Re: Solved. Odd, is it not? mkdir 'e:\' cannot be undone by rmdir 'e:\' ...

[Houder <houder at xs4all dot nl>]

Nothing new here; only correction of mistakes that I made (I decided
to review my e-mail because Ken Brwon took an interrest in the subject
matter).

On Fri, 06 Sep 2019 23:53:05, Houder  wrote:
> To those still interested! :-P
[snip]

> While I took a closer look at the source code, I found a BUG in
> path_conv::check() in winsup/cygwin/path.cc
> 
>     https://cygwin.com/ml/cygwin/2019-08/msg00418.html <==== wrong
>     ( Date: Sun, 01 Sep 2019 19:38:11 +0200 )

Correction:
https://cygwin.com/ml/cygwin/2019-09/msg00001.html

[snip]

> On September 3rd, I discovered that dropping Eric B.'s code snippet,
> would introduce a BUG:
> 
>     https://cygwin.com/ml/cygwin/2019-09/msg00015.html
>     ( Date: Tue, 03 Sep 2019 10:39:54 +0200 )
> 
> 64-@@ ln -s aap noot
> ..
> 64-@@ rmdir aap
> 64-@@ mkdir noot
> mkdir: cannot create directory ‘noot’: File exists
> 64-@@ mkdir noot/ <==== Whao! So that is what Eric indicated in his commit!
> 64-@@ ls -ld aap <==== WRONG! WRONG!
> drwxr-xr-x+ 1 Henri None 0 Sep  3 10:28 aap
> 
> Different from Posix, Linux does not allow the creation of the directory
> aap ...   (btw, neither should rmdir delete an existing directory aap if
> noot/ is specified)

Correction:
Linux is in agreement w/ Posix.
Cygwin is NOT in agreement w/ Posix (and Linux)i wrt to rmdir(2).

> While waiting for a reaction by Eric Blake, I decided to take a closer
> look at path_conv::check() ... Could a solution be found in this method?
> 
> (path arguments to (all?) commands are processed by this method)
> 
> Basically, this method consists of a 'double loop', as follows:
> 
>     for (;;) // outer loop
>       for (;;) // inner loop
> 
>  - the inner loop tests whether or not a path component is a symlnk
>  - if it is, the outer loop is reentered, where the symlnk part of
>    the path is replaced by the target
>  - finally, the algorithm bails out of both loops if a "real" path
>    is found (or not)
> 
> Or something very near to this explanation ...
> 
> In case the last component is a symlnk, the name of the symlnk is
> saved internally if the path had not been specified w/ a trailing
> slash. Otherwise the name of the target is saved internally.

Correction:
A symlnk is always followed if the pathname ends w/ a trailing slash;
if not, it depends on what the system call specified when it invoked
"path resolution" (path_conv::check() ).
If the system call specified "do not follow", "path resolution" does
not follow the symlnk (again, if path does NOT end w/ a trailing /).

> In short, there is a basic difference between specifying a path
> w/ a trailing slash or not ...

Correct! Look at how the response is different between stat final
and stat final/ in case of a symlnk.
(stat(1) basically calls lstat(2), which directs path resolution
 NOT to follow a symlnk; however that directive is ignored by path
 resolution if the pathname ends w/ a slash)

mkdir(2) and rmdir(2) are exceptions, in that these syscalls must
strip trailing slashes; they must also specify "do not follow".

The reason is, that these syscalls must not accept a symlnk as an
argument.

Henri

=====


--
Problem reports:       http://cygwin.com/problems.html
FAQ:                   http://cygwin.com/faq/
Documentation:         http://cygwin.com/docs.html
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple