Add Unicode property escape support to RegExp (#1295)

Summary:
Extends RegExp to handle `\p` and `\P` Unicode property escapes, standalone and within character classes, for Unicode mode. The implementation extends the existing idea of tables of codepoint ranges to cover all of the properties explicitly required by ES262.

The `genUnicodeTable.py` script was updated to generate code related to the binary and non-binary Unicode properties explicitly mentioned in ES262 ([ref](https://tc39.es/ecma262/multipage/text-processing.html#sec-runtime-semantics-unicodematchproperty-p)); based on the Unicode 15.1.0 data files.

Closes #1027

Pull Request resolved: #1295

Test Plan:
- Added `regexp_unicode_properties.js` to the Hermes test suite
	- Binary properties and the General_Category properties
	- Non-binary properties (Script=Latin, Script_Extensions=Thai)
	- Inverted character class escapes `\p{…}` and `\P{…}`
	- All of the above atom escapes in and out of character classes
	- Exercise the parser for the above forms, and incomplete forms
- test262 suite via `hermes/utils/testsuite/run_testsuite.py`
	- I had to use `--test-skiplist` otherwise around 417 tests (in `test262/test/built-ins/RegExp/property-escapes`) are skipped with the reason "Skipping test with 'const'". This was confusing to me since most of these do pass, but I couldn't find an explainer for this skip.
	- There are some failures here, at least one of which is caused by [test262 not being updated for Unicode 15.1.0](tc39/test262#3945), which I see as test suite issue rather than an implementation issue.

Reviewed By: avp

Differential Revision: D56493540

Pulled By: neildhar

fbshipit-source-id: 55c86e2d321fc03601b418e0244c099a98759f3f

CMakeLists.txt

@@ -275,6 +275,9 @@ set(EMSCRIPTEN_FASTCOMP OFF CACHE BOOL
 set(HERMES_ENABLE_INTL OFF CACHE BOOL
   "Enable JS Intl support (WIP)")
 
+set(HERMES_ENABLE_UNICODE_REGEXP_PROPERTY_ESCAPES ON CACHE BOOL
+  "Enable RegExp Unicode Property Escapes support")
+
 set(HERMES_ENABLE_TEST_SUITE ON CACHE BOOL
   "Enable the test suite")
 
@@ -470,6 +473,10 @@ if (HERMES_ENABLE_INTL)
   add_definitions(-DHERMES_ENABLE_INTL)
 endif()
 
+if (HERMES_ENABLE_UNICODE_REGEXP_PROPERTY_ESCAPES)
+  add_definitions(-DHERMES_ENABLE_UNICODE_REGEXP_PROPERTY_ESCAPES)
+endif()
+
 if (HERMES_ENABLE_WERROR)
   # Turn all warnings into errors on GCC-compatible compilers.
   if (GCC_COMPATIBLE)
@@ -735,6 +742,7 @@ if(HERMES_ENABLE_TEST_SUITE)
     unittests_dir=${CMAKE_CURRENT_BINARY_DIR}/unittests
     debugger_enabled=${HERMES_ENABLE_DEBUGGER}
     intl_enabled=${HERMES_ENABLE_INTL}
+    regexp_unicode_properties_enabled=${HERMES_ENABLE_UNICODE_REGEXP_PROPERTY_ESCAPES}
     use_flowparser=${HERMES_USE_FLOWPARSER}
     hbc_deltaprep=${HERMES_TOOLS_OUTPUT_DIR}/hbc-deltaprep
     dependency_extractor=${HERMES_TOOLS_OUTPUT_DIR}/dependency-extractor

doc/RegExp.md

@@ -21,7 +21,4 @@ As of this writing, Hermes regexp supports
 1. All of ES6, including global, case-insensitive, multiline, sticky, and Unicode (and legacy).
 1. ES9 lookbehinds.
 1. Named capture groups.
-
-Missing features from ES9 include:
-
 1. Unicode property escapes.

include/hermes/Platform/Unicode/CharacterProperties.h

@@ -10,6 +10,9 @@
 
 #include <cassert>
 #include <cstdint>
+#include <string>
+
+#include "llvh/ADT/ArrayRef.h"
 
 namespace hermes {
 
@@ -100,6 +103,16 @@ inline bool isUnicodeIDContinue(uint32_t cp) {
       cp == UNICODE_ZWNJ || cp == UNICODE_ZWJ;
 }
 
+/// \return true if the codepoint is valid in a unicode property name
+inline bool isUnicodePropertyName(uint32_t ch) {
+  return ch == '_' || ((ch | 32) >= 'a' && (ch | 32) <= 'z');
+}
+
+/// \return true if the codepoint is valid in a unicode property value
+inline bool isUnicodePropertyValue(uint32_t ch) {
+  return isUnicodePropertyName(ch) || isUnicodeDigit(ch);
+}
+
 /// \return the canonicalized value of \p cp, following ES9 21.2.2.8.2.
 uint32_t canonicalize(uint32_t cp, bool unicode);
 
