Re: [PATCH 3/4] --decorate now decorates ancestors, too

Hi,

On Thu, 12 Jul 2007, Junio C Hamano wrote:

Theodore Tso [off-list ref] writes:

quoted

On Wed, Jul 11, 2007 at 02:29:49AM +0100, Johannes Schindelin wrote:

quoted

The option --decorate changed default behavior: Earlier, it decorated
commits pointed to by any ref.  The new behavior is this: decorate the
with the given refs and its ancestors, i.e.

	git log --decorate next master

will show "next", "next^", "next~2", ..., "master", "master^", ...
in parenthesis after the commit name.

I'm wondering how useful the default is.  The arguments get used for
two things; both for git-log to decide what revisions to display, and
which refs to decorate, right?  I'm not sure that overloading is such
a great idea.

Also, I note that "git log --decorate" does nothing at all.  Maybe it
would be better to keep the default to be "any-ref" instead of "given"?

I think defaulting to "given" is a regression.  It could be
argued that "tag-ref" or "tag" might be a better default
(judging from my experience with "name-rev"), but keeping
"any-ref" would probably be the safest.

Okay, fair enough.  I kind of expected people to disagree as of the 
default mode.  My preference would have been "tag", since I need that 
quite often, but I am willing to put my wishes aside there.

Probably I should follow up with a replacement patch, and a config 
variable "commit.decorateMode" patch, leaving the default at "any-ref"?

But in general I do not see ("I haven't realized" might turn out to be a 
better expression) much value in this series yet except for the initial 
clean-up patches, while I think this option would be quite expensive in 
terms of memory footprints on projects with nontrivial size of history.  
I dunno.

There are a few points I want to make to convince you:

- I need this quite often to see which version introduced a certain 
  feature.  This is most visible on IRC, where people ask "how can I do X, 
  I have version Y", and me responding "There is a feature P, but 
  unfortunately, it is only available from revision Q~N" where Q > Y.

  I really want other people to do this easily, without having to know how 
  name-rev (which has a dash in its name, and thus is kind of 
  plumbing-ish) works.


- It is _not_ expensive.  It only ever allocates something in the case of 
  merges (at least that's how I designed it), and should free the used 
  memory when the commit name is not used at all (but maybe I forgot that 
  part...).  So you will have at most one allocation of a few bytes per 
  merge, and I really do not see how this could break down for 
  pathological, but real-world, cases.

- 40-character commit names seem to be really hard on people.

To drive the 3rd point home: I often see people confused as to what those 
long commit names are.  They do not even realise that they are _object_ 
names, actually, not only applicable to commit objects.

IMHO a great way to teach users that commit names are equivalent to 
shortcuts like "v1.5.0-rc1~20" would be to even enable decoration by tags 
by default.

Ciao,
Dscho