commit bash-20191009 snapshot

CWRU/CWRU.chlog

@@ -6726,3 +6726,59 @@ pathexp.[ch],{bashline,subst}.c
 subst.c
 	- glob_expand_word_list: call shell_glob_filename with QGLOB_CTLESC
 	  because quote removal hasn't been performed yet
+
+				   10/7
+				   ----
+pathexp.c
+	- quote_string_for_globbing: if we have an unquoted backslash followed
+	  by a CTLESC-quoted character (not CTLESC-CTLESC), just perform the
+	  usual CTLESC-to-backslash conversion instead of skipping over it.
+	  Fixes issue raised in austin-group discussion about globbing by
+	  Geoff Clare <gwc@opengroup.org> (austin-group issue 1234), though
+	  it's still inherently ambiguous
+
+				   10/8
+				   ----
+include/shmbutil.h
+	- xwcsrtombs: extern declaration, to match other functions in that file
+
+lib/glob/glob.c
+	- wcdequote_pathname: new function, actual backslash quote removal code
+	  from wdequote_pathname; wdequote_pathname calls it
+
+lib/glob/glob.c
+	- {udequote,wcdequote}_pathname: now public void functions
+
+lib/glob/smatch.c
+	- DEQUOTE_PATHNAME: defined appropriately to udequote_pathname or
+	  wcdequote_pathname
+
+lib/glob/sm_loop.c
+	- DEQUOTE_PATHNAME: appropriate extern declaration
+	- BRACKMATCH: call DEQUOTE_PATHNAME to dequote a character class name
+	  in a bracket expression. This is the result of a discussion on the
+	  austin-group mailing list, from Geoff Clare <gwc@opengroup.org> and
+	  Robert Elz <kre@bmunnari.oz.au>
+
+				   10/10
+				   -----
+execute_cmd.[ch]
+	- async_redirect_stdin: now a global function
+
+subst.c
+	- process_substitute: call async_redirect_stdin in the child to keep
+	  it from having stdin connected to the terminal, since it's not a
+	  job control process
+
+				   10/11
+				   -----
+subst.c
+	- process_substitute: in the child process, set interactive = 0, since
+	  an asynchronous process substitution process is not interactive.
+	  Seems to fix issue reported by Grisha Levit <grishalevit@gmail.com>
+
+lib/sh/shmatch.c
+	- sh_regmatch: implement a suggestion from Grisha Levit
+	  <grishalevit@gmail.com> and don't allow nocaseglob to enable case-
+	  insensitive regexp matching. It hasn't been documented that way
+	  in years

execute_cmd.c

@@ -126,8 +126,6 @@ static void cleanup_redirects PARAMS((REDIRECT *));
 static int restore_signal_mask PARAMS((sigset_t *));
 #endif
 
-static void async_redirect_stdin PARAMS((void));
-
 static int builtin_status PARAMS((int));
 
 static int execute_for_command PARAMS((FOR_COM *));
@@ -521,7 +519,7 @@ open_files ()
 }
 #endif
 
-static void
+void
 async_redirect_stdin ()
 {
   int fd;

execute_cmd.h

@@ -77,6 +77,7 @@ extern int execute_command __P((COMMAND *));
 extern int execute_command_internal __P((COMMAND *, int, int, int, struct fd_bitmap *));
 extern int shell_execve __P((char *, char **, char **));
 extern void setup_async_signals __P((void));
+extern void async_redirect_stdin __P((void));
 
 extern void undo_partial_redirects __P((void));
 extern void dispose_partial_redirects __P((void));

include/shmbutil.h

@@ -1,6 +1,6 @@
 /* shmbutil.h -- utility functions for multibyte characters. */
 
-/* Copyright (C) 2002-2004 Free Software Foundation, Inc.
+/* Copyright (C) 2002-2019 Free Software Foundation, Inc.
 
    This file is part of GNU Bash, the Bourne Again SHell.
 
@@ -29,12 +29,13 @@
 #if defined (HANDLE_MULTIBYTE)
 #include "shmbchar.h"
 
-extern size_t xmbsrtowcs __P((wchar_t *, const char **, size_t, mbstate_t *));
-extern size_t xdupmbstowcs __P((wchar_t **, char ***, const char *));
+extern size_t xwcsrtombs PARAMS((char *, const wchar_t **, size_t, mbstate_t *));
+extern size_t xmbsrtowcs PARAMS((wchar_t *, const char **, size_t, mbstate_t *));
+extern size_t xdupmbstowcs PARAMS((wchar_t **, char ***, const char *));
 
-extern size_t mbstrlen __P((const char *));
+extern size_t mbstrlen PARAMS((const char *));
 
