This is the mail archive of the
libc-hacker@sourceware.cygnus.com
mailing list for the glibc project.
wordexp is broken
- To: Ulrich Drepper <drepper@cygnus.com>
- Subject: wordexp is broken
- From: Tim Waugh <tim@cyberelk.demon.co.uk>
- Date: Sun, 13 Sep 1998 15:05:11 +0100 (BST)
- cc: libc-hacker <libc-hacker@gnu.org>
Hi again,
Field-splitting was broken for command substitution. Here's a patch that
fixes it, and also splits ':abc:' into three fields not two (which is what
happened before, owing to bugs).
Tim.
*/
Sun Sep 13 14:53:50 1998 Tim Waugh <tim@cyberelk.demon.co.uk>
* posix/wordexp-test.c: Field-splitting ':abc:' with IFS=: should
yield three fields, not two. Test both parameter expansion and
command substitution for correct field-splitting behaviour.
* posix/wordexp.c (w_emptyword): New function.
(parse_param): Use it.
(exec_comm): Likewise, for consistency with the way parse_param
splits fields.
(parse_param): Fix some memory leaks.
--- /home/tim/cvs/libc/posix/wordexp-test.c Sun Sep 13 00:26:31 1998
+++ ../libc/posix/wordexp-test.c Sun Sep 13 14:53:07 1998
@@ -51,8 +51,19 @@
{ 0, "two three", "one $var", 0, 3, { "one", "two", "three", }, IFS },
{ 0, "two three", "one \"$var\"", 0, 2, { "one", "two three", }, "" },
{ 0, "two three", "one $var", 0, 2, { "one", "two three", }, "" },
- { 0, ":abc:", "$var", 0, 2, { "", "abc", }, ":" }, /* cf. bash */
+ { 0, ":abc:", "$var", 0, 3, { "", "abc", "", }, ":" },
+ { 0, NULL, "$(echo :abc:)", 0, 3, { "", "abc", "", }, ":" },
+ { 0, NULL, "$(echo :abc:\\ )", 0, 3, { "", "abc", "", }, ": " },
+ { 0, NULL, "$(echo :abc\\ )", 0, 2, { "", "abc", }, ": " },
+ { 0, ":abc:", "$(echo $var)", 0, 3, { "", "abc", "", }, ":" },
{ 0, NULL, ":abc:", 0, 1, { " abc ", }, ":" },
+ { 0, NULL, "$(echo :abc:)def", 0, 3, { "", "abc", "def", }, ":" },
+ { 0, NULL, "$(echo abc:de)f", 0, 2, { "abc", "def", }, ":" },
+ { 0, NULL, "$(echo abc:de)f:ghi", 0, 2, { "abc", "def ghi", }, ":" },
+ { 0, "abc:", "$var$(echo def:ghi)", 0, 3, { "abc", "def", "ghi", }, ":" },
+ { 0, "abc:d", "$var$(echo ef:ghi)", 0, 3, { "abc", "def", "ghi", }, ":" },
+ { 0, "def:ghi", "$(echo abc:)$var", 0, 3, { "abc", "def", "ghi", }, ":" },
+ { 0, "ef:ghi", "$(echo abc:d)$var", 0, 3, { "abc", "def", "ghi", }, ":" },
/* Simple parameter expansion */
{ 0, "foo", "${var}", 0, 1, { "foo", }, IFS },
--- /home/tim/cvs/libc/posix/wordexp.c Sun Sep 13 00:26:32 1998
+++ ../libc/posix/wordexp.c Sun Sep 13 14:55:42 1998
@@ -74,6 +74,7 @@
#define W_CHUNK (100)
+/* Result of w_newword will be ignored if it the last word. */
static inline char *
w_newword (size_t *actlen, size_t *maxlen)
{
@@ -136,6 +137,20 @@
}
+/* Result of w_emptyword will not be ignored even if it is the last. */
+static inline char *
+w_emptyword (size_t *actlen, size_t *maxlen)
+{
+ char *word = malloc (1 + W_CHUNK);
+ *maxlen = W_CHUNK;
+ *actlen = 0;
+
+ if (word)
+ *word = '\0';
+
+ return word;
+}
+
static char *
internal_function
w_addstr (char *buffer, size_t *actlen, size_t *maxlen, const char *str)
@@ -811,6 +826,7 @@
int i;
char *buffer;
pid_t pid;
+ int keep_empty_word = 0;
/* Don't fork() unless necessary */
if (!comm || !*comm)
@@ -911,6 +927,11 @@
if (strchr (ifs_white, buffer[i]) == NULL)
{
/* Current character is IFS but not whitespace */
+
+ /* After this delimiter, another field must result.
