[Cryptech Tech] New Repository Structure

[Rob Austein]

At Tue, 24 Mar 2015 09:13:39 +0100, Linus Nordberg wrote:
> 
> Rob Austein <sra at hactrn.net> wrote
> Mon, 23 Mar 2015 20:53:36 -0500:
> 
> | - Routine software updates yesterday installed a new version of git on
> |   bikeshed (uh oh).
> 
> If it's not too cumbersome to install the older version again, that
> might be an interesting experiment.

Closer examination of installation dates and repository directory
modification dates suggest that we upgraded from git 2.3.0 to git
2.3.3 two days ago (as of this message) but that the most recent
updates to any repository were before that.  So it's not that the
version of git we're running now did something weird on push.

Further, HTTPS access to the repositories doesn't involve bikeshed's
copy of git at all, it's the client git binary talking to bikeshed's
Apache.

> | - Since this is showing up via HTTPS with a known certificate, it's
> |   not stream corruption by some middlebox, the problem is at one (or
> |   both) of the endpoints.
> 
> What's terminating the TLS connection server end? Seen anything weird in
> its logs? For information, I too have reproduced the problem (using git
> 2.1.4 and curl 7.38.0).

Good question.  Nothing obvious in bikeshed's Apache log.  Will try
re-running the test while watching logs (may be a few hours, unrelated
meeting coming up soon).

Oh, and of course OpenSSL (which Apache uses) was also upgraded in the
batch two days ago.  Apache itself didn't,

I will probably move this thread over to the ticket system.