-extern char *xstrchr __P((const char *, int));
+extern char *xstrchr PARAMS((const char *, int));
 
 extern int locale_mb_cur_max;	/* XXX */
 extern int locale_utf8locale;	/* XXX */

lib/glob/glob.c

@@ -115,8 +115,9 @@ static int skipname __P((char *, char *, int));
 #if HANDLE_MULTIBYTE
 static int mbskipname __P((char *, char *, int));
 #endif
+void udequote_pathname __P((char *));
 #if HANDLE_MULTIBYTE
-static void udequote_pathname __P((char *));
+void wcdequote_pathname __P((wchar_t *));
 static void wdequote_pathname __P((char *));
 #else
 #  define dequote_pathname udequote_pathname
@@ -409,7 +410,7 @@ mbskipname (pat, dname, flags)
 #endif /* HANDLE_MULTIBYTE */
 
 /* Remove backslashes quoting characters in PATHNAME by modifying PATHNAME. */
-static void
+void
 udequote_pathname (pathname)
      char *pathname;
 {
@@ -431,6 +432,26 @@ udequote_pathname (pathname)
 
 #if HANDLE_MULTIBYTE
 /* Remove backslashes quoting characters in PATHNAME by modifying PATHNAME. */
+void
+wcdequote_pathname (wpathname)
+     wchar_t *wpathname;
+{
+  int i, j;
+
+  for (i = j = 0; wpathname && wpathname[i]; )
+    {
+      if (wpathname[i] == L'\\')
+	i++;
+
+      wpathname[j++] = wpathname[i++];
+
+      if (wpathname[i - 1] == L'\0')
+	break;
+    }
+  if (wpathname)
+    wpathname[j] = L'\0';
+}
+
 static void
 wdequote_pathname (pathname)
      char *pathname;
@@ -458,18 +479,7 @@ wdequote_pathname (pathname)
     }
   orig_wpathname = wpathname;
 
-  for (i = j = 0; wpathname && wpathname[i]; )
-    {
-      if (wpathname[i] == L'\\')
-	i++;
-
-      wpathname[j++] = wpathname[i++];
-
-      if (wpathname[i - 1] == L'\0')
-	break;
-    }
-  if (wpathname)
-    wpathname[j] = L'\0';
+  wcdequote_pathname (wpathname);
 
   /* Convert the wide character string into unibyte character set. */
   memset (&ps, '\0', sizeof(mbstate_t));
@@ -686,7 +696,7 @@ glob_vector (pat, dir, flags)
      a filename `DIR/PAT'.  If there is, and we can access it, just make the
      vector to return and bail immediately. */
   hasglob = 0;
-  if (skip == 0 && (hasglob = glob_pattern_p (pat)) == 0 || hasglob == 2)
+  if (skip == 0 && ((hasglob = glob_pattern_p (pat)) == 0 || hasglob == 2))
     {
       int dirlen;
       struct stat finfo;

lib/glob/sm_loop.c

@@ -29,6 +29,8 @@ static CHAR *PARSE_COLLSYM __P((CHAR *, INT *));
 static CHAR *BRACKMATCH __P((CHAR *, U_CHAR, int));
 static int EXTMATCH __P((INT, CHAR *, CHAR *, CHAR *, CHAR *, int));
 
+extern void DEQUOTE_PATHNAME __P((CHAR *));
+
 /*static*/ CHAR *PATSCAN __P((CHAR *, CHAR *, INT));
 
 int
@@ -456,6 +458,9 @@ BRACKMATCH (p, test, flags)
 		{
 		  bcopy (p + 1, ccname, (close - p - 1) * sizeof (CHAR));
 		  *(ccname + (close - p - 1)) = L('\0');
+		  /* As a result of a POSIX discussion, char class names are
+		     allowed to be quoted (?) */
+		  DEQUOTE_PATHNAME (ccname);
 		  pc = IS_CCLASS (orig_test, (XCHAR *)ccname);
 		}
 	      if (pc == -1)
@@ -929,6 +934,7 @@ fprintf(stderr, "extmatch: flags = %d\n", flags);
 #undef PATSCAN
 #undef STRCOMPARE
 #undef EXTMATCH
+#undef DEQUOTE_PATHNAME
 #undef STRUCT
 #undef BRACKMATCH
 #undef STRCHR

lib/glob/smatch.c

@@ -260,6 +260,7 @@ is_cclass (c, name)
 #define PATSCAN			glob_patscan
 #define STRCOMPARE		strcompare
 #define EXTMATCH		extmatch
+#define DEQUOTE_PATHNAME	udequote_pathname
 #define STRUCT			smat_struct
 #define STRCHR(S, C)		strchr((S), (C))
 #define MEMCHR(S, C, N)		memchr((S), (C), (N))
@@ -500,6 +501,7 @@ posix_cclass_only (pattern)
 #define PATSCAN			glob_patscan_wc
 #define STRCOMPARE		wscompare
 #define EXTMATCH		extmatch_wc
+#define DEQUOTE_PATHNAME	wcdequote_pathname
 #define STRUCT			wcsmat_struct
 #define STRCHR(S, C)		wcschr((S), (C))
 #define MEMCHR(S, C, N)		wmemchr((S), (C), (N))

lib/sh/shmatch.c

@@ -64,7 +64,7 @@ sh_regmatch (string, pattern, flags)
 #endif
 
   rflags = REG_EXTENDED;
-  if (glob_ignore_case || match_ignore_case)
+  if (match_ignore_case)
     rflags |= REG_ICASE;
 #if !defined (ARRAY_VARS)
   rflags |= REG_NOSUB;

pathexp.c

@@ -243,6 +243,7 @@ quote_string_for_globbing (pathname, qflags)
 	}
       else if (pathname[i] == CTLESC)
 	{
+convert_to_backslash:
 	  if ((qflags & QGLOB_FILENAME) && pathname[i+1] == '/')
 	    continue;
 	  /* What to do if preceding char is backslash? */
@@ -358,6 +359,16 @@ quote_string_for_globbing (pathname, qflags)
 	     even when the first CTLESC is preceded by a backslash. */
 	  if ((qflags & QGLOB_CTLESC) && pathname[i] == CTLESC && (pathname[i+1] == CTLESC || pathname[i+1] == CTLNUL))
 	    i++;	/* skip over the CTLESC */
+	  else if ((qflags & QGLOB_CTLESC) && pathname[i] == CTLESC)
+	    /* A little more general: if there is an unquoted backslash in the
+	       pattern and we are handling quoted characters in the pattern,
+	       convert the CTLESC to backslash and add the next character on
+	       the theory that the backslash will quote the next character
+	       but it would be inconsistent not to replace the CTLESC with
+	       another backslash here. We can't tell at this point whether the
+	       CTLESC comes from a backslash or other form of quoting in the
+	       original pattern. */
+	    goto convert_to_backslash;
 	}
       else if (pathname[i] == '\\' && (qflags & QGLOB_REGEXP))
         last_was_backslash = 1;

subst.c

@@ -5898,12 +5898,15 @@ process_substitute (string, open_for_read_in_child)
   pid = make_child ((char *)NULL, 1);
   if (pid == 0)
     {
+interactive = 0;
       reset_terminating_signals ();	/* XXX */
       free_pushed_string_input ();
       /* Cancel traps, in trap.c. */
       restore_original_signals ();	/* XXX - what about special builtins? bash-4.2 */
       QUIT;	/* catch any interrupts we got post-fork */
       setup_async_signals ();
+      if (open_for_read_in_child == 0)
+	async_redirect_stdin ();
       subshell_environment |= SUBSHELL_COMSUB|SUBSHELL_PROCSUB;
 
       /* We don't inherit the verbose option for command substitutions now, so

tests/glob.right

@@ -20,16 +20,22 @@ ok 1
 [a
 [[:alpha:]
 ok 2
+ok 2.1
 ok 3
+ok 4
 == LANG=en_US.UTF-8 ==
 [[:alpha:]
 ok 1
 [a
 [[:alpha:]
 ok 2
+ok 2.1
 ok 3
+ok 4
 invalid character class
 == LANG=C ==
+p
+p
 ok 1
 ok 2
 ok 3
@@ -37,6 +43,8 @@ ok 4
 ok 5
 ok 6
 == LANG=en_US.UTF-8 ==
+p
+p
 ok 1
 ok 2
 ok 3

tests/glob3.sub

@@ -12,7 +12,13 @@
 #   along with this program.  If not, see <http://www.gnu.org/licenses/>.
 #
 : ${TMPDIR:=/var/tmp}
-cd $TMPDIR
+
+TESTDIR=${TMPDIR}/glob-test-$$
+mkdir ${TESTDIR}
+cd $TESTDIR ||  {
+        echo "$TESTDIR: cannot cd" >&2
+        exit 1
+}
 
 matchfunc()
 {
@@ -38,10 +44,21 @@ matchfunc()
 	*)		echo bad 2;;
 	esac
 
+	case x in
+	[[:aeioux:])	echo bad 2.1 ;;
+	*)		echo ok 2.1 ;;
+	esac
+
 	case [x in
 	[[:alpha:])	echo bad 3;;
 	*)		echo ok 3;;
 	esac
+
+	# unclosed bracket char class expression just matches against ":alpha"
+	case a in
+	[[:alpha])	echo ok 4;;
+	*)		echo bad 4;;
+	esac
 }
 
 echo invalid bracket expression
@@ -57,6 +74,12 @@ matchfunc()
 {
 	echo == LANG=$LANG ==
 
+	touch p
+	# quoted character classes work as if they were unquoted now
+	echo [[:alpha:]]
+	echo [[:"alpha":]]
+	rm -f p
+
 	case a] in
 	[[:aleph:]])	echo bad 1;;
 	*)		echo ok 1;;
@@ -72,9 +95,10 @@ matchfunc()
 	*)		echo ok 3;;
 	esac
 
-	case a in
-	[[:"alpha":]])	echo bad 4;;
-	*)		echo ok 4;;
+	# Posix says quoted character class names work now
+	case x in
+	[[:"alpha":]])	echo ok 4;;
+	*)		echo bad 4;;
 	esac
 
 	case a in
@@ -134,3 +158,6 @@ matchfunc
 
 export LANG=en_US.UTF-8
 matchfunc
+
+cd $OLDPWD
+rm -rf $TESTDIR