Re: [PATCH 3/3] Make clear_commit_marks() clean harder

Hi,

On Mon, 3 Jul 2006, Junio C Hamano wrote:

Johannes Schindelin [off-list ref] writes:

quoted

quoted

quoted

Don't care if objects have been parsed or not and don't stop when we
reach a commit that is already clean -- its parents could be dirty.

There is something quite wrong with this patch.

I always had the feeling that it was wrong to traverse not-yet-parsed 
parents: How could a revision walk possibly come to a certain commit 
without at least one continuous history of now-parsed objects?

Also, AFAIK the revision walk sets flags for each commit it touched, and 
we should not try to be smart-asses about the flags, but just unset these 
flags.

The main points were made by Linus already.

Traversing is not needed -- not clearing not-yet-parsed is
obviously wrong.

Traversing is actually wrong. Clearing the marks does not mean to clear 
them on commits we did not even mark!

But clearing on commits we _have_ -- but not parsed -- is important, 
obviously.

quoted

BTW some very quick tests showed that the clear_commit_marks() thing that 
I sent to the list was much faster than traversing all objects (which was 
in my original version).

I have a crude workaround pushed out last night but will be
replacing it with something less drastic.  I think the final
version should be what you had, perhaps minus not looking at the
parsed flag for unmarking purposes.

Isn't the right way to go about it to just clear the marks if we have a 
commit that has at least one of the marks set, but traverse further only 
if _in addition to having at least one mark set_ the commit has been 
parsed already?

That would not be a crude workaround.

BTW what cases could have a commit, being seen by the revision walk, not 
have the SEEN mark set?

Ciao,
Dscho