Re: [Gnu-arch-users] missing feature: text file handling

[John Meinel]

Aaron Bentley wrote:
> John Meinel wrote:
>
> > Well, I ran into a problem. precommit doesn't happen soon enough. I 
> > can change the file permissions, and they are change in the 
> > filesystem, but the changeset has already been committed. 
[...]
> Does this behaviour make sense to people?  True, it provides a chance to 
> edit the changeset, but you'd think people would rather have a chance to 
> edit the tree.  I guess we could also add a pre-precommit hook.
>
> > I also have to remember, that in a hardlinked tree, you can't change 
> > the mode of a file since it is still linked.
>
>
> But if it's already hardlinked, it's already committed.  So its 
> permissions are probably right.

This is true for all times except the *first* time you start using the 
hook. So for me that means when I go through and clean up my tree, it's 
a manual process. If it's only once, it's probably okay. As long as I 
detect it, and thus avoid corrupting the revision library (without it's 
knowledge).

> > What is the correct way to break a hard link? Copy the file, and then 
> > move it over the old one?
>
>
> That's right.
>
> Aaron

I would like to see a pre-precommit hook. I think both are useful, if 
only for different things. As you mentioned precommit could be used to 
modify the changeset, and pre-precommit can be used to verify the tree. 
Another possible hook time would be for pre-"changes". Something that 
lets you clean up a tree before it decides to compute the changes.

Also, if we wanted a hook-level for doing changeset patch munging (for 
CR/LF -> LF stuff), then we would need a hook somewhere around get 
changeset to allow you to "un-munge" on checkout.

I'll respond later, but I think using a different diff & patch is a 
better solution than munging the changeset.

John
=:->

signature.asc
Description: OpenPGP digital signature