Skip to contentSkip to conversation

The naming of at()s is a difficult matter

DISCUSSION H. Peter Anvin 5 participants 17 messages Feb 12, 2006 – Feb 14, 2006

lore
Share & tools
lore
#0 lore raw

I have noticed that the new ...at() system calls are named in what appears to be a completely haphazard fashion. In Unix system calls, an f- prefix means it operates on a file descriptor; the -at suffix (a prefix would have been more consistent, but oh well) similarly indicates it operates on a (directory fd, pathname) pair.

However, some system calls, in particular fchownat, futimesat, fchmodat and faccessat add the f- prefix for what appears to be absolutely no good reason. Logically, these system calls should be named chownat, utimesat, chmodat, and accessat.

I understand some of this braindamage comes from Solaris, but some of these calls do not. We should avoid it if at all possible, and I would recommend at least introducing aliases with the sane names.

-hpa

#1 lore raw
Quoted 7 lines

I have noticed that the new ...at() system calls are named in what appears to be a completely haphazard fashion. In Unix system calls, an f- prefix means it operates on a file descriptor; the -at suffix (a prefix would have been more consistent, but oh well) similarly indicates it operates on a (directory fd, pathname) pair.

shmat operates on dirfd/pathname?

Jan Engelhardt

#2 lore raw
Quoted 8 lines · Jan Engelhardt
Quoted 5 lines

I have noticed that the new ...at() system calls are named in what appears to be a completely haphazard fashion. In Unix system calls, an f- prefix means it operates on a file descriptor; the -at suffix (a prefix would have been more consistent, but oh well) similarly indicates it operates on a (directory fd, pathname) pair.

shmat operates on dirfd/pathname?

Convention collision. They unfortunately happen (yet another reason the -at convention was ill choosen); another pretty bad clash is the f- prefix for use on file descriptors versus use on FILE *...

-hpa

#3 lore raw
In reply to Jan Engelhardt
Quoted 8 lines · Jan Engelhardt
Quoted 7 lines

I have noticed that the new ...at() system calls are named in what appears to be a completely haphazard fashion. In Unix system calls, an f- prefix means it operates on a file descriptor; the -at suffix (a prefix would have been more consistent, but oh well) similarly indicates it operates on a (directory fd, pathname) pair.

shmat operates on dirfd/pathname?

Do you have a better proposal for naming the interfaces?

Jörg

Signature · 4 lines
 EMail:[email protected] (home) Jörg Schilling D-13353 Berlin
       [email protected]                (uni)  
       [email protected]     (work) Blog: http://schily.blogspot.com/
 URL:  http://cdrecord.berlios.de/old/private/ ftp://ftp.berlios.de/pub/schily
#4 lore raw
Quoted 10 lines · Joerg Schilling
Quoted 7 lines · Jan Engelhardt
Quoted 6 lines

I have noticed that the new ...at() system calls are named in what appears to be a completely haphazard fashion. In Unix system calls, an f- prefix means it operates on a file descriptor; the -at suffix (a prefix would have been more consistent, but oh well) similarly indicates it operates on a (directory fd, pathname) pair.

Quoted 5 lines
Quoted 1 line

shmat operates on dirfd/pathname?

Do you have a better proposal for naming the interfaces?

Isn't it obvious? Drop the misleading f- prefixes.

-hpa

#5 lore raw
In reply to Joerg Schilling
Quoted 11 lines
Quoted 8 lines
Quoted 7 lines

I have noticed that the new ...at() system calls are named in what appears to be a completely haphazard fashion. In Unix system calls, an f- prefix means it operates on a file descriptor; the -at suffix (a prefix would have been more consistent, but oh well) similarly indicates it operates on a (directory fd, pathname) pair.

shmat operates on dirfd/pathname?

Do you have a better proposal for naming the interfaces?

chownfn maybe. (fd + name)

Jan Engelhardt

#6 lore raw
Quoted 7 lines · Jan Engelhardt
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines

I have noticed that the new ...at() system calls are named in what appears to be a completely haphazard fashion. In Unix system calls, an f- prefix means it operates on a file descriptor; the -at suffix (a prefix would have been more consistent, but oh well) similarly indicates it operates on a (directory fd, pathname) pair.

Quoted 6 lines
Quoted 4 lines
Quoted 1 line

shmat operates on dirfd/pathname?

Do you have a better proposal for naming the interfaces?

chownfn maybe. (fd + name)

I am not shure if this would match the rules from the Opengroup. Solaris has these interfaces since at least 5 years.

