Re: stat() replacement with _LARGE_FILES on AIX

[Eric Blake]

Michael Haubenwallner <michael.haubenwallner <at> salomon.at> writes:

> > If stat is defined, we can't be positive it was defined to stat64 or some
> > other spelling, like __stat64.  I'd feel safer blindly replacing all known
> > spellings, as and when we encounter them.
> 
> Fine with me.
> 
> > So that I can fully understand the situation on AIX, what does the system
> > header show for stat64()?  Is there a struct stat64?  I guess if the
> > #define stat stat64 occurs before either declaration of stat() or struct
> > stat, that would explain what you are seeing.
> 
> The declarations are in this order in <sys/stat.h> on AIX5.3:

Thanks - that confirms what I wanted to know.

> 
> ---
> #if !defined(_LARGE_FILES)
> struct  stat { ... };
> #endif

struct stat only exists if _LARGE_FILES is undefined...

> 
> #if defined(_LARGE_FILES) || defined(_LARGE_FILE_API)
> struct  stat64 { ... };
> #endif /* _LARGE_FILES || LARGE_FILE_API */

struct stat64 exists for _LARGE_FILES and also if you can see both stat() and 
stat64()...

> 
> #ifdef _LARGE_FILES
> # define stat    stat64
> # define fstat   fstat64

This is overriding only stat(), since struct stat was undefined to begin with 
on this preprocessor path, but picking up on the fact that only struct stat64 
was declared...


> # if _XOPEN_SOURCE_EXTENDED==1
> #  define lstat   lstat64
> # endif
> #endif  /* _LARGE_FILES */
> 
>     extern int  stat(const char *__restrict__, struct stat *__restrict__);

This line declares:

stat(struct stat)

if _LARGE_FILES was undefined, otherwise it declares two synonyms:

stat(struct stat64) aka stat64(struct stat64)

>     extern int  fstat(int, struct stat *);
> #ifdef _LARGE_FILE_API
>     extern int  stat64(const char *__restrict__, struct stat64 *__restrict__);

If _LARGE_FILES, this is a repeat delaration, and we have exactly:

stat(struct stat64)
stat64(struct stat64)

visible to the app.  If _LARGE_FILE_API but not _LARGE_FILES, then we have:

stat(struct stat)
stat64(struct stat64)

visible to the app.  And with neither, then we see only stat(struct stat).

> Do as you like, have attached another patch anyway.

Your second patch looks like it solves the problem nicely.  We end up with this 
with _LARGE_FILES:

stat(struct stat) => rpl_stat(struct stat64)

which is exactly what we want.  And there is no change if _LARGE_FILES is 
undefined.

> 
> Do you still need (offlist?) the whole header file?

Nope, you've done enough.  Thanks!  Now all I have to do is test that your 
patch doesn't break Linux or Solaris (that is, either autoconf doesn't set 
_LARGE_FILES for those platforms, or the use of _LARGE_FILES on those platforms 
doesn't break anything).

-- 
Eric Blake