Re: USE_SHA1DC is broken in pu

From: Jonathan Nieder <jrnieder@gmail.com>
Date: Mar 23, 2017, 22:22 UTC
Message-ID: <20170323222229.GB20794@aiede.mtv.corp.google.com>
In-Reply-To: <alpine.DEB.2.20.1703231724350.3767@virtualbox>

Johannes Schindelin wrote:

> On Wed, 22 Mar 2017, Jonathan Nieder wrote:
> > Johannes Schindelin wrote:

Show 13 quoted lines

>>> As to the default of seriously slowing down all SHA-1 computations:
>>> since you made that the default, at compile time, with no way to turn
>>> on the faster computation, this will have a major, negative impact.
>>> Are you really, really sure you want to do that?
>>>
>>> I thought that it was obvious that we would have at least a runtime
>>> option to lessen the load.
>>
>> It's not obvious to me.  I agree that the DC_SHA1 case can be sped up,
>> e.g. by turning off the collision detection for sha1 calculations that
>> are not part of fetching, receiving a push, or running fsck.
>
> And in those cases, using OpenSSL instead is *even* faster.

[...]

> The index is 300MB. If you have to experience a sudden drop in performance
> of `git add`, even by "only" 30%, relative to OpenSSL, it is very
> noticeable. It is painful.

[...]

> It gets even worse when it comes to fetching, let alone cloning.

[...]

Show 5 quoted lines

> And by "switching collision detection off", I of course refer to *not*
> using SHA1DC's routines at all, but what would have been used originally,
> in Git for Windows' case: (hardware-accelerated) OpenSSL.
>
> Did I manage to clarify the problem?

Yes.  Thank you for explaining.

Sincerely, Jonathan