[Nix-dev] OpenSSL dowgrade summary

Lluís Batlle i Rossell viriketo at gmail.com
Wed Apr 28 22:57:08 CEST 2010


On Wed, Apr 28, 2010 at 05:37:35PM +0200, Marc Weber wrote:
> Today Lluís Batlle dowgraded OpenSSL because some packages don't compile
> yet and Sander found that 1.0.0 might be a bigger change than expected.
Sorry that I did not send this letter myself. At that time we were having a
discussion similar to what we had with Marc Weber last week, and we had the
trunk quite broken almost for 24h.

I remember I also took back a change of Simons some time ago, without asking the
commiter before.

In both cases I reverted changes, I think there was quite a clear reason why to
do that, and I don't think anyone should take a change revert as offensive. We
all agreed enough on a stable trunk, and it may be not easy to realise that a
change can end up breaking for too long trunk. Whether more patches would have
fixed that in five minutes, if I had not reverted the code, who knows. But hours
passed already.

I also think that reapplying the code, more revised, in a branch apart, tested by
hydra, does not take much effort. I understand that we may want the openssl
bugfix as fast as we can, but I think we do not need openssl 1.0.0 today. It can
take two or three more days of testing in a branch apart, and noone should get
nervous for that change being into trunk immediately. If I break something in
trunk, and anyone takes my change back, I could say nothing more than "sorry for
the noise", and I'll test my change again. Reverting the revert reapplies it,
and it goes quite easy and immediately in terms of telling that command to the
computer.

Marc thinks that 'svn' makes him loose time, while git does that less. Well, I
know those timings on svn or git operations, but I would never blame a program
taking 30s for a not very common operation (annotate) or merge (from trunk, to
trunk) on 'stealing my time'. I think that we can use 'svn' branches. There
are lots of people in the world using those. It does not work that bad. We
create a branch, put changes there, we review them ('Mrs. hydra' may specially
do a good review), and we merge into trunk. Maybe in this situation the
nixos.org owners can say a little words on "we are not going to change to git
because ...", to keep the someone's spirits quiet. :)

We have had those branch operations rarely until now in nixpkgs or nixos, but I
don't think anyone would complain for the branches. We can use the structure
"branches/viric/mybranch1" if we feel like creating too many branches in the
branches root; but that does not happen by now.

If I'm told (on agreement) I'd better not reverted what I reverted, in the same
way anyone can test a change in his own computer without harming others, I can
apply the revert in my own working directory. I may even create branches "trunk
without the breaking change", if you want. :)

OTH, I wasn't in the irc or I did not write to the list, because today I felt
quite ill - nothing important, but I felt heavily on cold at that time, with
enough headache, and decided to rest. I think that caused me to forget about
nix-dev or notifying the revert beyond irc, because now I'd prefer I had
written to nix-dev.

Regards,
Lluís.



More information about the nix-dev mailing list