This is a regression from 2.15. Test case below, does not assert with glibc-2.15, but asserts with current trunk. If I use fopen() instead of fdopen() ["./a.out -o"] or if I fflush() after fwrite() ["./a.out -f"], then assert goes away. /* --- cut --- */ #include <assert.h> #include <fcntl.h> #include <stdio.h> #include <unistd.h> int main(int argc, char *argv[]) { FILE *fp; size_t written; off_t off; int do_flush = 0, do_fopen = 0; int opt; const char *const fname = "/tmp/output"; while ((opt = getopt (argc, argv, "fo")) != -1) { switch (opt) { case 'f': do_flush = 1; break; case 'o': do_fopen = 1; break; } } fp = fopen (fname, "w"); written = fwrite ("abcabc", 1, 6, fp); assert (written == 6); fclose (fp); if (do_fopen) fp = fopen (fname, "a"); else { int fd = open (fname, O_WRONLY, 0); assert (fd != -1); fp = fdopen (fd, "a"); } assert (fp != NULL); written = fwrite ("ghi", 1, 3, fp); assert (written == 3); if (do_flush) fflush (NULL); off = ftello (fp); assert (off == 9); return 0; } /* --- cut --- */
This broke with commit adb26faefe47b7d34c941cbfc193ca7a5fde8e3f: 2012-09-28 Siddhesh Poyarekar <siddhesh@redhat.com> [BZ #5298] * libio/fileops.c (_IO_new_file_seekoff): Don't flush buffer for ftell. Compute offsets from write pointers instead. * libio/wfileops.c (_IO_wfile_seekoff): Likewise.
Thanks, I'll look into it.
*** Bug 16605 has been marked as a duplicate of this bug. ***
Fixed in master: commit 000232b9bcbf194f1e5fd0ff380000f341505405 Author: Siddhesh Poyarekar <siddhesh@redhat.com> Date: Tue Mar 4 07:45:58 2014 +0530 Separate ftell from fseek logic and avoid modifying FILE data (#16532) ftell semantics are distinct from fseek(SEEK_CUR) especially when it is called on a file handler that is not yet active. Due to this caveat, much care needs to be taken while modifying the handler data and hence, this first iteration on separating out ftell focusses on maintaining handler data integrity at all times while it figures out the current stream offset. The result is that it makes a syscall for every offset request. There is scope for optimizing this by caching offsets when we know that the handler is active. A simple way to find out is when the buffers have data. It is not so simple to find this out when the buffer is empty without adding some kind of flag.
I am afraid the provided fix has caused a memory leak. See https://sourceware.org/bugzilla/show_bug.cgi?id=17370. The malloc in the patch has not been freed.