Jörg

Signature · 4 lines
 EMail:[email protected] (home) Jörg Schilling D-13353 Berlin
       [email protected]                (uni)  
       [email protected]     (work) Blog: http://schily.blogspot.com/
 URL:  http://cdrecord.berlios.de/old/private/ ftp://ftp.berlios.de/pub/schily
#7 lore raw
Quoted 4 lines · Joerg Schilling

I am not shure if this would match the rules from the Opengroup. Solaris has these interfaces since at least 5 years.

Surely you're not suggesting that TOG's job is to rubber-stamp bad Solaris decisions...

-hpa

#8 lore raw
Quoted 8 lines · --- "H. Peter Anvin
Quoted 4 lines · Joerg Schilling

I am not shure if this would match the rules from the Opengroup. Solaris has these interfaces since at least 5 years.

Surely you're not suggesting that TOG's job is to rubber-stamp bad Solaris decisions...

No, he's suggesting that precedent should apply across unixen. It makes sense to Joerg, who programs for the lot of them like they share object inheritance. This is sometimes more problematic than others (vis a vis cdrecord). Joerg is disinclined to support the kind of compulsive re-engineering that Linus encourages, even though you might do it because it would make more sense re-engineered a certain way. If I've understood correctly (and charitably), he prefers compatibility over novelty.

I myself like functional novelty over conformant compatibility in the adiaphora.

Matt

Quoted 9 lines

-hpa
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

#9 lore raw
In reply to H. Peter Anvin
Quoted 8 lines · H. Peter Anvin
Quoted 4 lines · Joerg Schilling

I am not shure if this would match the rules from the Opengroup. Solaris has these interfaces since at least 5 years.

Surely you're not suggesting that TOG's job is to rubber-stamp bad Solaris decisions...

Are you interested in forcing Solaris to change all interfaces?

Jörg

Signature · 4 lines
 EMail:[email protected] (home) Jörg Schilling D-13353 Berlin
       [email protected]                (uni)  
       [email protected]     (work) Blog: http://schily.blogspot.com/
 URL:  http://cdrecord.berlios.de/old/private/ ftp://ftp.berlios.de/pub/schily
#10 lore raw
Quoted 5 lines · Joerg Schilling
Quoted 2 lines

Surely you're not suggesting that TOG's job is to rubber-stamp bad Solaris decisions...

Are you interested in forcing Solaris to change all interfaces?

In this case, I think it would be entirely appropriate. It's hardly a hardship to maintain the legacy names for backwards compatibility; since those names are nonsensical it is unlikely they'll ever want to be claimed, in which case they can be maintained indefinitely.

However, I would find it much, much worse if names that are *actively misleading and confusing* to the programmer would be included in an international standard and therefore forced upon the rest of the world forever.

-hpa

#11 lore raw
In reply to Joerg Schilling
Quoted 7 lines
Quoted 3 lines
Quoted 1 line

Do you have a better proposal for naming the interfaces?

chownfn maybe. (fd + name)

I am not shure if this would match the rules from the Opengroup. Solaris has these interfaces since at least 5 years.

This is not the cdrecord thread so Solaris is a no-go in this very one.

Thank you,
Jan Engelhardt

Signature · 3 lines
| Software Engineer and Linux/Unix Network Administrator
| Alphagate Systems, http://alphagate.hopto.org/
| jengelh's site, http://jengelh.hopto.org/
#12 lore raw
Quoted 9 lines · Jan Engelhardt
Quoted 6 lines
Quoted 3 lines
Quoted 1 line

Do you have a better proposal for naming the interfaces?

chownfn maybe. (fd + name)

I am not shure if this would match the rules from the Opengroup. Solaris has these interfaces since at least 5 years.

This is not the cdrecord thread so Solaris is a no-go in this very one.

FWIW, I think the -at() suffix is just fine, and well established by now (yes, there is shmat, but the SysV shared memory interfaces are bizarre to begin with -- hence POSIX shared memory which has real names.)

What I object to is the random, meaningless and misleading application of the f- suffix.

-hpa

#13 lore raw
Quoted 10 lines
Quoted 5 lines
Quoted 4 lines
Quoted 2 lines
Quoted 1 line

Do you have a better proposal for naming the interfaces?

chownfn maybe. (fd + name)

I am not shure if this would match the rules from the Opengroup. Solaris has these interfaces since at least 5 years.

This is not the cdrecord thread so Solaris is a no-go in this very one.

FWIW, I think the -at() suffix is just fine, and well established by now (yes, there is shmat, but the SysV shared memory interfaces are bizarre to begin with -- hence POSIX shared memory which has real names.)

