Re: Enhance chgrp/chown?

[Gus Heck <gu...@cognition.olin.edu>]

Stefan Bodewig wrote:
> On Wed, 21 May 2003, Gus Heck <gu...@cognition.olin.edu> wrote:
> 
>>Stefan Bodewig wrote:
>>
>>>>currently chmod has a fair number of additional features.
>>>
>>>Which are?
>>>
>>
>>dir, includes, excludes, defaultexcludes...
> 
> 
> Please don't!  These attributes come from the pre-fileset time and are
> only kept for backwards compatibility.  Please look at the special
> case in Chmod's execute that is needed to keep the old behavior of the
> dir attribute.
> 
> 

Ah glad I asked.

>>Another logicical addition to all 3 of them might be allowing
>>dirsets.
> 
> 
> I think the type="dir" attribute covers this, doesn't it?
> 

I guess multiple ways of doing things just isn't popular with anyone 
around here. Makes me wonder why we have dirsets at all if we can't use 
them here, and only in a narrow way for subant...

I suppose type="dir" is a pre-dirset feature on chmod, which I may have 
inadvertently propogated to my tasks because dirset was very new at the 
time I wrote them. (which was 4-5 months before they were commited due 
to the 1.5 release feature freezes).  These tasks are not yet released, 
so maybe they should take DirSets instead?

Granted that one would then need to supply a dirset and a fileset to get 
the "both" effect, but it would seem more consistant with the 
documentation of fileset which entirely fails to mention that 
directories are matched too. (unless you count the fact that /CVS is one 
of the default excludes).

So what is the purpose of DirSet? Maybe fileset should have had a 
type="dir|file|both" instead of having Dirsets. (of course with dirset 
released, it cant' be undone)

What direction should I be taking with chgrp/chown here?

Thinking out loud... It looks like we are not winding up with one way of 
doing things, but several ways that cannot be mixed between tasks. This 
will force the user to remember which flavor of file or directory 
specification a particular task finds tasty.

To put it another way, I think we want to make using ant as simple as 
possible. To me life is simplest when any logical answer will do, and 
failing that, simple if there is one answer that always works. The worst 
case is a hashtable of case specific answers. The hashtable is what we 
are in danger of here.

Given the constraints of back compatability, it seems that the only way 
to make life easy for the user is keep adding the new ways of 
referencing things (ie dirsets) to the old tasks in which they make 
sense. (much like has been done to chmod with filesets). This way at 
least, A dirset would always work and older ways you might or might not.

If my logic is correct, and the goal is to minimize the number of ways 
to acomplish the same thing, we should not release chgrp and chown with 
the old method (type=) but rather convert them to a DirSet accepting 
version. (and add dirsets to chmod). This change would break builds 
(like mine particularly), but hey it is an alpha version... such is life.

The pruning of things like chmod's dir atribute can be part of ant 2.0 
where back compatability is not one of our promises... Of course if 
dirset is prunable, then I have it all backwards :).

> 
>>>It has been a "core task" of Ant virtually since Ant exists, so
>>>there is no way to move it from "core" to "optional" unless we
>>>restructure the set of built-in tasks completely IMHO.
> 
> 
> To clarify.  I don't think we should move any core task to optional at
> all - until we restructure our task system completely, which I'd
> expect to go along with antlibs, yes.  There are more tasks than just
> chmod that are in core but don't really fit there (cvs for example).
> 

ok I understand. Have you found out anything about tandem/nonstop and 
chgrp chown yet?

-Gus

Re: Enhance chgrp/chown?

[Stefan Bodewig <bo...@apache.org>]

On 23 May 2003, Stefan Bodewig <bo...@apache.org> wrote:

> On Thu, 22 May 2003, Gus Heck <gu...@cognition.olin.edu> wrote:
>
>> Have you found out anything about tandem/nonstop and chgrp chown
>> yet?
> 
> No, sorry.  I'll keep you posted.

OK, Tandem supports chown and chgrp, but at the same time,
Os.isFamily("unix") will be true for it so the special case isn't
needed (and not needed in Chown either where it is going to be removed
in a few minutes).

Stefan

Re: Enhance chgrp/chown?

[Stefan Bodewig <bo...@apache.org>]

On Thu, 22 May 2003, Gus Heck <gu...@cognition.olin.edu> wrote:
> Stefan Bodewig wrote:
>> On Wed, 21 May 2003, Gus Heck <gu...@cognition.olin.edu>
>> wrote:

>>>Another logicical addition to all 3 of them might be allowing
>>>dirsets.
>> I think the type="dir" attribute covers this, doesn't it?
> 
> I guess multiple ways of doing things just isn't popular with anyone
> around here.

Uhm, Perl used to be my favorite scripting language for quite some
time.  But probably not because of TIMTOWTDI 8-)

If there is an existing way to do a thing I have no problem with
adding another way if it improves consistency accross tasks.  But
there are not that many tasks supporting <dirset> (yet?).

> Makes me wonder why we have dirsets at all if we can't use them
> here, and only in a narrow way for subant...

The main reason for <dirset> has been as a nested element in <path>.

> I suppose type="dir" is a pre-dirset feature on chmod,

Yes.

> These tasks are not yet released, so maybe they should take DirSets
> instead?

They inherit it from ExecuteOn, so they cannot un-use the attribute
(well, they could override the setter and throw an exception).

> Maybe fileset should have had a type="dir|file|both" instead of
> having Dirsets.

<fileset> always does the implict type="both" - it is the task that
decides whether it invokes getIncludedFiles or
getIncludedDirectories.  Take a look at the code of FileSet and
DirSet, they are identical.

We considered making dirsets and filesets interchangeable by having
dirset return the included directories in getIncludedFiles but decided
against it - not just because it was hackish but also because tasks
might rely on the fact that getIncludedFiles returns files and not
directories.

> What direction should I be taking with chgrp/chown here?

Thinking about it a second time - add <dirset> support to ExecuteOn.

> Given the constraints of back compatability, it seems that the only
> way to make life easy for the user is keep adding the new ways of
> referencing things (ie dirsets) to the old tasks in which they make
> sense.

True.

> ok I understand. Have you found out anything about tandem/nonstop
> and chgrp chown yet?

No, sorry.  I'll keep you posted.

Stefan