Re: [PATCH] revision.c: drop missing objects from cmdline

[Junio C Hamano <gitster@pobox.com>]

Matthew DeVore <matvore@google.com> writes:

> No code which reads cmdline in struct rev_info can handle NULL objects
> in cmdline.rev[i].item, so stop adding them to the cmdline.rev array.

"The code is not prepared to have cmdline.rev[].item that is NULL"
is something everybody would understand and agree with, but that
does not automatically lead to "so ignoring or rejecting and dying
is OK", though.  The cmdline thing is used for the commands to learn
the end-user intent that cannot be learned by the resulting objects
in the object array (e.g. the user may have said 'master' but the
pending[] (and later revs.commits) would only have the object names,
and some callers would want to know if it was a branch name, a
refname refs/heads/master, or the hexadecimal object name), so
unless absolutely needed, I'm hesitant to take a change that loses
information (e.g. the user named this object that is not locally
available, we cannot afford to add it to the pending[] and add it to
revs.commits to traverse from there, but we still want to know what
object was given by the user).

> Objects in cmdline are NULL when the given object is promisor and
> --exclude-promisor-objects is enabled.

A "promisor" is a remote repository.  It promises certain objects
that you do not have are later retrievable from it.  The way you can
see if the promisor promised to later give you an object is to see
if that missing object is reachable from an object in a packfile the
promisor gave you earlier.  

"The given object" is never a "promisor", so I am not sure what the
above wants to say.  Is 

    When an object is given on the command line and if it is missing
    from the local repository, add_rev_cmdline() receives NULL in
    its "item" parameter.

what you meant?  Is that the _only_ case in which "item" could be
NULL, or is it also true for any missing object due to repository
corruption?