[Toybox] Fixes to mkdir, chown, env

[Rob Landley]

On 09/05/2013 06:02:39 AM, Jacek Bukarewicz wrote:
>> 
>> $ touch walrus
>> $ sudo chown root.bin walrus
>> $ ls -l walrus
>> -rw-rw-r-- 1 root bin 0 Sep  5 05:16 walrus
>> 
>> It's not posix but busybox supports it too and it's used in the  
>> ./configure stage of various package builds. (Do you have a specific  
>> reason to remove it?)
>> 
> Indeed, busybox and GNU coreutils also allow dot as delimiter  
> (probably for
> historical reasons)
> I encountered this issue while testing commands that we use in our  
> environment
> and which could possibly be replaced by toybox. I tried to chown a  
> file to my
> user so I typed ./toybox chown j.bukarewicz xxx and toybox complained  
> about
> nonexisting user 'j'.

Hmmm... That's fiddly to get right.

You can't have : in a username because that's the /etc/passwd field  
delimeter.

> Busybox also fails, GNU coreutils works as expected.

Which they can pretty much only be doing by either trying multiple  
lookups or loading the entire /etc/passwd username list and comparing  
against that. Which we're trying not to do because android doesn't  
store users in /etc/password and instead has a horrible invisible  
binary database inspired by the windows registry, because that turned  
out to be such a stunningly maintainable idea...

> I wasn't aware at that point that dot is allowed as delimiter,  
> especially when
> manual or --help only say about colon.

Possibly the right thing to do is remove . but I remember it was added  
because the ./configure stage of some package build broke. Or maybe it  
was the portage build? Hmmm... There was some reason for it...

> It's nothing big. If we ever need another behavior we can probably  
> maintain
> patches locally.

I'd like to do the right thing. I'm just not sure what the right thing  
is here. Coreutils is doing something subtle and I can't guess off the  
top of my head what it is. I'd rather _not_ do something subtle, but I  
need a good test environment to show that removing a feature won't  
break stuff.

The linux from scratch plus beyond linux from scratch buidls are  
probably a reasonable test, and LFS 7.4 (just released) has a  
corresponding BLFS release for the first time since 2008.

Let me get control images implemented for those over on the aboriginal  
linux side of things, then feel free to re-raise the issue of removing  
support for . as a delimeter if I forget. :)

(Natively building sabotage under aboriginal would be good too, they've  
ot buckets of packages, but I believe their approach is "build  
coreutils, replace busybox/toybox in the $PATH, then build more stuff  
on top of it"... Doesn't prove as much as I'd like.

Thanks,

Rob
 1378747936.0