#34 stat fails on directory ending with slash

[Pascal Obry]

See attached patch which fixes this issue on _stat.c and _wstat.c.

[Pascal Obry]

Fix problem reported.

[Pascal Obry]

I forgot to say that this has been tested on i686 and x86_64.

[Kai Tietz]

Thanks for the patch. I will give it some test. Maybe it would be good to add here a testcase for those stat function- I want to verify if the code here handles '/' and UNC correct.

Kai

[Ozkan Sezer]

As I understand it, stat("C:\\") or stat("C:/") are OK (or, ignored by the workaround). How about stat("/") or stat("\\") ?

[Pascal Obry]

Thanks for the quick feedback. Here is a small test case I have used:

#include <stdio.h>
#include <sys/stat.h>

int
main (int argc, char **argv)
{
struct stat buf;
char *f1 = "c:\\windows\\";
char *f2 = "c:\\";
wchar_t *wf1 = L"c:\\windows\\";
wchar_t *wf2 = L"c:\\";

printf ("stat(%s) = %d\n", f1, stat(f1, &buf));
printf ("stat(%s) = %d\n", f2, stat(f2, &buf));

wprintf (L"stat(%s) = %d\n", wf1, wstat(wf1, &buf));
wprintf (L"stat(%s) = %d\n", wf2, wstat(wf2, &buf));
}

[Pascal Obry]

> As I understand it, stat("C:\\") or stat("C:/") are OK (or, ignored by the
> workaround).

Right.

> How about stat("/") or stat("\\") ?

Not tested, I'll update my patch which is probably broken here.

[Pascal Obry]

Patch V2 - fix handling of / and \

[Kai Tietz]

I suggest the following pattern (here for _stat, but same for _wstat with different type)

[code]
static char*
_mingw_no_trailing_slash (const char* _path)
{
int len;
char *p;
int start;

p = (char*)_path;

if (_path && *_path) {
len = strlen (_path);

/* Ignore X:\ */

if ((len == 2 || len == 3) && _path[1] == ':')
return p;
if (len <= 1)
return p;
/* Check UNC \\abc\<name>\ */
if ((_path[0] == '\\' || _path[0] == '/')
&& (_path[1] == '\\' || _path[1] == '/'))
{
const char *r = &_path[2];
while (*r != 0 && *r != '\\' && *r != '/')
++r;
if (*r != 0)
++r;
if (*r == 0)
return p;
while (*r != 0 && *r != '\\' && *r != '/')
++r;
if (*r != 0)
++r;
if (*r == 0)
return p;
}

if (_path [len - 1] == '/' || _path [len - 1] == '\\')
{
p = (char*)malloc (len);
memcpy (p, _path, len - 1);
p [len - 1] = '\0';
}
}

return p;
}
[/code]

[Pascal Obry]

New patch proposed. Fixes / and \. Here is the new testcase. I have not tested the UNC path.

<<
#include <stdio.h>
#include <sys/stat.h>

int
main (int argc, char **argv)
{
struct stat buf;
char *f1 = "c:\\windows\\";
char *f2 = "c:\\";
char *f3 = "\\";
char *f4 = "/";
wchar_t *wf1 = L"c:\\windows\\";
wchar_t *wf2 = L"c:\\";
wchar_t *wf3 = L"\\";
wchar_t *wf4 = L"/";

printf ("stat(%s) = %d\n", f1, stat(f1, &buf));
printf ("stat(%s) = %d\n", f2, stat(f2, &buf));
printf ("stat(%s) = %d\n", f3, stat(f3, &buf));
printf ("stat(%s) = %d\n", f4, stat(f4, &buf));

wprintf (L"stat(%s) = %d\n", wf1, wstat(wf1, &buf));
wprintf (L"stat(%s) = %d\n", wf2, wstat(wf2, &buf));
wprintf (L"stat(%s) = %d\n", wf3, wstat(wf3, &buf));
wprintf (L"stat(%s) = %d\n", wf4, wstat(wf4, &buf));
}
>>

[Pascal Obry]

Ktietz70,

You code seems to be ok expect that for an UNC path without trailing slash : \\abc\<name>

In this case the stat() routine fails whereas a dir \\abc\<name> works!

I think that this is a minor case not worth covering.

New patch attached.

[Pascal Obry]

Patch V3 - support for UNC path too

[Pascal Obry]

New testcase here:

<<
#include <stdio.h>
#include <sys/stat.h>

void
test (char *path)
{
struct stat buf;
if (stat(path, &buf) == 0)
printf ("OK [%s]\n", path);
else
printf ("ERROR [%s]\n", path);
}

void
wtest (wchar_t *path)
{
struct stat buf;
if (wstat(path, &buf) == 0)
wprintf (L"OK [%s]\n", path);
else
wprintf (L"ERROR [%s]\n", path);
}

int
main (int argc, char **argv)
{
test ("c:\\windows\\");
test ("c:\\");
test ("\\");
test ("/");

wtest (L"c:\\windows\\");
wtest (L"c:\\");
wtest (L"\\");
wtest (L"/");
}
>>

I have tested with UNC too.

[Kai Tietz]

Patch applied to trunk at revision 3832. Possibly something for our v1 branch, too.

Thanks,
Kai

[Pascal Obry]

Thanks, very responsive team!

[Ozkan Sezer]

The problem is that we have these same functions as inlines in sys/stat.h, too, and there is no workaround applied to them. The patch may be a candidate for the release branch if the inline situation is handled along with the static libmingwex versions.

[Kai Tietz]

Yes, good catch. So we have here two different ways to solve it. Make inlines using the path-adjustment routine from libmingwex (inlining it makes IMHO no sense here). Or remove inlines for them, which is also not a good choice IMHO. Btw the _stat64i32 (and its wide-character variant) needs same handling in my opinion. So we should make those unify-functions global and add them to sys/stat.h header.

Kai

[Kai Tietz]

Issue is fixed. Therefore close thread.