[PATCH 2 of 2 STABLE] store: add auto detection for hardlink count blindness (issue1866)

[Matt Mackall]

On Sun, 2010-10-31 at 20:26 +0100, Adrian Buehlmann wrote:
> > - Test function is too heavy-weight - do the simplest fastest test
> > possible. Probably something like:
> > 
> >  f, fn = getasecurefilehandleandname(path)
> >  fn2 = fn + "l"
> >  link(fn, fn2)
> >  stat
> > 
> > Even this is pretty expensive.
> 
> Near as I can tell, there's no test in my code that is not needed.
> 
> But I guess if even the above is too heavyweight, there's not much point
> in explaining in detail every case I covered.

Here's a quick test I made. It catches the Linux bug and should work on
Windows shares too.

def checknlink(testfile):
    rm = None
    try:
        for x in xrange(10):
            try:
                f = "%s.hgtmp%s" % (testfile, x)
		os.link(testfile, f)
                rm = f
            except OSError, inst:
                if inst == errno.EEXISTS:
                    continue
		elif inst == errno.EPERM:
                    return false
                raise
            break
        else:
            return False

	try:
            fd = open(f, "r", 0)
            s = os.stat(f)
            if s.st_nlink > 1:
		return True
        finally:
            fd.close()

    finally:
	if rm:
            os.unlink(rm)

    return False

Typically, this will do the bare minimum of syscalls: link, open, stat,
close, and unlink. Note how it uses an existing file (which we should
always have when doing this test) to avoid the tempfile issues (and the
poking around at /dev/urandom that tempfile.mkstemp wants to do).

-- 
Mathematics is the supreme nostalgia of our time.