pwd output changed with coreutils upgrade from 5.2.1-4 to 5.3.0-2

Max Bowsher
Wed Feb 23 14:19:00 GMT 2005 wrote:
>> Hi all,
>> I'm using the pwd.exe command from a standard winXP command prompt
>> ( for building Mozilla ).
>> However, when I upgrade to coreutils 5.3.0-2 I get this weird result:
>> C:\devel\lgo\mozilla>pwd
>> pwd: reading directory `../../../..'
>> Even more strange, is that when I execute pwd from the command prompt, I 
>> hear
>> my floppy disk being accessed.
>> Can somebody help me out on this? Is this expected behaviour?
> I can reproduce the bug.  It is because coreutils switched in 5.3.0 to 
> using
> opendir("..")/readdir()/closedir() to find the child directory in .. with 
> the
> same inode as stat(".").  However, using the readdir() sequence on
> "/cygdrive":
> a) scans all Windows drives, including floppies (hence the sound of your 
> disk
> access)
> b) tickles a cygwin bug where closedir() incorrectly returns -1 without
> setting errno (hence the cryptic error message from pwd)
> The bug never bit me in testing because I tested from mount points other 
> than
> /cygdrive.
> Also, there is a known bug in coreutils,
> that
> pwd(1) is not POSIX compliant because it does not have a -L option that
> defaults to reading $PWD rather than assuming -P and scanning inodes.
> Until such time as a cygwin snapshot fixes closedir() to work correctly
> (hopefully by cygwin 1.5.13), or such time as I release a coreutils with a
> compiled in hack to ignore the return of closedir if errno is unchanged, 
> you
> have two workarounds:
> a) rather than using /bin/pwd, use the builtin pwd of /bin/sh or /bin/bash
> b) use mount(1) to ensure that your files can be accessed without having 
> to
> use /cygdrive as their prefix

Wouldn't it be far more appropriate to hold back 5.3.0 as a test package, 
leaving 5.2.1 as current until such time as 5.3.0 can be considered ready?

In fact, unless you think that is a bad idea, I'll go ahead and change 
setup.hint accordingly.
Waiting an hour for possible reply....


Unsubscribe info:
Problem reports:

More information about the Cygwin mailing list