Implement the getopts builtin, with tests.

Uncovered a fundamental design flaw with the global variable OPTIND. Didn't implement some stuff like extra args and OPTERR. Alpine doesn't seem to use these. Fixes issue #39.
oilshell · Sep 26, 2017 · a36fe9a45f404db7f7ba65d378987842557d52d6 · a36fe9a
1 parent fc3fc85
commit a36fe9a45f404db7f7ba65d378987842557d52d6
Unified Split

Showing with 293 additions and 2 deletions.

+1 −0 core/args.py

+106 −1 core/builtin.py

+3 −0 core/cmd_exec.py

+2 −0 core/state.py

+180 −0 spec/builtin-getopts.test.sh

+1 −1 test/spec.sh
diff --git a/core/args.py b/core/args.py
@@ -472,3 +472,4 @@ def Parse(self, argv):
        break

    return out, state.i
+
diff --git a/core/builtin.py b/core/builtin.py
@@ -42,6 +42,9 @@
 scope = runtime.scope
 var_flags = runtime.var_flags
 log = util.log
+e_die = util.e_die
+
+

 # NOTE: NONE is a special value.
 # TODO:
@@ -57,7 +60,7 @@
 COMPLETE COMPGEN DEBUG_LINE
 TRUE FALSE
 COLON
-TEST BRACKET
+TEST BRACKET GETOPTS
 TYPE HELP
 """.split())

@@ -197,6 +200,9 @@ def Resolve(argv0):
  elif argv0 == "[":
    return EBuiltin.BRACKET

+  elif argv0 == "getopts":
+    return EBuiltin.GETOPTS
+
  elif argv0 == "type":
    return EBuiltin.TYPE

@@ -746,6 +752,105 @@ def Umask(argv):
  raise args.UsageError('umask: unexpected arguments')


+def _ParseOptSpec(spec_str):
+  spec = {}
+  i = 0
+  n = len(spec_str)
+  while True:
+    if i >= n:
+      break
+    c = spec_str[i]
+    key = '-' + c
+    spec[key] = False
+    i += 1
+    if i >= n:
+      break
+    # If the next character is :, change the value to True.
+    if spec_str[i] == ':':
+      spec[key] = True
+      i += 1
+  return spec
+
+
+def _GetOpts(spec, mem, optind):
+  optarg = ''  # not set by default
+
+  v2 = mem.GetArgNum(optind)
+  if v2.tag == value_e.Undef:  # No more arguments.
+    return 1, '?', optarg, optind
+  assert v2.tag == value_e.Str
+
+  current = v2.s
+
+  if not current.startswith('-'):  # The next arg doesn't look like a flag.
+    return 1, '?', optarg, optind
+
+  # It looks like an argument.  Stop iteration by returning 1.
+  if current not in spec:  # Invalid flag
+    optind += 1
+    return 0, '?', optarg, optind
+
+  optind += 1
+  opt_char = current[-1]
+
+  needs_arg = spec[current]
+  if needs_arg:
+    v3 = mem.GetArgNum(optind)
+    if v3.tag == value_e.Undef:
+      util.error('getopts: option %r requires an argument', current)
+      # Hm doesn't cause status 1?
+      return 0, '?', optarg, optind
+    assert v3.tag == value_e.Str
+
+    optarg = v3.s
+    optind += 1
+
+  return 0, opt_char, optarg, optind
+
+
+# spec string -> {flag, arity}
+_GETOPTS_CACHE = {}
+
+def GetOpts(argv, mem):
+  """
+  Vars to set:
+    OPTIND - initialized to 1 at startup
+    OPTARG - argument
+  Vars used:
+    OPTERR: disable printing of error messages
+  """
+  try:
+    # NOTE: If first char is a colon, error reporting is different.  Alpine
+    # might not use that?
+    spec_str = argv[0]
+    var_name = argv[1]
+  except IndexError:
+    raise args.UsageError('getopts optstring name [arg]')
+
+  try:
+    spec = _GETOPTS_CACHE[spec_str]
+  except KeyError:
+    spec = _ParseOptSpec(spec_str)
+    _GETOPTS_CACHE[spec_str] = spec
+
+  # These errors are fatal errors, not like the builtin exiting with code 1.
+  # Because the invariants of the shell have been violated!
+  v = mem.GetVar('OPTIND')
+  if v.tag != value_e.Str:
+    e_die('OPTIND should be a string, got %r', v)
+  try:
+    optind = int(v.s)
+  except ValueError:
+    e_die("OPTIND doesn't look like an integer, got %r", v.s)
+
+  status, opt_char, optarg, optind = _GetOpts(spec, mem, optind)
+
+  state.SetGlobalString(mem, var_name, opt_char)
+  state.SetGlobalString(mem, 'OPTARG', optarg)
+  state.SetGlobalString(mem, 'OPTIND', str(optind))
+  return status
+
+
 def Help(argv, loader):
  # TODO: Need $VERSION inside all pages?
  try:
diff --git a/core/cmd_exec.py b/core/cmd_exec.py
@@ -297,6 +297,9 @@ def _RunBuiltin(self, builtin_id, argv):
    elif builtin_id == EBuiltin.BRACKET:
      status = test_builtin.Test(argv, True)  # need_right_bracket

