GetLastError strangeness with FindClose

[Alex Pasadyn]

Hi all,
I observed a program that was bailing out with an absolutely baffling 
error message.  I have traced it to its cause, but now I'm stuck. 
Basically, it was telling me that it had an error in FindNextFile of 
"success".  I did a relay trace and found out that it was calling 
FindFirstFile, FindNextFile a bunch of times, FindClose, and then 
GetLastError.

So, I put together a little test:

#include <windows.h>
#include <stdio.h>

WIN32_FIND_DATA fd;
HANDLE h;

int main(int argc, char *argv[])
{

h = FindFirstFile("*", &fd);
     if (h == INVALID_HANDLE_VALUE)
     {
         printf("Error finding first file: %d\n", GetLastError());
         return 1;
     }
     printf("found file: %s\n", fd.cFileName);
     while (FindNextFile(h, &fd))
     {
         printf("found file: %s\n", fd.cFileName);
     }
     FindClose(h);
     switch (GetLastError())
     {
     case ERROR_NO_MORE_FILES:
         printf("all done!\n");
         return 0;
     default:
         printf("Error finding next file: %d\n", GetLastError());
         return 1;
     }
     return 1;
}

Now, on Windows, the output is:
G:\testfnf>testfnf
found file: .
found file: ..
found file: testfnf.cpp
found file: testfnf.exe
all done!

And with Wine the output is:
[ajp@toaster testfnf]$ ~/wine/wine testfnf.exe
found file: .
found file: ..
found file: testfnf.cpp
found file: testfnf.exe
Error finding next file: 0

So, it looks like somewhere in FindClose Wine is resetting the last 
error.  I started digging through there and the functions it calls but I 
was unable to find where it was getting reset.  Can anyone shed some 
light on this?  Is there any way to break on that value changing?

Thanks,
-ajp

[Eric Pouech]

Alex Pasadyn wrote:
> So, it looks like somewhere in FindClose Wine is resetting the last 
> error.  I started digging through there and the functions it calls but I 
> was unable to find where it was getting reset.  Can anyone shed some 
> light on this?  Is there any way to break on that value changing?

GlobalUnlock does set last error to 0 if it's unlocking the last lock.

A+

[Dmitry Timoshkov]

"Eric Pouech"  wrote:
> > So, it looks like somewhere in FindClose Wine is resetting the last 
> > error.  I started digging through there and the functions it calls but I 
> > was unable to find where it was getting reset.  Can anyone shed some 
> > light on this?  Is there any way to break on that value changing?
>
> GlobalUnlock does set last error to 0 if it's unlocking the last lock.

A way to go might be to use HeapAlloc instead of GlobalAlloc and add
a critical section to the allocated data structure to protect the object.

Alex, are you willing to do the job?

[Alex Pasadyn]

Dmitry Timoshkov wrote:
> "Eric Pouech"  wrote:
> > > So, it looks like somewhere in FindClose Wine is resetting the last 
> > > error.  I started digging through there and the functions it calls but I 
> > > was unable to find where it was getting reset.  Can anyone shed some 
> > > light on this?  Is there any way to break on that value changing?
> >
> > GlobalUnlock does set last error to 0 if it's unlocking the last lock.
>
> A way to go might be to use HeapAlloc instead of GlobalAlloc and add
> a critical section to the allocated data structure to protect the object.
>
> Alex, are you willing to do the job?

I'll try that and see how it turns out.

-ajp

[Kelly Leahy]

Your test doesn't make much sense.  The call to FindClose should be done
after you check the result of GetLastError, since FindClose <should> change
the last error to ERROR_SUCCESS if the close operation was successful.
> From your test results, it seems that the FindClose implementation on the

version of windows you're using doesn't seem to do this.  I would consider
this a bug in windows, not a bug in Wine.  If a programmer relies on this
behavior, it's their fault, not MS's or Wine's fault that their code does
not work.

----- Original Message ----- 
From: "Alex Pasadyn" 
To: 
Sent: Saturday, August 16, 2003 11:55 AM
Subject: GetLastError strangeness with FindClose
> Hi all,
> I observed a program that was bailing out with an absolutely baffling
> error message.  I have traced it to its cause, but now I'm stuck.
> Basically, it was telling me that it had an error in FindNextFile of
> "success".  I did a relay trace and found out that it was calling
> FindFirstFile, FindNextFile a bunch of times, FindClose, and then
> GetLastError.
>
> So, I put together a little test:
>
> #include <windows.h>
> #include <stdio.h>
>
> WIN32_FIND_DATA fd;
> HANDLE h;
>
> int main(int argc, char *argv[])
> {
>  h = FindFirstFile("*", &fd);
>  if (h == INVALID_HANDLE_VALUE)
>  {
>      printf("Error finding first file: %d\n", GetLastError());
>      return 1;
>  }
>  printf("found file: %s\n", fd.cFileName);
>  while (FindNextFile(h, &fd))
>  {
>      printf("found file: %s\n", fd.cFileName);
>  }
>  FindClose(h);
>  switch (GetLastError())
>  {
>  case ERROR_NO_MORE_FILES:
>      printf("all done!\n");
>      return 0;
>  default:
>      printf("Error finding next file: %d\n", GetLastError());
>      return 1;
>  }
>  return 1;
>
> }
>
> Now, on Windows, the output is:
> G:\testfnf>testfnf
> found file: .
> found file: ..
> found file: testfnf.cpp
> found file: testfnf.exe
> all done!
>
> And with Wine the output is:
> [ajp@toaster testfnf]$ ~/wine/wine testfnf.exe
> found file: .
> found file: ..
> found file: testfnf.cpp
> found file: testfnf.exe
> Error finding next file: 0
>
> So, it looks like somewhere in FindClose Wine is resetting the last
> error.  I started digging through there and the functions it calls but I
> was unable to find where it was getting reset.  Can anyone shed some
> light on this?  Is there any way to break on that value changing?
>
> Thanks,
> -ajp