+ * Make a note. */
+ keep_empty_word = 1;
+
if (copying == 2)
{
/* current character
@@ -935,25 +956,12 @@
if (copying != 1)
continue;
- /* End of field (search for non-IFS afterwards) */
+ /* End of field (search for non-ws IFS afterwards) */
copying = 2;
}
/* First IFS white space, or IFS non-whitespace.
- * Delimit the field. */
- if (!*word)
- {
- /* This field is null, so make it an empty string */
- *word = w_addchar (*word, word_length, max_length, 0);
- if (*word == NULL)
- {
- __kill (pid, SIGKILL);
- __waitpid (pid, NULL, 0);
- __close (fildes[0]);
- return WRDE_NOSPACE;
- }
- }
-
+ * Delimit the field. Nulls are converted by w_addword. */
if (w_addword (pwordexp, *word) == WRDE_NOSPACE)
{
__kill (pid, SIGKILL);
@@ -962,13 +970,20 @@
return WRDE_NOSPACE;
}
- *word = w_newword (word_length, max_length);
+ if (keep_empty_word)
+ *word = w_emptyword (word_length, max_length);
+ else
+ *word = w_newword (word_length, max_length);
/* fall back round the loop.. */
}
else
{
/* Not IFS character */
copying = 1;
+
+ if (buffer[i] != '\n')
+ keep_empty_word = 0;
+
*word = w_addchar (*word, word_length, max_length,
buffer[i]);
if (*word == NULL)
@@ -985,7 +1000,18 @@
/* Bash chops off trailing newlines, which seems sensible. */
while (*word_length > 0 && (*word)[*word_length - 1] == '\n')
- (*word)[--*word_length] = '\0';
+ {
+ (*word)[--*word_length] = '\0';
+
+ /* If the last word was entirely newlines, and the previous word
+ * wasn't delimited with IFS non-whitespace, turn it into a new word
+ * which can be ignored if there's nothing following it. */
+ if (!keep_empty_word && *word_length == 0)
+ {
+ free (*word);
+ *word = w_newword (word_length, max_length);
+ }
+ }
__close (fildes[0]);
return 0;
@@ -1723,7 +1749,7 @@
free (value);
if (value_copy == NULL)
- return WRDE_NOSPACE;
+ goto no_space;
do
{
@@ -1736,10 +1762,10 @@
if (w_addword (pwordexp, *word) == WRDE_NOSPACE)
{
free (value_copy);
- return WRDE_NOSPACE;
+ goto no_space;
}
- *word = w_newword (word_length, max_length);
+ *word = w_emptyword (word_length, max_length);
}
/* Skip IFS whitespace before the field */
@@ -1773,7 +1799,7 @@
if (*word == NULL && *field_begin != '\0')
{
free (value_copy);
- return WRDE_NOSPACE;
+ goto no_space;
}
field_begin = next_field;
@@ -2217,7 +2243,7 @@
/* End of string */
/* There was a word separator at the end */
- if (word == NULL)
+ if (word == NULL) /* i.e. w_newword */
return 0;
/* There was no field separator at the end */