Add ArgDefCase conversion to tensorflow::str_util

PiperOrigin-RevId: 166905993

3 files changed, 95 insertions, 0 deletions

diff --git a/tensorflow/core/lib/strings/str_util.cc b/tensorflow/core/lib/strings/str_util.cc
index c68e14f09f..1159304724 100644
--- a/tensorflow/core/lib/strings/str_util.cc
+++ b/tensorflow/core/lib/strings/str_util.cc
@@ -248,6 +248,58 @@ string Uppercase(StringPiece s) {
   return result;
 }
 
+string ArgDefCase(StringPiece s) {
+  const int n = s.size();
+
+  // Compute the size of resulting string.
+  // Number of extra underscores we will need to add.
+  int extra_us = 0;
+  // Number of non-alpha chars in the beginning to skip.
+  int to_skip = 0;
+  for (int i = 0; i < n; ++i) {
+    // If we are skipping and current letter is non-alpha, skip it as well
+    if (i == to_skip && !isalpha(s[i])) {
+      ++to_skip;
+      continue;
+    }
+
+    // If we are here, we are not skipping any more.
+    // If this letter is upper case, not the very first char in the
+    // resulting string, and previous letter isn't replaced with an underscore,
+    // we will need to insert an underscore.
+    if (isupper(s[i]) && i != to_skip && i > 0 && isalnum(s[i - 1])) {
+      ++extra_us;
+    }
+  }
+
+  // Initialize result with all '_'s. There is no string
+  // constructor that does not initialize memory.
+  string result(n + extra_us - to_skip, '_');
+  // i - index into s
+  // j - index into result
+  for (int i = to_skip, j = 0; i < n; ++i, ++j) {
+    DCHECK_LT(j, result.size());
+    char c = s[i];
+    // If c is not alphanumeric, we don't need to do anything
+    // since there is already an underscore in its place.
+    if (isalnum(c)) {
+      if (isupper(c)) {
+        // If current char is upper case, we might need to insert an
+        // underscore.
+        if (i != to_skip) {
+          DCHECK_GT(j, 0);
+          if (result[j - 1] != '_') ++j;
+        }
+        result[j] = tolower(c);
+      } else {
+        result[j] = c;
+      }
+    }
+  }
+
+  return result;
+}
+
 void TitlecaseString(string* s, StringPiece delimiters) {
   bool upper = true;
   for (string::iterator ss = s->begin(); ss != s->end(); ++ss) {
diff --git a/tensorflow/core/lib/strings/str_util.h b/tensorflow/core/lib/strings/str_util.h
index 669f0d3c52..8cea0f0718 100644
--- a/tensorflow/core/lib/strings/str_util.h
+++ b/tensorflow/core/lib/strings/str_util.h
@@ -81,6 +81,17 @@ string Lowercase(StringPiece s);
 // Return upper-cased version of s.
 string Uppercase(StringPiece s);
 
+// Converts "^2ILoveYou!" to "i_love_you_". More specifically:
+// - converts all non-alphanumeric characters to underscores
+// - replaces each occurence of a capital letter (except the very
+//   first character and if there is already an '_' before it) with '_'
+//   followed by this letter in lower case
+// - Skips leading non-alpha characters
+// This method is useful for producing strings matching "[a-z][a-z0-9_]*"
+// as required by OpDef.ArgDef.name. The resulting string is either empty or
+// matches this regex.
+string ArgDefCase(StringPiece s);
+
 // Capitalize first character of each word in "*s".  "delimiters" is a
 // set of characters that can be used as word boundaries.
 void TitlecaseString(string* s, StringPiece delimiters);
diff --git a/tensorflow/core/lib/strings/str_util_test.cc b/tensorflow/core/lib/strings/str_util_test.cc
index 040f7447e4..5c735a87a3 100644
--- a/tensorflow/core/lib/strings/str_util_test.cc
+++ b/tensorflow/core/lib/strings/str_util_test.cc
@@ -338,6 +338,38 @@ TEST(Uppercase, Basic) {
   EXPECT_EQ("HELLO WORLD", str_util::Uppercase("Hello World"));
 }
 
+TEST(SnakeCase, Basic) {
+  EXPECT_EQ("", str_util::ArgDefCase(""));
+  EXPECT_EQ("", str_util::ArgDefCase("!"));
+  EXPECT_EQ("", str_util::ArgDefCase("5"));
+  EXPECT_EQ("", str_util::ArgDefCase("!:"));
+  EXPECT_EQ("", str_util::ArgDefCase("5-5"));
+  EXPECT_EQ("", str_util::ArgDefCase("_!"));
+  EXPECT_EQ("", str_util::ArgDefCase("_5"));
+  EXPECT_EQ("a", str_util::ArgDefCase("_a"));
+  EXPECT_EQ("a", str_util::ArgDefCase("_A"));
+  EXPECT_EQ("i", str_util::ArgDefCase("I"));
+  EXPECT_EQ("i", str_util::ArgDefCase("i"));
+  EXPECT_EQ("i_", str_util::ArgDefCase("I%"));
+  EXPECT_EQ("i_", str_util::ArgDefCase("i%"));
+  EXPECT_EQ("i", str_util::ArgDefCase("%I"));
+  EXPECT_EQ("i", str_util::ArgDefCase("-i"));
+  EXPECT_EQ("i", str_util::ArgDefCase("3i"));
+  EXPECT_EQ("i", str_util::ArgDefCase("32i"));
+  EXPECT_EQ("i3", str_util::ArgDefCase("i3"));
+  EXPECT_EQ("i_a3", str_util::ArgDefCase("i_A3"));
+  EXPECT_EQ("i_i", str_util::ArgDefCase("II"));
+  EXPECT_EQ("i_i", str_util::ArgDefCase("I_I"));
+  EXPECT_EQ("i__i", str_util::ArgDefCase("I__I"));
+  EXPECT_EQ("i_i_32", str_util::ArgDefCase("II-32"));
+  EXPECT_EQ("ii_32", str_util::ArgDefCase("Ii-32"));
+  EXPECT_EQ("hi_there", str_util::ArgDefCase("HiThere"));
+  EXPECT_EQ("hi_hi", str_util::ArgDefCase("Hi!Hi"));
+  EXPECT_EQ("hi_hi", str_util::ArgDefCase("HiHi"));
+  EXPECT_EQ("hihi", str_util::ArgDefCase("Hihi"));
+  EXPECT_EQ("hi_hi", str_util::ArgDefCase("Hi_Hi"));
+}
+
 TEST(TitlecaseString, Basic) {
   string s = "sparse_lookup";
   str_util::TitlecaseString(&s, "_");