[Bug testsuite/45841] [4.6 Regression]: r164529 cris-elf libstdc++ 27_io/basic_filebuf/seekoff/char/2-io.cc

["potswa at mac dot com" <gcc-bugzilla at gcc dot gnu dot org>]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45841

--- Comment #16 from David Krauss <potswa at mac dot com> 2010-10-05 00:24:35 UTC ---
(In reply to comment #15)
> r164528:
> ...
> read(4, "1", 1)                         = 1
> lseek(4, 4294967295, SEEK_CUR)          = 4294967303
> write(4, "x", 1)                        = 1
> write(4, "\n", 1)                       = 1
> lseek(4, 0, SEEK_CUR)                   = 4294967305
> read(4, "", 1)                          = 0
> read(4, "", 1)                          = 0
> lseek(4, 0, SEEK_END)                   = 4294967305
> ...
> 
> r164529:
> ...
> read(4, "1", 1)                         = 1
> lseek(4, 0, SEEK_CUR)                   = 8
> lseek(4, 4294967295, SEEK_CUR)          = 4294967303
> write(4, "x", 1)                        = 1
> write(4, "\n", 1)                       = 1
> lseek(4, 0, SEEK_CUR)                   = 4294967305
> read(4, "", 1)                          = 0
> read(4, "", 1)                          = 0
> write(2, "assertion \"", 11assertion ")            = 11

The failure had nothing to do with the additional seek. The excerpts above show
that the added operation didn't move the file pointerâ the failure occurred at
the same position as success before the patch.

What the testcase ends up doing is extending the file and then reading past
EOF. That's not illegal behavior and the case as written should still pass
after comparing EOF == EOF, just as it did before my patch.

Hmm, yep, caught a bug in my patch! Thanks!