[Gnu-arch-users] Re: Creating and joining branches...

[Aaron Bentley]

Miles Bader wrote:
> > Yeah, the aba change-version script is essentially what you wrote above. 

> And the `tla-switch' script in tla-tools.  :-)

And then there's all the scripts for forking... :-)

> >  In Fai, you can do "fai update project--branch-cheese--0"

> This is worrying, because `tla update project--branch-cheese--0' does
> something quite different (and very useful in its own right).

I haven't yet seen a use case where update was better than star-merge or 
replay VERSION for the purpose.

> tla's
> interpretation is quite consistent with the way other commands in tla
> work when given a version argument, so I think it's arguably a good one.

This behavior is consistent with "tla get".  And no, the commands aren't 
all that consistent.  Somtimes they use the latest revision that has a 
patchlog in the tree.  Somtimes they use the latest revision in the 
archive.  Update does both; it's roughly

tla apply-delta $(tla logs -f |tail -n1) $(tla revisions -f|tail -n 1) 
$(tla tree-root)

> I don't know what your goals are for fai, but it seems like avoiding
> pointless incompatibilities with identically-named commands in tla would
> be a good one...

I'm trying to provide a set of commands that are useful for everyday work.

Having them behave similarly to tla commands is fine, but making them 
identical is pointless-- it's just reimplementing tla in Python on top 
of tla.

I believe the functionality "fai update" provides is useful-- the number 
of times it's been reimplemented should be proof of that.  I think that 
the behavior of "tla update" is strange:

- In a tla--devo--1.3 tree, "tla update tla--devo--1.3--patch-4" does 
something, but "tla update tla--devo--1.2--patch-115" does nothing.
- In a tla--devo--1.2 tree, "tla update tla--devo--1.3" reports "tree 
has no common history with version"
- Perhaps the merge behavior has a use case, but it's counterintuitive, 
and the use case is unusual.  On the other hand, people update all the 
time, and I believe this is more useful for that case.

I always thought of update as "turn this tree into the latest revision 
of foo, but keep my local changes intact".  I think that's what people 
want from an update command.  It took me months to find out that tla's 
update had these odd behavours in the corner cases, and I'm not the only 
one.

Aaron