[KinoSearch] test failure in svn trunk t/core/051-fs_file_handle.t

[Marvin Humphrey]

On Fri, Jan 15, 2010 at 11:18:32PM -0600, Peter Karman wrote:
> The test label suggests that FSFH_Read() should be returning false, ostensibly 
> before it gets to the memcpy() that causes the segfault. 

That's right.  The API for FileHandle's Read() is that it sets Err_error and
returns false when something goes wrong.  Such as attempting to read from a
write-only handle.

> I looked at the other FSFH_Read() functions defined (this is calling the
> 64bit version) and they are all very different. 

Yup.  I screwed up the 64-bit version.  (It was coded blind.)  At least the
test caught it, though.  :\  Thanks for troubleshooting.

> I am guessing the write-only flag is set by the underlying 
> open() call on the filehandle and that KS isn't actually tracking that, since I 
> see no check for it in the KS code.

We should check for the read-only flag up front in the 64-bit branch.

  if (!(self->flags & FH_READ_ONLY)) {
      /* ... */
  }

For the 32-bit branches, we get to rely on a system call failing, but for
read-only 64-bit FSFileHandle objects we always mmap the whole file up front
-- so there's no system call to fail during Read().

Marvin Humphrey