Hi Eric,
lseek on BeOS has another deficiency: In some cases it can modify the
file offset although the same call fails. POSIX says this should not happen.
But I don't think it's worth working around it in lib/lseek.c. Here is
instead a proposed workaround for the unit test:
2007-08-18 Bruno Haible <[EMAIL PROTECTED]>
* tests/test-lseek.c (main): Disable a test for BeOS.
* doc/functions/lseek.texi: Document the BeOS bug.
*** tests/test-lseek.c 2 Jun 2007 00:42:13 -0000 1.4
--- tests/test-lseek.c 18 Aug 2007 22:56:22 -0000
***************
*** 49,55 ****
--- 49,59 ----
ASSERT (lseek (0, (off_t)-4, SEEK_CUR) == -1);
ASSERT (errno == EINVAL);
errno = 0;
+ #if ! defined __BEOS__
+ /* POSIX says that the last lseek call, when failing, does not change
+ the current offset. But BeOS sets it to 0. */
ASSERT (lseek (0, (off_t)0, SEEK_CUR) == 2);
+ #endif
#if 0 /* leads to SIGSYS on IRIX 6.5 */
ASSERT (lseek (0, (off_t)0, (SEEK_SET | SEEK_CUR | SEEK_END) + 1) ==
-1);
ASSERT (errno == EINVAL);
***************
*** 59,65 ****
--- 63,73 ----
ASSERT (lseek (1, (off_t)-4, SEEK_CUR) == -1);
ASSERT (errno == EINVAL);
errno = 0;
+ #if ! defined __BEOS__
+ /* POSIX says that the last lseek call, when failing, does not change
+ the current offset. But BeOS sets it to 0. */
ASSERT (lseek (1, (off_t)0, SEEK_CUR) == 2);
+ #endif
#if 0 /* leads to SIGSYS on IRIX 6.5 */
ASSERT (lseek (1, (off_t)0, (SEEK_SET | SEEK_CUR | SEEK_END) + 1) ==
-1);
ASSERT (errno == EINVAL);
*** doc/functions/lseek.texi 2 Jun 2007 00:42:13 -0000 1.4
--- doc/functions/lseek.texi 18 Aug 2007 22:56:22 -0000
***************
*** 27,30 ****
--- 27,35 ----
@code{errno} to @code{EINVAL} and return -1, but in this situation a
@code{SIGSYS} signal is raised on some platforms:
IRIX 6.5.
+ @item
+ When the @code{lseek} function fails, POSIX says that the file offset remains
+ unchanged. But on some platforms, attempting to set a negative file offset
+ fails and sets the file offset to 0:
+ BeOS.
@end itemize
