check reports spurious unreferenced texts

[Matthew Fuller]

check output from below script shows unreferenced text versions where there shouldn't be any:

     2 revisions
     2 file-ids
     2 unique file texts
     1 repeated file texts
     2 unreferenced text versions

#!/bin/sh -x
bzr="/home/fullermd/src/bzr/bzr.dev/bzr --no-aliases"

${bzr} init A
(
        cd A ;
        touch foo ;
        ${bzr} add ;
        ${bzr} ci -m '1' ;
        touch bar ;
        ${bzr} add ;
        ${bzr} ci -m '2' ;
        ${bzr} check
)

[John A Meinel (jameinel)]

Isn't this a duplicate of a bug that Alexander reported?

I know I've already written up a simple patch which fixes this. (no tests, etc)

Basically, I think when Andrew refactored the check code he changed it so that VersionedFileChecker was called with the complete list of revision ids in the repository. Rather than just the versions that were present in the knit file itself.

as a rough example

vfc.check_file_verison_parents(a_knit, file_id, repo.all_revision_ids())
versus
vfc.check_file_verison_parents(a_knit, file_id, a_knit.versions())

The former one gives an unreferenced text version for every version present in all_revision_ids() that is not present in a_knit.versions().

The patch should be at least on the mailing list, if not already associated with a bug.

[Robert Collins (lifeless)]

On Mon, 2007-11-26 at 20:26 +0000, John A Meinel wrote:
> Isn't this a duplicate of a bug that Alexander reported?
>
> I know I've already written up a simple patch which fixes this. (no
> tests, etc)
>
> Basically, I think when Andrew refactored the check code he changed it
> so that VersionedFileChecker was called with the complete list of
> revision ids in the repository. Rather than just the versions that were
> present in the knit file itself.
>
> as a rough example
>
> vfc.check_file_verison_parents(a_knit, file_id, repo.all_revision_ids())
> versus
> vfc.check_file_verison_parents(a_knit, file_id, a_knit.versions())
>
> The former one gives an unreferenced text version for every version
> present in all_revision_ids() that is not present in a_knit.versions().
>
> The patch should be at least on the mailing list, if not already
> associated with a bug.

I have replaced the guts of this part of check; so possibly it is the
same, but possibly it isn't :).

-Rob
--
GPG key available at: <http://www.robertcollins.net/keys.txt>.