declare canonicalize_file_name

declare canonicalize_file_name

[Richard Henderson]

GDB uses canonicalize_file_name if it detects it in libc,
but doesn't define _GNU_SOURCE to get it declared.  This
results in the compiler thinking that the function returns
an integer, which results in the pointer value being 
truncated on ia64.

I figured providing a declaration was safer than forcing
_GNU_SOURCE.

Ok?


r~


	* configure.in (canonicalize_file_name): Look for declaration.
	* configure, config.in: Rebuild.
	* utils.c (canonicalize_file_name): Provide decl if required.

Index: configure.in
===================================================================
RCS file: /cvs/src/src/gdb/configure.in,v
retrieving revision 1.83
diff -c -p -d -r1.83 configure.in
*** configure.in	2002/03/06 21:16:00	1.83
--- configure.in	2002/03/14 21:24:49
*************** BFD_NEED_DECLARATION(free)
*** 241,247 ****
  BFD_NEED_DECLARATION(strerror)
  BFD_NEED_DECLARATION(strdup)
  BFD_NEED_DECLARATION(strstr)
! 
  
  # The following save_state_t checkery is only necessary for HPUX 
  # versions earlier than 10.20.  When those fade from memory, this 
--- 241,247 ----
  BFD_NEED_DECLARATION(strerror)
  BFD_NEED_DECLARATION(strdup)
  BFD_NEED_DECLARATION(strstr)
! BFD_NEED_DECLARATION(canonicalize_file_name)
  
  # The following save_state_t checkery is only necessary for HPUX 
  # versions earlier than 10.20.  When those fade from memory, this 
Index: utils.c
===================================================================
RCS file: /cvs/src/src/gdb/utils.c,v
retrieving revision 1.68
diff -c -p -d -r1.68 utils.c
*** utils.c	2002/03/01 06:19:24	1.68
--- utils.c	2002/03/14 21:24:49
*************** extern PTR realloc ();
*** 81,86 ****
--- 81,91 ----
  #ifdef NEED_DECLARATION_FREE
  extern void free ();
  #endif
+ /* Actually, we'll never have the decl, since we don't define _GNU_SOURCE.  */
+ #if defined(HAVE_CANONICALIZE_FILE_NAME) \
+     && defined(NEED_DECLARATION_CANONICALIZE_FILE_NAME)
+ extern char *canonicalize_file_name (const char *);
+ #endif
  
  #undef XMALLOC
  #define XMALLOC(TYPE) ((TYPE*) xmalloc (sizeof (TYPE)))

Re: declare canonicalize_file_name

[Andrew Cagney]

> GDB uses canonicalize_file_name if it detects it in libc,
> but doesn't define _GNU_SOURCE to get it declared.  This
> results in the compiler thinking that the function returns
> an integer, which results in the pointer value being 
> truncated on ia64.
> 
> I figured providing a declaration was safer than forcing
> _GNU_SOURCE.

I believe RichardE's fixed this (or will when he wakes up :-).  GDB 
should have been using realpath() in preference to 
canonicalize_file_name() .

enjoy,
Andrew

(See gdb/409)

Re: declare canonicalize_file_name

[Richard Henderson]

On Thu, Mar 14, 2002 at 05:33:29PM -0500, Andrew Cagney wrote:
> I believe RichardE's fixed this (or will when he wakes up :-).  GDB 
> should have been using realpath() in preference to 
> canonicalize_file_name() .

The problem will still hold with canonicalize_file_name
should it be used.

Of course, since it's specific to glibc, and glibc also
provides realpath, it won't ever be used.  Why not remove
support for it then?


r~

Re: declare canonicalize_file_name

[Andrew Cagney]

> On Thu, Mar 14, 2002 at 05:33:29PM -0500, Andrew Cagney wrote:
> 
>> I believe RichardE's fixed this (or will when he wakes up :-).  GDB 
>> should have been using realpath() in preference to 
>> canonicalize_file_name() .
> 
> 
> The problem will still hold with canonicalize_file_name
> should it be used.
> 
> Of course, since it's specific to glibc, and glibc also
> provides realpath, it won't ever be used.  Why not remove
> support for it then?

The call was only just added for the hurd :-(  Hmm, 
http://sources.redhat.com/ml/gdb-patches/2002-02/msg00631.html

Oops, ulgh! if RichardE tweeks to code (as I suggested) to simply prefer 
realpath() it will break the hurd again.  I think prefering realpath() 
is correct (only use an obscure function when forced too) but that logic 
is going to need to be scrambled a bit more :-(.

RichardE, I guess another re-think.

