Optimize hotspot in DependencySet.process().

--
MOS_MIGRATED_REVID=122050015

1 files changed, 52 insertions, 72 deletions

diff --git a/src/main/java/com/google/devtools/build/lib/util/DependencySet.java b/src/main/java/com/google/devtools/build/lib/util/DependencySet.java
index cb8994b3f0..a51e98649b 100644
--- a/src/main/java/com/google/devtools/build/lib/util/DependencySet.java
+++ b/src/main/java/com/google/devtools/build/lib/util/DependencySet.java
@@ -24,8 +24,6 @@ import java.nio.charset.StandardCharsets;
 import java.util.ArrayList;
 import java.util.Collection;
 import java.util.Collections;
-import java.util.regex.Matcher;
-import java.util.regex.Pattern;
 
 /**
  * Representation of a set of file dependencies for a given output file. There
@@ -49,10 +47,6 @@ import java.util.regex.Pattern;
  */
 public final class DependencySet {
 
-  private static final Pattern DOTD_MERGED_LINE_SEPARATOR = Pattern.compile("\\\\[\n\r]+");
-  private static final Pattern DOTD_LINE_SEPARATOR = Pattern.compile("[\n\r]+");
-  private static final Pattern DOTD_DEP = Pattern.compile("(?:[^\\s\\\\]++|\\\\ |\\\\)+");
-
   /**
    * The set of dependent files that this DependencySet embodies. They are all
    * Path with the same FileSystem  A tree set is used to ensure that we
@@ -73,7 +67,7 @@ public final class DependencySet {
   public void setOutputFileName(String outputFileName) {
     this.outputFileName = outputFileName;
   }
-  
+
   /**
    * Constructs a new empty DependencySet instance.
    */
@@ -130,71 +124,57 @@ public final class DependencySet {
     if (content.length > 0 && content[content.length - 1] != '\n') {
       throw new IOException("File does not end in a newline");
     }
-    // true if there is a CR in the input.
-    boolean cr = content.length > 0 && content[0] == '\r';
-    // true if there is more than one line in the input, not counting \-wrapped lines.
-    boolean multiline = false;
-
-    byte prevByte = ' ';
-    for (int i = 1; i < content.length; i++) {
-      byte b = content[i];
-      if (cr || b == '\r') {
-        // CR found, abort since our little loop here does not deal with CR/LFs.
-        cr = true;
-        break;
-      }
-      if (b == '\n') {
-        // Merge lines wrapped using backslashes.
-        if (prevByte == '\\') {
-          content[i] = ' ';
-          content[i - 1] = ' ';
-        } else {
-          multiline = true;
-        }
-      }
-      prevByte = b;
-    }
-
-    if (!cr && content.length > 0 && content[content.length - 1] == '\n') {
-      content[content.length - 1] = ' ';
-    }
-
-    String s = new String(content, StandardCharsets.UTF_8);
-    if (cr) {
-      s = DOTD_MERGED_LINE_SEPARATOR.matcher(s).replaceAll(" ").trim();
-      multiline = true;
-    }
-    return process(s, multiline);
-  }
-
-  private DependencySet process(String contents, boolean multiline) {
-    String[] lines;
-    if (!multiline) {
-      // Microoptimization: skip the usually unnecessary expensive-ish splitting step if there is
-      // only one target. This saves about 20% of CPU time.
-      lines = new String[] { contents };
-    } else {
-      lines = DOTD_LINE_SEPARATOR.split(contents);
-    }
-
-    for (String line : lines) {
-      // Split off output file name.
-      int pos = line.indexOf(':');
-      if (pos == -1) {
-        continue;
-      }
-      outputFileName = line.substring(0, pos);
-      
-      String deps = line.substring(pos + 1);
-
-      Matcher m = DOTD_DEP.matcher(deps);
-      while (m.find()) {
-        String token = m.group();
-        // Process escaped spaces.
-        if (token.contains("\\ ")) {
-          token = token.replace("\\ ", " ");
-        }
-        addDependency(new PathFragment(token).normalize());
+    int w = 0;
+    for (int r = 0; r < content.length; ++r) {
+      final byte c = content[r];
+      switch (c) {
+        case ' ':
+        case '\n':
+        case '\r':
+          if (w > 0) {
+            String s = new String(content, 0, w, StandardCharsets.UTF_8);
+            addDependency(new PathFragment(s).normalize());
+            w = 0;
+          }
+          break;
+        case ':':
+          // Normally this indicates the output file, but it might be part of a filename on Windows.
+          // Peek ahead at the next character.  This is in bounds because we checked for a
+          // terminating newline above.
+          switch (content[r + 1]) {
+            case ' ':
+            case '\n':
+            case '\r':
+              if (w > 0) {
+                outputFileName = new String(content, 0, w, StandardCharsets.UTF_8);
+                w = 0;
+              }
+              continue;
+            default:
+              content[w++] = c;  // copy to filename
+              continue;
+          }
+        case '\\':
+          // Peek ahead at the next character.  This is in bounds because we checked for a
+          // terminating newline above.
+          switch (content[r + 1]) {
+            // Backslashes are taken literally except when followed by whitespace.
+            // See the Windows tests for some of the nonsense we have to tolerate.
+            case ' ':
+              content[w++] = ' ';  // copy a space into the filename
+              ++r;  // skip the space in the input
+              continue;
+            case '\n':
+            case '\r':
+              // Let the newline act as a terminator.  Technically we could have an escaped newline
+              // with no adjacent space, but compilers don't seem to generate that.
+              continue;
+            default:
+              content[w++] = c;  // copy to filename
+              continue;
+          }
+        default:
+          content[w++] = c;
       }
     }
     return this;