[Eric Pouech]

Kelly Leahy wrote:
> Your test doesn't make much sense.  The call to FindClose should be done
> after you check the result of GetLastError, since FindClose <should> change
> the last error to ERROR_SUCCESS if the close operation was successful.
> > From your test results, it seems that the FindClose implementation on the
>
> version of windows you're using doesn't seem to do this.  I would consider
> this a bug in windows, not a bug in Wine.  If a programmer relies on this
> behavior, it's their fault, not MS's or Wine's fault that their code does
> not work

AFAIK, it's never said that a successful function call should set the 
last error to 0. This only thing MS claims is that when a function 
fails, the last error is set to a error code explaining that error.

Moreover, wine's goal is not to make a better design than windows, but 
to mimic windows behavior. So, if FindClose doesn't change the last 
error on success, wine should do the same (especially if some known 
program depends on it)

A+

[Kelly Leahy]

> AFAIK, it's never said that a successful function call should set the
> last error to 0. This only thing MS claims is that when a function
> fails, the last error is set to a error code explaining that error.
>
> Moreover, wine's goal is not to make a better design than windows, but
> to mimic windows behavior. So, if FindClose doesn't change the last
> error on success, wine should do the same (especially if some known
> program depends on it)

You're missing my point (or maybe you're not and you just don't agree with
me - in either case, let me make my point in a clearer way).  If windows
doesn't say it won't change the last error, we, as programmers, should
assume that it might.  In my experience, a large number of functions that
set last error on failure set it to ERROR_SUCCESS on success.  For the few
that don't, I would consider this an oversight on MS's part, and I wouldn't
expect it to exist in all implementations of a particular function (across
OS versions).

Just my 0.02.

Kelly

[Eric Pouech]

> You're missing my point (or maybe you're not and you just don't agree with
> me - in either case, let me make my point in a clearer way).  If windows
> doesn't say it won't change the last error, we, as programmers, should
> assume that it might.  In my experience, a large number of functions that
> set last error on failure set it to ERROR_SUCCESS on success.  For the few
> that don't, I would consider this an oversight on MS's part, and I wouldn't
> expect it to exist in all implementations of a particular function (across
> OS versions).

let's just test Alex small test program on various platforms and see 
what gives...
A+

[Alex Pasadyn]

Eric Pouech wrote:
> > You're missing my point (or maybe you're not and you just don't agree 
> > with
> > me - in either case, let me make my point in a clearer way).  If windows
> > doesn't say it won't change the last error, we, as programmers, should
> > assume that it might.  In my experience, a large number of functions that
> > set last error on failure set it to ERROR_SUCCESS on success.  For the 
> > few
> > that don't, I would consider this an oversight on MS's part, and I 
> > wouldn't
> > expect it to exist in all implementations of a particular function 
> > (across
> > OS versions).
>
> let's just test Alex small test program on various platforms and see 
> what gives...
> A+

FWIW, the original program that was giving me a problem was the Diablo 2 
1.09 to 1.10beta updater.  The workaround I found was to move all the 
files from my "save" subdirectory to somewhere else, run the patch, and 
then move them all back.  Having any files in the save directory caused 
it to bomb out with a message about FindNextFile failing due to "success".

I made that test program based on the series of calls I saw in the relay 
trace, and it gave a different result with Windows (9x was all I had to 
test) from Wine.

Anyway, if I were writing it I would put the check for the error code 
before the FindClose, but it wasn't my program...

Cheers,
-ajp

[Francois Gouget]

On Sat, 16 Aug 2003, Eric Pouech wrote:
[...]
> let's just test Alex small test program on various platforms and see
> what gives...

Alex's program works fine on NT4 so it seems to be a valid test. It
looks like a good candidate for Wine's conformance tests. Alex, do you
feel up to it?

[Ferenc Wagner]

Alex Pasadyn  writes:
> it was telling me that it had an error in FindNextFile of
> "success".  I did a relay trace and found out that it was
> calling FindFirstFile, FindNextFile a bunch of times,
> FindClose, and then GetLastError.

What happens if you change the Windows version?  To
http://www.winehq.org/hypermail/wine-patches/2003/05/0334.html
(at the bottom) I got the response that it is probably
version dependent, and your problem seems to be similar.

Feri.