Andrew

Re: declare canonicalize_file_name

[rth]

On Thu, Mar 14, 2002 at 06:43:55PM -0500, Andrew Cagney wrote:
> Oops, ulgh! if RichardE tweeks to code (as I suggested) to simply prefer 
> realpath() it will break the hurd again.  I think prefering realpath() 
> is correct (only use an obscure function when forced too) but that logic 
> is going to need to be scrambled a bit more :-(.

What about something like this?


r~


Index: utils.c
===================================================================
RCS file: /cvs/src/src/gdb/utils.c,v
retrieving revision 1.68
diff -c -p -d -r1.68 utils.c
*** utils.c	2002/03/01 06:19:24	1.68
--- utils.c	2002/03/15 00:06:43
*************** extern PTR realloc ();
*** 81,86 ****
--- 81,91 ----
  #ifdef NEED_DECLARATION_FREE
  extern void free ();
  #endif
+ /* Actually, we'll never have the decl, since we don't define _GNU_SOURCE.  */
+ #if defined(HAVE_CANONICALIZE_FILE_NAME) \
+     && defined(NEED_DECLARATION_CANONICALIZE_FILE_NAME)
+ extern char *canonicalize_file_name (const char *);
+ #endif
  
  #undef XMALLOC
  #define XMALLOC(TYPE) ((TYPE*) xmalloc (sizeof (TYPE)))
*************** string_to_core_addr (const char *my_stri
*** 2532,2551 ****
  char *
  gdb_realpath (const char *filename)
  {
! #ifdef HAVE_CANONICALIZE_FILE_NAME
!   return canonicalize_file_name (filename);
! #elif defined (HAVE_REALPATH)
! #if defined (PATH_MAX)
    char buf[PATH_MAX];
! #elif defined (MAXPATHLEN)
    char buf[MAXPATHLEN];
! #elif defined (HAVE_UNISTD_H) && defined(HAVE_ALLOCA)
    char *buf = alloca ((size_t)pathconf ("/", _PC_PATH_MAX));
! #else
! #error "Neither PATH_MAX nor MAXPATHLEN defined"
! #endif
    char *rp = realpath (filename, buf);
    return xstrdup (rp ? rp : filename);
  #else
    return xstrdup (filename);
  #endif
--- 2537,2560 ----
  char *
  gdb_realpath (const char *filename)
  {
! #if defined(HAVE_REALPATH)
! # if defined (PATH_MAX)
    char buf[PATH_MAX];
! #  define USE_REALPATH
! # elif defined (MAXPATHLEN)
    char buf[MAXPATHLEN];
! #  define USE_REALPATH
! # elif defined (HAVE_UNISTD_H) && defined(HAVE_ALLOCA)
    char *buf = alloca ((size_t)pathconf ("/", _PC_PATH_MAX));
! #  define USE_REALPATH
! # endif
! #endif /* HAVE_REALPATH */
! 
! #if defined(USE_REALPATH)
    char *rp = realpath (filename, buf);
    return xstrdup (rp ? rp : filename);
+ #elif defined(HAVE_CANONICALIZE_FILE_NAME)
+   return canonicalize_file_name (filename);
  #else
    return xstrdup (filename);
  #endif

Re: declare canonicalize_file_name

[Andrew Cagney]

> On Thu, Mar 14, 2002 at 06:43:55PM -0500, Andrew Cagney wrote:
> 
>> Oops, ulgh! if RichardE tweeks to code (as I suggested) to simply prefer 
>> realpath() it will break the hurd again.  I think prefering realpath() 
>> is correct (only use an obscure function when forced too) but that logic 
>> is going to need to be scrambled a bit more :-(.
> 
> 
> What about something like this?

Yes, lets see.  Thanks!

Andrew

Re: declare canonicalize_file_name

[Alfred M. Szmidt]

* Andrew Cagney writes:
> The call was only just added for the hurd :-( Hmm,
> http://sources.redhat.com/ml/gdb-patches/2002-02/msg00631.html

This was added because realpath() is horribly broken, and that the Hurd
does not define any MAXPATHLEN/PATH_MAX, which in turn breaks realpath().
Which the old code depended on to be defined.

The only obscure function is realpath() that depends on the values of
MAXPATHLEN/PATH_MAX to be defined, which are not required by POSIX,
and on any system that actually provides canonicalize_file_name() it
should be used. Specially on GNU and GNU/Linux.

> Oops, ulgh! if RichardE tweeks to code (as I suggested) to simply
> prefer realpath() it will break the hurd again.  I think prefering
> realpath() is correct (only use an obscure function when forced too)
> but that logic is going to need to be scrambled a bit more :-(.

The code that has been checked in now looks OK, other then the fact that
GNU/Linux will use the obscure function realpath().

-- 
Alfred M. Szmidt