Thanks for the report and analysis.
The problem is that you're using Windows, and the assumptions
that i-node and device numbers are unique (per POSIX) don't hold there.
FYI, the latest release won't fix this problem but does have
some bug fixes and other improvements:
ftp://alpha.gnu.org/gnu/fetish/
Igor Afanasyev <[EMAIL PROTECTED]> writes:
| I use the following software:
| W2K, cygwin-1.1.3-1, textutils-2.0.
|
| To spy on log files I run:
| tail --retry --follow=name /y/logs/cf.log
| where cf.log is a file in a remote file system mounted to directory /y.
|
| Periodically I rename cf.log to something like cf-current-date.log
| and when new cf.log appears tail crashes with message
| ==
| assertion "valid_file_spec (f)" failed: file "tail.c", line 700
| 0 [sig] tail 1007 stackdump: Dumping stack trace to tail.exe.stackdump
| ==
|
| I traced tail in gdb and found the following:
| when I rename cf.log to cf-yyyy-mm-dd.log and create new cf.log this
| new cf.log has the same st_ino and st_dev as old cf.log before
| renaming, so tail cannot determine that cf.log is a new file and
| the conditional expression in tail.c (line 761) results false
| ==
| else if (f->ino != new_stats.st_ino || f->dev != new_stats.st_dev)
| ==
| statements from line 787 to 798 do not execute
| ==
| if (new_file)
| {
| /* Record new file info in f. */
| f->fd = fd;
| f->size = 0; /* Start at the beginning of the file... */
| f->dev = new_stats.st_dev;
| f->ino = new_stats.st_ino;
| f->n_unchanged_stats = 0;
| f->n_consecutive_size_changes = 0;
| /* FIXME: check lseek return value */
| lseek (f->fd, f->size, SEEK_SET);
| }
| ==
| f->errnum becomes 0 and f->fd remains -1 though cf.log was opened
| successfully.
|
| In the next call of recheck assertion on line 700 prints an error
| message.
| ==
| assert (valid_file_spec (f));
| ==
|
| When cf.log places on a local file system tail works correctly.
|
| Igor
