Re: BUG in "cvs co -r" (CVS 1.11.0)?

[Larry Jones]

Greg A. Woods writes:
> 
> So, therefore the first and most important question to ask Alexei is
> whether or not there was some other filesystem object called "Attic"
> before the manual mkdir of the new empty Attic directory.  Only if not
> then could there may be a bug in the OS.

He strongly implied there was not -- particularly since he was able to
create the Attic directory by hand, which he would not have been able to
do if there were some other object with that name.

> existence_error() is defined with this rather ugly but no doubt more
> portable mess:
[...]
> (why can't POSIX-compatible systems stick to the basics in basic
> situations like this?!?!?!??!  having to go to such lengths for the sake
> of POSIX portability is stupid)

I think the mess is due to systems that try to be Unix-like without
actually conforming to POSIX.  As far as I can see, ENOTEXIST, EOS2ERR,
and EVMSERR (which are the cause of the mess) are definitely not part of
POSIX.  (I can guess where EOS2ERR and EVMSERR came from, does anyone
know where ENOTEXIST came from?)

> On most systems opendir() calls open(2), which says:
> 
>      [ENOTDIR]     A component of the path prefix is not a directory.
> 
> suggesting again that the thing called "Attic" was not a directory in
> Alexei's repository (until after it was manually made to be one).

POSIX (or at least SUS2, which is all I have on-line access to) is
unfortunately a bit fuzzy in this area.  opendir() says:

    [ENOENT]    A component of dirname does not name an existing
                directory or dirname is an empty string. 

    [ENOTDIR]   A component of dirname is not a directory.   

which seems to leave it ambiguous as to which is returned for a non-
existent directory.  Of course, we all know that ENOENT is the right
choice.

> If not then Alexei's operating system does have a serious bug or is at
> least not very POSIX compatible.

I've taken a quick look at the cygwin source (which is what Alexei is
using) and it appears that opendir() has a pechant for returning ENOTDIR
for nearly all errors whereas ENOENT would probably be a better choice
in most cases.

-Larry Jones

I'm so disappointed. -- Calvin