apr-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "William A. Rowe, Jr." <wr...@rowe-clan.net>
Subject Re: Report of incorrect behaviour of apr_dir_read on win32
Date Wed, 05 Sep 2001 04:44:29 GMT
Thanks for your report!  I'd fixed with a similar patch some time ago for the
growing dots problem, but the problem with GetNamedSecurityInfoW should now
be resolved as well.

Bill

----- Original Message ----- 
From: "Mats Nilsson" <mats.nilsson@xware.se>
To: <dev@apr.apache.org>
Sent: Monday, August 06, 2001 12:20 PM
Subject: Report of incorrect behaviour of apr_dir_read on win32


> Hi
> 
> During my testing the subversion (http://subversion.tigris.org) on the 
> Win32 platform, I ran into some problems with the apr_dir_open/apr_dir_read 
> functions as implemented on Win32.
> 
> The following is tested on a WinNT 4.0 sp6, using MSVC6 sp3.
> As for which version of APR I'm using, I can't really tell. However, I 
> retrieved a APR CVS snapshot in april, and file_io/win32/dir.c has version 
> 1.56.
> 
> This test program demonstrate these problems:
> 
> ----- snip
> 
> #include "../apr/include/apr_pools.h"
> #include "../apr/include/apr_file_info.h"
> 
> #include <assert.h>
> 
> void test(const char *dirname) {
> apr_pool_t *pool; apr_pool_create(&pool, NULL);
> apr_dir_t *dir;
> apr_status_t err;
> apr_finfo_t this_entry;
> 
> err = apr_dir_open(&dir, dirname, pool);
> 
> /* Problem 1:
> * This call fails when dirname == "c:/temp", but not when it is "c:\\temp"
>          */
> err = apr_dir_read(&this_entry, APR_FINFO_NORM & ~APR_FINFO_IDENT, dir);
> assert(APR_STATUS_IS_SUCCESS(err));
> /* assert(strcmp(this_entry.name, ".") == 0);  This works */
> 
> err = apr_dir_read(&this_entry, APR_FINFO_NORM & ~APR_FINFO_IDENT, dir);
> 
> /* Problem 2: name contains "..." instead of ".." */
> assert(strcmp(this_entry.name, "..") == 0);
> 
> apr_dir_close(dir); apr_pool_destroy(pool);
> }
> 
> 
> int main(int argc, char **argv) {
> test("c:\\temp");
> test("c:/temp");
> return 0;
> }
> 
> ----- snip
> 
> 1. Sometimes the apr translates the directory name into something
> that some Win32 functions don't like.
> 
> Specifically if the directory given to apr_dir_open is given like 
> "d:/some/dir",
> this gets translated inside apr into "\\?\d:\some\dir", a filename format
> that GetNamedSecurityInfoW will not accept.
> 
> On the other hand, if the directory name is given like "d:\some\dir", the
> translation will not take place, and the above mentioned function works.
> 
> 
> 2. There is a bug in apr\file_io\win32\dir.c (apr_dir_read). Subsequent
> calls to this function will cause the internal variable wdirname to build up.
> 
> The first time wdirname contains    d:\some\dir\.
> Second time                         d:\some\dir\...
> Third time                          d:\some\dir\...somefile.txt
> 
> The following patch fixes this problem
> 
> *** apr.orig/file_io/win32/dir.c Thu Apr 12 18:26:22 2001
> --- apr/file_io/win32/dir.c Fri Jun 29 10:08:28 2001
> ***************
> *** 222,227 ****
> --- 222,228 ----
>              */
>     #if APR_HAS_UNICODE_FS
>             if (os_level >= APR_WIN_NT) {
> +             apr_status_t err;
>                 /* Almost all our work is done.  Tack on the wide file name
>                  * to the end of the wdirname (already / delimited)
>                  */
> ***************
> *** 228,234 ****
>                 if (!eos)
>                     eos = wcschr(wdirname, '\0');
>                 wcscpy(eos, thedir->w.entry->cFileName);
> !             return more_finfo(finfo, wdirname, wanted, MORE_OF_WFSPEC,
> os_leve
> l);
>             }
>             else {
>                 /* Don't waste stack space on a second buffer, the one we set
> --- 229,237 ----
>                 if (!eos)
>                     eos = wcschr(wdirname, '\0');
>                 wcscpy(eos, thedir->w.entry->cFileName);
> !             err = more_finfo(finfo, wdirname, wanted, MORE_OF_WFSPEC,
> os_level
> );
> !             eos[0] = '\0';
> !             return err;
>             }
>             else {
>                 /* Don't waste stack space on a second buffer, the one we set
> 
> 
> Sincerely
> 
> Mats Nilsson
> 
> 


Mime
View raw message