Can telldir() == 0 value be made special?

[Conrad Meyer]

One could also make it reserve the zero slot for the loc == 0 entry,
but not create it dynamically.  Just set loccnt=1 at initialization
and then special-case dd_loc==0.  Like this:

--- a/lib/libc/gen/opendir.c
+++ b/lib/libc/gen/opendir.c
@@ -295,7 +295,7 @@ __opendir_common(int fd, int flags, bool use_current_pos)
        dirp->dd_lock = NULL;
        dirp->dd_td = (struct _telldir *)((char *)dirp + sizeof(DIR));
        LIST_INIT(&dirp->dd_td->td_locq);
-       dirp->dd_td->td_loccnt = 0;
+       dirp->dd_td->td_loccnt = 1;
        dirp->dd_compat_de = NULL;

        /*
--- a/lib/libc/gen/telldir.c
+++ b/lib/libc/gen/telldir.c
@@ -76,7 +76,10 @@ telldir(DIR *dirp)
                                _pthread_mutex_unlock(&dirp->dd_lock);
                        return (-1);
                }
-               lp->loc_index = dirp->dd_td->td_loccnt++;
+               if (dirp->dd_loc == 0)
+                       lp->loc_index = 0;
+               else
+                       lp->loc_index = dirp->dd_td->td_loccnt++;
                lp->loc_seek = dirp->dd_seek;
                lp->loc_loc = dirp->dd_loc;
                if (flp != NULL)

Best,
Conrad


On Tue, Aug 22, 2017 at 11:35 AM, Eitan Adler <lists at eitanadler.com> wrote:
> On 22 August 2017 at 14:30, Nikolaus Rath <Nikolaus at rath.org> wrote:
>> On Aug 22 2017, Conrad Meyer <cem at freebsd.org> wrote:
>>> Hi Nikolaus,
>>>
>>> As you have surmised, DIR* seekpoints are created dynamically whenever
>>> requested by user's telldir() call:
>>
>> Good to know, thanks!
>>
>>> I believe we could special case zero without breaking ABI
>>> compatibility of correct programs.  Something like this:
>>>
>>> --- a/lib/libc/gen/telldir.c
>>> +++ b/lib/libc/gen/telldir.c
>>> @@ -70,6 +70,20 @@ telldir(DIR *dirp)
>>>                 }
>>>         }
>>>         if (lp == NULL) {
>>> +               /* Create special zero telldir entry, similar to Linux */
>>> +               if (dirp->dd_td->td_loccnt == 0 && dirp->dd_loc != 0) {
>>> +                       lp = malloc(sizeof(struct ddloc));
>>> +                       if (lp == NULL) {
>>> +                               if (__isthreaded)
>>> +                                       _pthread_mutex_unlock(&dirp->dd_lock);
>>> +                               return (-1);
>>> +                       }
>>> +                       lp->loc_index = dirp->dd_td->td_loccnt++;
>>> +                       lp->loc_seek = 0;
>>> +                       lp->loc_loc = 0;
>>> +                       LIST_INSERT_HEAD(&dirp->dd_td->td_locq, lp, loc_lqe);
>>> +               }
>>> +
>>>                 lp = malloc(sizeof(struct ddloc));
>>>                 if (lp == NULL) {
>>>                         if (__isthreaded)
>>>
>>> I don't know if there are any downsides to special-casing zero like
>>> this, other than additional code complexity.
>>
>> This seems a little more complex than I would have expected.
>
>> Is this
>> maybe turning zero into a valid argument for seekdir() even if telldir()
>> has never been called?
>
> No, this code is called inside of telldir. I'd be against adding that
> behavior by contract since it complicates the contract of seekdir. No
> objection to this behavior by implementation though if it is simpler
> than modifying telldir (doubtful).
>
> What this is doing is ensuring that there will always be a 0 entry
> *after* telldir is called, and that seekdir(0) will return to the
> start of the directory if called with an argument of 0.
>
>
>> That would be even better, but the minimal change that I need is just to
>> ensure that telldir() either (a) never returns zero, or (b) only returns
>> zero if called at the beginning of the stream.