@@ -108,6 +121,21 @@ class CodePointSet;
 /// any character in \p set, following ES9 21.2.2.8.2.
 CodePointSet makeCanonicallyEquivalent(const CodePointSet &set, bool unicode);
 
+struct UnicodeRangePoolRef;
+
+// Create a codepoint range array from a Unicode \p propertyName and \p
+// propertyValue.
+llvh::ArrayRef<UnicodeRangePoolRef> unicodePropertyRanges(
+    std::string_view propertyName,
+    std::string_view propertyValue);
+
+/// Add a codepoint range array of codepoints to \p receiver, typically used in
+/// conjuction with unicodePropertyRanges.
+void addRangeArrayPoolToBracket(
+    CodePointSet *receiver,
+    const llvh::ArrayRef<UnicodeRangePoolRef> rangeArrayPool,
+    bool inverted);
+
 } // namespace hermes
 
 #endif // HERMES_PLATFORMUNICODE_CHARACTERPROPERTIES_H

include/hermes/Regex/RegexNode.h

@@ -827,6 +827,12 @@ class BracketNode : public Node {
     classes_.push_back(cls);
   }
 
+  void addCodePointRanges(
+      llvh::ArrayRef<UnicodeRangePoolRef> rangeArray,
+      bool inverted = false) {
+    addRangeArrayPoolToBracket(&codePointSet_, rangeArray, inverted);
+  }
+
   virtual MatchConstraintSet matchConstraints() const override {
     MatchConstraintSet result = 0;
     if (!canMatchASCII())

include/hermes/Regex/RegexTypes.h

@@ -22,6 +22,9 @@
 #include "llvh/ADT/SmallString.h"
 
 namespace hermes {
+
+struct UnicodeRangePoolRef;
+
 namespace regex {
 namespace constants {
 
@@ -122,7 +125,10 @@ enum class ErrorType {
   InvalidNamedReference,
 
   /// Reference to nonexistent capture group.
-  NonexistentNamedCaptureReference
+  NonexistentNamedCaptureReference,
+
+  /// Invalid Unicode property name or value
+  InvalidPropertyName,
 };
 
 /// \return an error message for the given \p error.
@@ -158,6 +164,8 @@ inline const char *messageForError(ErrorType error) {
       return "Invalid named reference";
     case ErrorType::NonexistentNamedCaptureReference:
       return "Nonexistent named capture reference";
+    case ErrorType::InvalidPropertyName:
+      return "Invalid property name";
     case ErrorType::None:
       return "No error";
   }
@@ -209,6 +217,19 @@ struct CharacterClass {
   CharacterClass(Type type, bool invert) : type_(type), inverted_(invert) {}
 };
 
+// Type wrapping up a Unicode codepoint range array.
+struct CharacterClassCodepointRanges {
+  llvh::ArrayRef<UnicodeRangePoolRef> rangeArray;
+
+  // Whether the class is inverted (\P instead of \p).
+  bool inverted_;
+
+  CharacterClassCodepointRanges(
+      llvh::ArrayRef<UnicodeRangePoolRef> rangeArray,
+      bool inverted)
+      : rangeArray(rangeArray), inverted_(inverted) {}
+};
+
 // Struct representing flags which may be used when constructing the RegExp
 class SyntaxFlags {
  private:

lib/CompilerDriver/CompilerDriver.cpp

@@ -2184,6 +2184,9 @@ void printHermesVersion(
 #endif
 #ifdef HERMESVM_CONTIGUOUS_HEAP
       << "    Contiguous Heap\n"
+#endif
+#ifdef HERMES_ENABLE_UNICODE_REGEXP_PROPERTY_ESCAPES
+      << "    Unicode RegExp Property Escapes\n"
 #endif
       << "    Zip file input\n";
   }

lib/Platform/Unicode/CharacterProperties.cpp

@@ -11,6 +11,7 @@
 #include <algorithm>
 #include <climits>
 #include <iterator>
+#include <string>
 #include <utility>
 
 namespace hermes {
@@ -210,4 +211,129 @@ uint32_t canonicalize(uint32_t cp, bool unicode) {
   }
 }
 
