[builtin/read] Exit code is 1 only when EOF is reached.

Addresses issue #705.

osh/builtin_misc.py

@@ -159,27 +159,28 @@ def _AppendParts(s, spans, max_results, join_next, parts):
 # in C?  Less garbage probably.
 # NOTE that dash, mksh, and zsh all read a single byte at a time.  It appears
 # to be required by POSIX?  Could try libc getline and make this an option.
+
 def ReadLineFromStdin(delim_char):
   # type: (Optional[str]) -> Tuple[str, bool]
   """Read a portion of stdin.
   
   If delim_char is set, read until that delimiter, but don't include it.
   If not set, read a line, and include the newline.
   """
-  found_delim = False
+  eof = False
   chars = []
   while True:
     c = posix.read(0, 1)
     if not c:
+      eof = True
       break
 
     if c == delim_char:
-      found_delim = True
       break
 
     chars.append(c)
 
-  return ''.join(chars), found_delim
+  return ''.join(chars), eof
 
 
 class Read(object):
@@ -249,19 +250,19 @@ def Run(self, cmd_val):
     # it's not -r).
     parts = []
     join_next = False
+    status = 0
     while True:
-      line, found_delim = ReadLineFromStdin(delim_char)
-      #log('LINE %r', line)
-      if len(line) == 0:  # EOF
-        status = 1
-        break
+      line, eof = ReadLineFromStdin(delim_char)
 
-      status = 0
-      if not found_delim:
-        # odd bash behavior: fail if no newline, even though we're setting
-        # variables.
+      if eof:
+        # status 1 to terminate loop.  (This is true even though we set
+        # variables).
         status = 1
 
+      #log('LINE %r', line)
+      if len(line) == 0:
+        break
+
       spans = self.splitter.SplitForRead(line, not arg.r)
       done, join_next = _AppendParts(line, spans, max_results, join_next, parts)
 

osh/word_parse_test.py

@@ -518,7 +518,8 @@ def testMultiLine(self):
     test_lib.AssertAsdlEqual(self, t, w)
 
   def testUnicode(self):
-    words = u'\u007a \u03bb \u4e09 \U0001f618'.encode('utf-8')
+    words = 'z \xce\xbb \xe4\xb8\x89 \xf0\x9f\x98\x98'
+
     w_parser = test_lib.InitWordParser(words)
     w = w_parser.ReadWord(lex_mode_e.ShCommand)
     self.assertEqual('z', w.parts[0].val)

qsn_/qsn.py

@@ -25,8 +25,6 @@
   implementation could relax this.
 
 TODO:
-
-  - Optimize common case encoding with native regex.  IsPlainWord()
   - maybe_decode() in addition to decode()
 
 TODO for other implementations:

spec/builtin-io.test.sh

@@ -216,9 +216,32 @@ echo "[$x]"
 echo -n '' > $TMP/empty.txt
 read x < $TMP/empty.txt
 argv.py "status=$?" "$x"
-## stdout: ['status=1', '']
+
+# No variable name, behaves the same
+read < $TMP/empty.txt
+argv.py "status=$?" "$REPLY"
+
+## STDOUT:
+['status=1', '']
+['status=1', '']
+## END
+## OK dash STDOUT:
+['status=1', '']
+['status=2', '']
+## END
 ## status: 0
 
+
+#### read with zero args
+echo | read
+echo status=$?
+## STDOUT:
+status=0
+## END
+## BUG dash STDOUT:
+status=2
+## END
+
 #### Read builtin with no newline.
 # This is odd because the variable is populated successfully.  OSH/Oil might
 # need a separate put reading feature that doesn't use IFS.