[Gnu-arch-users] Re: Design proposal to kill pristine trees

[Pau Aliagas]

On 8 Oct 2003, Miles Bader wrote:

> I'm confused as to why you're keying the revlib location on archive 
> name.

Because you need that location to add the revisions to the library if you 
don't want to ask it every time. The same happens to add temporary revlib 
entries (what now are pristine trees).

> As you've seen from my postings, there are various ways I'd like to
> arrange my revlibs/pristine trees, but which archive the revisions come
> from rarely seems like a useful way to categorize things.  In most cases
> it's way too _narrow_ a focus, because I use stuff from many libraries
> which I'd like to use the same path for -- so having the specify it anew
> for each new archive seems like a needless annoyance.  In other cases
> it's too broad -- if I have several project trees from the same archive
> located on different devices, I'd like the revlibs to be on the same
> device as the project tree, not forced into a single location.

You are describing how to look for a revlib entry and giving an algorithm
for it. What I'm trying to do first is to store those revlib entries
somewhere and not asking again. If you have multiple paths, how do you do
it? Fix it per archive.

> I'd much rather have tla just invoke a hook so that I can use my _own_
> algorithm for locating revlibs (e.g. one of the scripts I posted
> earlier).  tla can cache the path returned, so this needn't be
> inefficient, and it seems far more useful than the simple algorithm
> you've described.  [Who is that algorithm intended for?  I'd think most
> people would either want maxium control or maximum simplicity, and it
> offers neither.]

We are in different stages: you already take for granted that these 
revisions are in place; I want to put them on place automatically, without 
user intervention but to define its location the fist time, if the default 
does not fit him.

For revision libraries, explicitly added, we could override the default 
location with an option, so maybe we could overcome it. But for 
automatically created revlibs (ex pristine trees), we cannot do this.

> The question of course is what should happen for people that _don't_
> supply a hook?  As far as I can see, a single static path is probably
> reasonable for that case.

Let's focus first the discussion on the storge and then we'll go through 
the other issues.

Pau