Re: [Tcllib-devel] [TCLCORE] Github mirror broken (Tcl, Tk, Tcllib)

[Andreas K. <aku...@sh...>]

> (03) ref=$(git --bare --git-dir /home/www/state/tcl/git rev-parse HEAD)
> (04) git --bare --git-dir /home/www/state/tcl/tmp cat-file -e $ref 
> 
> Ever since Jun 18 10:48 the command (04) [cat-file] exits with status 1,
> causing the overall controller to abort the whole operation.
> 
> I suspect that the $ref is not in the repo.  If that is true then that
> might be because of a fossil change generating a different change
> set. I hope not.

Heh. I should read the code better. Just before the commands doing the
check.

    # Ensure that the new repository contains the HEAD of the old
    # repository.  If something goes wrong in the import then all the
    # commit ids get perturbed from the point of corruption on up and
    # this test will fail. If all is ok then this id will be present
    # in the new repo and we can push the new commits.

With that there is a solution to unlock ... Partial reset the peering
state (i.e. knowledge of the last uuid sync'd). Then the check is not
made.

And yes, the updated git repo will require a rebase, force pull.


Tcllib has already completed the update now, and is rebased.
Tcl and Tk will have the same. ... Completed now, yep.

>From the logs

Tcl

Tue Aug  4 22:48:04 PDT 2015 @          + 110d64c...1550256 master -> master 
(forced update)
Tue Aug  4 22:48:04 PDT 2015 @          + 80839d9...2c14a08 mistake -> mistake 
(forced update)

Tk

Tue Aug  4 23:06:25 PDT 2015 @          + 8317147...df236e2 core_8_5_branch -> 
core_8_5_branch (forced update)
Tue Aug  4 23:06:25 PDT 2015 @          + 8c7d442...4f9159f master -> master 
(forced update)

Tcllib

... Pretty much all tags and branches, looks like. Representative example:
Tue Aug  4 23:03:14 PDT 2015 @          + 3ea59cc...dd37920 master -> master 
(forced update)


 
> While waiting on responses I will now think about how to capture this
> type of problem better, i.e. make the "fx" app generate a mail to the
> maintainer of the mirror (here: me) as a proper notification when the
> issue happens instead of when somebody notes the mirror being
> out-of-date.

Done. Meaning, next time this happens I should get a mail immediately.


-- 
So long,
	Andreas Kupries <aku...@sh...>
			<http://www.purl.org/NET/akupries/>
	Developer @	<http://www.activestate.com/>

Tcl'2015, Oct 19-23, Manassas, VA, USA, http://www.tcl.tk/community/tcl2015/
-------------------------------------------------------------------------------