Syslinux - Dec 2010 - CMENU file handling problems.

I'm writing something that's based on Murali Krishnan Ganapathy's
CMENU
system,
except mine's driven by a menu.txt file that I supply (mostly so I can
edit the
menu.txt wherever I happen to be and so I won't need a syslinux build
environment
everytime the menu structure changes).

I'm able to build up menus that do what I expect, so that's good.

However, I want to be able to grey out menus (mark them INACTIVE) or
hide them
altogether (mark them INVISIBLE). I want to do this based on the
presence or
absence of a file (or maybe files). So, for example, if I don't include
SLAX
on today's DVD, then /SLAX/boot/initrd.gz will not exist, so I might
choose
to write menu.txt so as to hide that option.

In my original testing the feature worked. With a more complete
menu.txt,
the feature fails.

Poking around a bit further it seems that after ~29 attempts like this:

FILE* ifp;
const char* fname = "/some/file/that/does/NOT/exist";

if ((ifp = fopen(fname, "r")) == 0) {
  // code for failure
} else {
  // code for success
  fclose(ifp);
}

then a simple fopen(name, "r") of a file that *does* exist fails,
even though that exact same fopen will succeed if I try it before
the ~29 fopen()s on a file that foes not exist.

If I change the deliberatly failing pathname to one that succeeds, then
there seems to be no problem.

So it looks like I've hit a bug. Possibly the file descriptor table
entry
is not cleared when a file is closed or maybe older entries do not get
reused even though they are free?

Is this a known issue? (I'd go hunting but I'm not sure where in that
tree
the relevant code lives).

Opening a file is just a simple proxy for determining its existence: if
there's
an alternative way (that will work for ~150 files) then I'm happy to
change.

This is all under syslinux 4.03 and the C32 build environment is Ubuntu
10.10.

Thanks,

Antonio Carlini
arcarlini at iee.org

On 12/05/2010 03:23 PM, arcarlini at iee.org wrote:
> 
> Poking around a bit further it seems that after ~29 attempts like this:
> 
> FILE* ifp;
> const char* fname = "/some/file/that/does/NOT/exist";
> 
> if ((ifp = fopen(fname, "r")) == 0) {
>   // code for failure
> } else {
>   // code for success
>   fclose(ifp);
> }
> 
> then a simple fopen(name, "r") of a file that *does* exist fails,
> even though that exact same fopen will succeed if I try it before
> the ~29 fopen()s on a file that foes not exist.
> 
Sounds like some kind of file descriptor leak, indeed.  The number 29 is
particularly suspicious: 32-3 (for stdin/stdout/stderr).

	-hpa

H. Peter Anvin wrote:
> It seems to be *almost* that simple.  I just pushed a patch
> into the git tree and released syslinux-4.04-pre2, if you
> could test it that would be great.
Looks good both in my noddy test C32 and the full program (which
managed to successfully open() a file after ~64 failed open()s).

Many thanks,

Antonio
arcarlini at iee.org