Yep. Someday, Linux - or rather glibc! for that matter, as it is the one which translates FUNCTIONNAME() into a syscall -- will be like the Windows API. Full of compatibility stuff. And you can't do anything about it :)

Jan Engelhardt

#14 lore raw
In reply to H. Peter Anvin
Quoted 6 lines · H. Peter Anvin
Quoted 6 lines
Quoted 6 lines
Quoted 3 lines
Quoted 1 line

Do you have a better proposal for naming the interfaces?

chownfn maybe. (fd + name)

I am not shure if this would match the rules from the Opengroup. Solaris has these interfaces since at least 5 years.

...

Quoted 6 lines

FWIW, I think the -at() suffix is just fine, and well established by now (yes, there is shmat, but the SysV shared memory interfaces are bizarre to begin with -- hence POSIX shared memory which has real names.)

What I object to is the random, meaningless and misleading application of the f- suffix.

This is what I would concur.

I could live with the meaningless f- prefixes being removed for the POSIX variant of the interface.

Jörg

Signature · 4 lines
 EMail:[email protected] (home) Jörg Schilling D-13353 Berlin
       [email protected]                (uni)  
       [email protected]     (work) Blog: http://schily.blogspot.com/
 URL:  http://cdrecord.berlios.de/old/private/ ftp://ftp.berlios.de/pub/schily
#15 lore raw
In reply to Jan Engelhardt
Quoted 8 lines · Jan Engelhardt
Quoted 7 lines
Quoted 3 lines
Quoted 1 line

Do you have a better proposal for naming the interfaces?

chownfn maybe. (fd + name)

I am not shure if this would match the rules from the Opengroup. Solaris has these interfaces since at least 5 years.

This is not the cdrecord thread so Solaris is a no-go in this very one.

I encourage you to read a mail before you answer to it. This may help you to avoid writing incorrect replies.

Jörg

Signature · 4 lines
 EMail:[email protected] (home) Jörg Schilling D-13353 Berlin
       [email protected]                (uni)  
       [email protected]     (work) Blog: http://schily.blogspot.com/
 URL:  http://cdrecord.berlios.de/old/private/ ftp://ftp.berlios.de/pub/schily
#16 lore raw
In reply to H. Peter Anvin
Quoted 6 lines · H. Peter Anvin

I have noticed that the new ...at() system calls are named in what appears to be a completely haphazard fashion. In Unix system calls, an f- prefix means it operates on a file descriptor; the -at suffix (a prefix would have been more consistent, but oh well) similarly indicates it operates on a (directory fd, pathname) pair.

Quoted 8 lines

However, some system calls, in particular fchownat, futimesat, fchmodat and faccessat add the f- prefix for what appears to be absolutely no good reason. Logically, these system calls should be named chownat, utimesat, chmodat, and accessat.

I understand some of this braindamage comes from Solaris, but some of these calls do not. We should avoid it if at all possible, and I would recommend at least introducing aliases with the sane names.

This has bothered me, too.
But what would the semantics be? Using an alias named `chownat' to get lchown-like functionality (with the AT_SYMLINK_NOFOLLOW flag) seems undesirable.

If we're considering aliases,
then how about a pair of `f'-less names for each of the f*at names. E.g., chownat and lchownat corresponding to the use (or not) of AT_SYMLINK_NOFOLLOW in the last parameter of an fchownat function call.

Here's some code from coreutils/lib/openat.h:

/* Using these function names makes application code
slightly more readable than it would be with fchownat (..., 0) or fchownat (..., AT_SYMLINK_NOFOLLOW). */
static inline int
chownat (int fd, char const *file, uid_t owner, gid_t group)

  {
    return fchownat (fd, file, owner, group, 0);

}

static inline int
lchownat (int fd, char const *file, uid_t owner, gid_t group)

  {
    return fchownat (fd, file, owner, group, AT_SYMLINK_NOFOLLOW);

}

static inline int
chmodat (int fd, char const *file, mode_t mode)

  {
    return fchmodat (fd, file, mode, 0);

}

static inline int
lchmodat (int fd, char const *file, mode_t mode)

  {
    return fchmodat (fd, file, mode, AT_SYMLINK_NOFOLLOW);

}

[Note that afaik, faccessat is available only in glibc so far. ] Unfortunately, this doesn't generalize as well to faccessat, which has four (not just two) possible values of its last parameter. Those correspond to settings of the AT_SYMLINK_NOFOLLOW and AT_EACCESS bits. AT_EACCESS means faccessat checks for access by the effective IDs rather than by the real IDs.

With faccessat, we'd need 4 different names, e.g.,

elaccessat or leaccessat
ulaccessat or luaccessat
eaccessat
uaccessat

Personally, I do find it easier to read names like lstatat and statat with only one extra argument (the file descriptor) than fstatat (... with 0 or AT_SYMLINK_NOFOLLOW. With fstatat uses, it seems like I always have to convert mentally that the presence of AT_SYMLINK_NOFOLLOW means that a particular use acts like lstat.