+    elif builtin_id == EBuiltin.GETOPTS:
+      status = builtin.GetOpts(argv, self.mem)
+
    elif builtin_id == EBuiltin.TYPE:
      path = self.mem.GetVar('PATH')
      status = builtin.Type(argv, self.funcs, path)
diff --git a/core/state.py b/core/state.py
@@ -245,6 +245,8 @@ def _InitDefaults(self):
    # NOTE: Should we put these in a namespace for Oil?
    SetGlobalString(self, 'UID', str(os.getuid()))
    SetGlobalString(self, 'EUID', str(os.geteuid()))
+    # For getopts builtin
+    SetGlobalString(self, 'OPTIND', '1')

  def _InitEnviron(self, environ):
    # This is the way dash and bash work -- at startup, they turn everything in
diff --git a/spec/builtin-getopts.test.sh b/spec/builtin-getopts.test.sh
@@ -4,6 +4,36 @@
 # 
 # NOTE: Aliases don't work in batch mode!  Interactive only.

+### getopts empty
+set -- 
+getopts 'a:' opt
+echo "status=$? opt=$opt OPTARG=$OPTARG"
+# stdout: status=1 opt=? OPTARG=
+
+### getopts sees unknown arg
+set -- -Z
+getopts 'a:' opt
+echo "status=$? opt=$opt OPTARG=$OPTARG"
+# stdout: status=0 opt=? OPTARG=
+
+### getopts three invocations
+set -- -h -c foo
+getopts 'hc:' opt
+echo status=$? opt=$opt
+getopts 'hc:' opt
+echo status=$? opt=$opt
+getopts 'hc:' opt
+echo status=$? opt=$opt
+# stdout-json: "status=0 opt=h\nstatus=0 opt=c\nstatus=1 opt=?\n"
+
+### getopts resets OPTARG
+set -- -c foo -h
+getopts 'hc:' opt
+echo status=$? opt=$opt OPTARG=$OPTARG
+getopts 'hc:' opt
+echo status=$? opt=$opt OPTARG=$OPTARG
+# stdout-json: "status=0 opt=c OPTARG=foo\nstatus=0 opt=h OPTARG=\n"
+
 ### Basic getopts invocation
 set -- -h -c foo x y z
 FLAG_h=0
@@ -17,3 +47,153 @@ done
 shift $(( OPTIND - 1 ))
 echo h=$FLAG_h c=$FLAG_c optind=$OPTIND argv=$@
 # stdout: h=1 c=foo optind=4 argv=x y z
+
+### getopts with invalid flag
+set -- -h -x
+while getopts "hc:" opt; do
+  case $opt in
+    h) FLAG_h=1 ;;
+    c) FLAG_c="$OPTARG" ;;
+    '?') echo ERROR $OPTIND; exit 2; ;;
+  esac
+done
+echo status=$?
+# stdout: ERROR 3
+# status: 2
+
+### getopts missing required argument
+set -- -h -c
+while getopts "hc:" opt; do
+  case $opt in
+    h) FLAG_h=1 ;;
+    c) FLAG_c="$OPTARG" ;;
+    '?') echo ERROR $OPTIND; exit 2; ;;
+  esac
+done
+echo status=$?
+# stdout: ERROR 3
+# status: 2
+
+### getopts doesn't look for flags after args
+set -- x -h -c y
+FLAG_h=0
+FLAG_c=''
+while getopts "hc:" opt; do
+  case $opt in
+    h) FLAG_h=1 ;;
+    c) FLAG_c="$OPTARG" ;;
+  esac
+done
+shift $(( OPTIND - 1 ))
+echo h=$FLAG_h c=$FLAG_c optind=$OPTIND argv=$@
