On Thu, Jul 28, 2011 at 7:46 PM, Eric Blake <ebl...@redhat.com> wrote: > On 07/28/2011 07:30 AM, Eric Blake wrote: > This topic came up again in today's Austin Group meeting, reaffirming that > Solaris behavior is correct and glibc 2.14 behavior, while fixing one aspect > of fclose, caused an an unintentional regression in another aspect when > compared to 2.13 behavior: [...] > That is, fclose() should _not_ call lseek() to the stream position unless > the stream has done any I/O or fseek() which would qualify as making the > stream an active handle, such that the stream position has a reason to be > reflected back into the file description position.
Thanks for the clarification. I see you also updated the bugzilla entry: http://sourceware.org/bugzilla/show_bug.cgi?id=12724#c2 Regarding the Ruby issue, then what they do is OK (still I think ugly historical cruft). Csaba
