NOTE: I notice this problem ONLY the first time I load a directory
using a file url the file i am looking for is not listed
if i reload the page the file then appears.

to recreate:
1. load file:/usr/sqe/doc
2. notice hotjava1.0fcs_testplan.html is not listed
3. reload page
4. notice hotjava1.0fcs_testplan.html is listed

FYI this bug is difficult to recreate after the first time you reload the page
the problem seems to go away for the remainder of the session.

sometimes even if you relaunch hotjava the problem no longer occurs
so i have saved 2 files that demonstrate the difference if needed

the files are
file:/home/pkelley/DIR.one = before the reload (missing the filename)
file:/home/pkelley/DIR.two = after the reload (contains the file in question)

------------------------------------------------------------------------
Another user reports the same symtoms:

Synopsis: File.list() can return incomplete lists

Description:
When trying to list a large, NFS-mounted directory using
File.list(), on a very busy machine, the native code call
to readdir() may return EINTR (Interrupted System Call).
In this case, File.list() will just return what it's got
rather than retrying.

We have a copy of the JDK1.1.something sources. Here's the
code in question, from src/share/java/lang/filesystem.c:

    /* scan the directory */
    while ((ptr = sysReadDir(dir)) != NULL) {
   if (strcmp(ptr->d_name, ".") && strcmp(ptr->d_name, "..")) {
   if (len == maxlen) {
  narr = (HArrayOfString *) ArrayAlloc(T_CLASS, maxlen <<= 1);
  if (narr == NULL) {
char utfPath[128];
javaString2UTF(thisptr->path, utfPath, sizeof(utfPath));
   sysCloseDir(dir);
   SignalError(0, JAVAPKG "OutOfMemoryError", utfPath);
  arr = NULL; /* Hide arr from GC */
   return NULL;
   }
  memcpy(unhand(narr), unhand(arr), len * sizeof(HString *));
  arr = narr;
  narr = 0; /* Hide narr from GC */
   }
  
   /* add it to the list */
unhand(arr)->body[len++] =
makeJavaStringFromPlatformCString
(ptr->d_name, strlen(ptr->d_name));
   }
    }

What's happening here is that sysReadDir (which is
just a macro mapping to readdir()) is returning NULL, and
errno is set. This code doesn't check for errno, but just
blindly assumes that this is the end of the directory. In
real life I'm getting an EINTR, and we should be trying
again.

This causes a P1 bug for us, even in our prerelease copy of
the JDK1.1.2C.

I've tried to create a simple test case that just does a
File.list() of a large (1000-file) NFS-mounted directory,
but it doesn't reproduce the problem. However, running
some of our AWT-oriented apps definitely does, every single
time (they just aren't small test cases, sorry.) But I can
tell you that I've edited the VM code above and recompiled it
and confirmed that this is exactly what's going on.