RE: [Mingw-users] gcc -save-temps foo.c fails to build foo.o

[Danny Smith <danny_r_smith_2001 at yahoo dot co dot nz>]

The problem is indeed in st_ino.  Its always 0 in windows implementation
of stat (unless the file doesn't exist).  Using MSVCRT implemntation of
_fullpath to get the absolute pathname seems to be enough to ensure uniqueness
on mingw32

This patch to gcc.c  solves --save-temps problem for me, but needs to be made a
bit  more palatable.

Danny

Index: gcc.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/gcc.c,v
retrieving revision 1.344
diff -c -3 -p -r1.344 gcc.c
*** gcc.c	14 Oct 2002 07:15:38 -0000	1.344
--- gcc.c	16 Nov 2002 06:38:33 -0000
*************** do_spec_1 (spec, inswitch, soft_matched_
*** 4668,4673 ****
--- 4668,4677 ----
  		      {
  		      	struct stat st_temp;
  
+ #ifdef  __MINGW32__
+ 			char temp_fullpath[MAX_PATH];
+ 			char input_fullpath[MAX_PATH];
+ #endif
  		      	/* Note, set_input() resets input_stat_set to 0.  */
  		      	if (input_stat_set == 0)
  		      	  {
*************** do_spec_1 (spec, inswitch, soft_matched_
*** 4683,4690 ****
--- 4687,4702 ----
  
  			if (input_stat_set != 1
  			    || stat (temp_filename, &st_temp) < 0
+ #ifndef __MINGW32__
  			    || input_stat.st_dev != st_temp.st_dev
  			    || input_stat.st_ino != st_temp.st_ino)
+ #else
+ 		      	    || (strcmp (_fullpath (input_fullpath,
+ 						   input_filename, MAX_PATH),
+ 					_fullpath (temp_fullpath,
+ 						   temp_filename, MAX_PATH))
+ 				 != 0 )) 
+ #endif		
  			  {
  			    temp_filename = save_string (temp_filename,
  							 temp_filename_length + 1);




> "Rekha Deshmukh" <RekhaD@kpit.com> wrote:
> 
> > Hi Earnie,
> > 
> > Thanks for the reply. 
> > From your mail I understood that the patch should be modified,
> > where the st_ino of two files are compared and apply some other
> > logic so as to know the file's unique existance. 
> > I searched through the gcc source but was unable to find the
> > code, where two files were compared for uniqueness.
> 
>  The GNU ld has this problem too, when handling '.so' files
> on MinGW-host, the DT_NEEDED entries in the files don't work.
> The code causing this seems to be in 'emultemps/elf32.em' and
> is:
>  
> -------------------- clip ---------------------------
> /* See if an input file matches a DT_NEEDED entry by running stat on
>    the file.  */
> 
> static void
> gld${EMULATION_NAME}_stat_needed (s)
>      lang_input_statement_type *s;
> {
>   struct stat st;
>   const char *suffix;
>   const char *soname;
> 
>   if (global_found)
>     return;
>   if (s->the_bfd == NULL)
>     return;
> 
>   if (bfd_stat (s->the_bfd, &st) != 0)
>     {
>       einfo ("%P:%B: bfd_stat failed: %E\n", s->the_bfd);
>       return;
>     }
> 
>   if (st.st_dev == global_stat.st_dev		<----- !!
>       && st.st_ino == global_stat.st_ino)	<----- !!
>     {
>       global_found = true;
>       return;
>     }
> -------------------- clip ---------------------------
> 
>  If only the plain vanilla file name, without the absolute
> path, is the input here, how the 'uniqueness' will be
> solved ?  Or is it necessary -- in this case it would be
> enough if a file with the right name could be found. So
> wrapping the '&& st.st_ino == global_stat.st_ino)' between
> '#ifndef __MINGW32__' and '#endif' could fix this case.
> Assuming that the 'st.st_dev' has the right disk number...
> 
>  Haven't tried this yet though...
> 
> Cheers, Kai
> 
> 

http://careers.yahoo.com.au - Yahoo! Careers
- 1,000's of jobs waiting online for you!