[Gnu-arch-users] Re: implicit discussion

[Miles Bader]

Martin Pool <address@hidden> writes:
> > > [and of course you can also just always `mv' always and then use a tool
> > > like `tla-update-ids' before committing.]
> > 
> > Hmmm.  How does that work?  How can tla-update-ids differentiate a moved
> > file from a deleted one?
> 
> I think it notices that an identical file has been added under a
> different name.

It actually uses `most similar with less than N% change' rule (using
diff to do the comparison).  This allow renames to be detected even if
the file was also modified a bit, which is a pretty common occurance.

> > > If your tree is chock-full of binary files which you keep moving around,
> > > I guess you're going to have get in the habit of using `tla mv' (or some
> > > other tool), but taglines don't cramp that style.  Might as well use
> > > taglines anyway for their other benefits.
> > 
> > What other benefits?
> 
> Putting them in doesn't stop you using tla mv, and it accomodates
> people who do like them.

Some advantages of taglines (besides the generally more user friendly
mv/rm behavior):

 (1) With the current TLA, they're faster

 (2) Don't bloat up your disk with tons of small-files-that-the-file-
     system-makes-large (and even if _you_ use reiserfs or whatever,
     consider your other developers -- some of who might be stuck using
     ext2).

 (3) Fewer magic files polluting your source tree.  I know, some people
     feel just the opposite: they feel that the taglines are `pollution',
     and prefer the magic files.  To some degree this will always be a
     religious issue, but I find I _far_ prefer a small comment at the end
     of source files -- which you basically never see -- to lots of little
     files.

 (4) Changes can be tracked even in external distributions that deleted
     all the arch-specific files

It obvious depends on just exactly your tree looks like -- if you have 90%
binary files, taglines will be an exception rather than the rule, so you'd
probably be better off just using explicit for everything -- but having
used them for quite a while now, I really, really, like them; AFAIC,
they're quite a bit better than the bad old cvs-style.

BTW, points 1 - 3 are to some degree dependent on the current
implementation of explicit tags used in tla; for instance, changing to a
representation that used a single file per directory to hold explicit tags
would improve things quite a bit.  However there doesn't appear to be any
rush to change the implementation.

-Miles
-- 
We live, as we dream -- alone....