATA) {
node.header.type = htods(RES_XML_CDATA_TYPE);
node.header.size = htodl(sizeof(node) + extSize + (sizeof(attr)*NA));
- dest->writeData(&node, sizeof(node));
- if (extSize > 0) {
- dest->writeData(extData, extSize);
+ if (writeCurrentNode) {
+ dest->writeData(&node, sizeof(node));
+ if (extSize > 0) {
+ dest->writeData(extData, extSize);
+ }
}
for (i=0; i<NA; i++) {
dest->writeData(&node, sizeof(node));
dest->writeData(&endElementExt, sizeof(endElementExt));
} else if (type == TYPE_NAMESPACE) {
- node.header.type = htods(RES_XML_END_NAMESPACE_TYPE);
- node.lineNumber = htodl(getEndLineNumber());
- node.comment.index = htodl((uint32_t)-1);
- node.header.size = htodl(sizeof(node)+extSize);
- dest->writeData(&node, sizeof(node));
- dest->writeData(extData, extSize);
+ if (writeCurrentNode) {
+ node.header.type = htods(RES_XML_END_NAMESPACE_TYPE);
+ node.lineNumber = htodl(getEndLineNumber());
+ node.comment.index = htodl((uint32_t)-1);
+ node.header.size = htodl(sizeof(node)+extSize);
+ dest->writeData(&node, sizeof(node));
+ dest->writeData(extData, extSize);
+ }
}
return NO_ERROR;