+# stdout: h=0 c= optind=1 argv=x -h -c y
+
+### getopts with explicit args
+# NOTE: Alpine doesn't appear to use this
+FLAG_h=0
+FLAG_c=''
+arg=''
+while getopts "hc:" opt -h -c foo x y z; do
+  case $opt in
+    h) FLAG_h=1 ;;
+    c) FLAG_c="$OPTARG" ;;
+  esac
+done
+echo h=$FLAG_h c=$FLAG_c optind=$OPTIND argv=$@
+# stdout: h=1 c=foo optind=4 argv=
+
+### OPTIND
+echo $OPTIND
+# stdout: 1
+
+### OPTIND after multiple getopts with same spec
+while getopts "hc:" opt; do
+  echo '-'
+done
+echo $OPTIND
+
+set -- -h -c foo x y z
+while getopts "hc:" opt; do
+  echo '-'
+done
+echo $OPTIND
+
+set --
+while getopts "hc:" opt; do
+  echo '-'
+done
+echo $OPTIND
+# stdout-json: "1\n-\n-\n4\n1\n"
+# BUG mksh/osh stdout-json: "1\n-\n-\n4\n4\n"
+
+### OPTIND after multiple getopts with different spec
+# Wow this is poorly specified!  A fundamental design problem with the global
+# variable OPTIND.
+set -- -a
+while getopts "ab:" opt; do
+  echo '.'
+done
+echo $OPTIND
+
+set -- -c -d -e foo
+while getopts "cde:" opt; do
+  echo '-'
+done
+echo $OPTIND
+
+set -- -f
+while getopts "f:" opt; do
+  echo '_'
+done
+echo $OPTIND
+# stdout-json: ".\n2\n-\n-\n5\n2\n"
+# BUG ash/dash stdout-json: ".\n2\n-\n-\n-\n5\n_\n2\n"
+# BUG mksh/osh stdout-json: ".\n2\n-\n-\n5\n5\n"
+
+### OPTIND narrowed down
+FLAG_a=
+FLAG_b=
+FLAG_c=
+FLAG_d=
+FLAG_e=
+set -- -a
+while getopts "ab:" opt; do
+  case $opt in
+    a) FLAG_a=1 ;;
+    b) FLAG_b="$OPTARG" ;;
+  esac
+done
+# Bash doesn't reset optind!  It skips over c!  mksh at least warns about this!
+# You have to reset OPTIND yourself.
+
+set -- -c -d -e E
+while getopts "cde:" opt; do
+  case $opt in
+    c) FLAG_c=1 ;;
+    d) FLAG_d=1 ;;
+    e) FLAG_e="$OPTARG" ;;
+  esac
+done
+
+echo a=$FLAG_a b=$FLAG_b c=$FLAG_c d=$FLAG_d e=$FLAG_e
+# stdout: a=1 b= c=1 d=1 e=E
+# BUG bash/mksh/osh stdout: a=1 b= c= d=1 e=E
+
+
+### Getopts parses the function's arguments
+# NOTE: GLOBALS are set, not locals!  Bad interface.
+FLAG_h=0
+FLAG_c=''
+myfunc() {
+  while getopts "hc:" opt; do
+    case $opt in
+      h) FLAG_h=1 ;;
+      c) FLAG_c="$OPTARG" ;;
+    esac
+  done
+}
+set -- -h -c foo x y z
+myfunc -c bar
+echo h=$FLAG_h c=$FLAG_c opt=$opt optind=$OPTIND argv=$@
+# stdout: h=0 c=bar opt=? optind=3 argv=-h -c foo x y z
+
diff --git a/test/spec.sh b/test/spec.sh
@@ -259,7 +259,7 @@ builtin-vars() {

 builtin-getopts() {
  sh-spec spec/builtin-getopts.test.sh --osh-failures-allowed 1 \
-    ${REF_SHELLS[@]} $OSH "$@"
+    ${REF_SHELLS[@]} $BUSYBOX_ASH $OSH "$@"
 }

 builtin-test() {