+#ifdef HERMES_ENABLE_UNICODE_REGEXP_PROPERTY_ESCAPES
+
+/// Find a matching entry (such as \p NameMapEntry or \p RangeMapEntry) by
+/// matching a string \p name against the entry's \p name field.
+template <class T>
+static const T *findMapEntry(
+    const llvh::ArrayRef<T> &arrayRef,
+    const std::string_view name) {
+  auto it = std::lower_bound(
+      std::begin(arrayRef),
+      std::end(arrayRef),
+      name,
+      [](const T &a, std::string_view b) {
+        return UNICODE_DATA_STRING_POOL.compare(a.name.offset, a.name.size, b) <
+            0;
+      });
+  if (it == std::end(arrayRef) ||
+      UNICODE_DATA_STRING_POOL.compare(it->name.offset, it->name.size, name) !=
+          0) {
+    return nullptr;
+  }
+  return it;
+}
+
+llvh::ArrayRef<UnicodeRangePoolRef> unicodePropertyRanges(
+    std::string_view propertyName,
+    std::string_view propertyValue) {
+  const NameMapEntry *canonicalNameEntry;
+  llvh::ArrayRef<RangeMapEntry> rangeMap;
+
+  if (propertyValue.empty()) {
+    // There was no property value, this is either a binary property or a value
+    // from General_Category, as per `LoneUnicodePropertyNameOrValue`.
+    if ((canonicalNameEntry = findMapEntry(
+             llvh::ArrayRef(canonicalPropertyNameMap_BinaryProperty),
+             propertyName))) {
+      rangeMap = unicodePropertyRangeMap_BinaryProperty;
+    } else if ((canonicalNameEntry = findMapEntry(
+                    llvh::ArrayRef(canonicalPropertyNameMap_GeneralCategory),
+                    propertyName))) {
+      rangeMap = unicodePropertyRangeMap_GeneralCategory;
+    }
+  } else {
+    // There was a property value, assume the name is a category.
+    if ((propertyName == "General_Category" || propertyName == "gc") &&
+        (canonicalNameEntry = findMapEntry(
+             llvh::ArrayRef(canonicalPropertyNameMap_GeneralCategory),
+             propertyValue))) {
+      rangeMap = unicodePropertyRangeMap_GeneralCategory;
+    } else if (
+        (propertyName == "Script" || propertyName == "sc") &&
+        (canonicalNameEntry = findMapEntry(
+             llvh::ArrayRef(canonicalPropertyNameMap_Script), propertyValue))) {
+      rangeMap = unicodePropertyRangeMap_Script;
+    } else if (
+        (propertyName == "Script_Extensions" || propertyName == "scx") &&
+        // Since Script_Extensions is a superset of Script, they share
+        // a name map.
+        (canonicalNameEntry = findMapEntry(
+             llvh::ArrayRef(canonicalPropertyNameMap_Script), propertyValue))) {
+      rangeMap = unicodePropertyRangeMap_ScriptExtensions;
+    } else {
+      return llvh::ArrayRef<UnicodeRangePoolRef>();
+    }
+  }
+
+  if (canonicalNameEntry == nullptr) {
+    return llvh::ArrayRef<UnicodeRangePoolRef>();
+  }
+
+  // Look up the range arrays for the property.
+  auto rangeMapEntry = findMapEntry(
+      rangeMap,
+      UNICODE_DATA_STRING_POOL.substr(
+          canonicalNameEntry->canonical.offset,
+          canonicalNameEntry->canonical.size));
+  if (rangeMapEntry == nullptr) {
+    return llvh::ArrayRef<UnicodeRangePoolRef>();
+  }
+
+  return llvh::ArrayRef{
+      &UNICODE_RANGE_ARRAY_POOL[rangeMapEntry->rangeArrayPoolOffset],
+      rangeMapEntry->rangeArraySize};
+}
+
+void addRangeArrayPoolToBracket(
+    CodePointSet *receiver,
+    const llvh::ArrayRef<UnicodeRangePoolRef> rangeArrayPool,
+    bool inverted) {
+  for (auto rangePoolRef : rangeArrayPool) {
+    auto rangePool = llvh::ArrayRef<UnicodeRange>{
+        &UNICODE_RANGE_POOL[rangePoolRef.offset], rangePoolRef.size};
+
+    if (inverted) {
+      uint32_t last = 0;
+      for (auto range : rangePool) {
+        receiver->add(CodePointRange{last, range.first - last});
+        last = range.second + 1;
+      }
+      // Add the final range.
+      receiver->add(CodePointRange{last, UNICODE_MAX_VALUE - last});
+    } else {
+      for (auto range : rangePool) {
+        const uint32_t length = range.second - range.first + 1;
+        receiver->add(CodePointRange{range.first, length});
+      }
+    }
+  }
+}
+
+#else
+
+llvh::ArrayRef<UnicodeRangePoolRef> unicodePropertyRanges(
+    std::string_view propertyName,
+    std::string_view propertyValue) {
+  return llvh::ArrayRef<UnicodeRangePoolRef>();
+}
+
+void addRangeArrayPoolToBracket(
+    CodePointSet *receiver,
+    const llvh::ArrayRef<UnicodeRangePoolRef> rangeArrayPool,
+    bool inverted) {}
+
+#endif
+
 } // namespace hermes