"git fetch" performance regression

[Jonathan Nieder]

[Impact]

Updating git from lucid to precise brought in the change

  6b67e0dc068d fetch: verify we have everything we need before updating our ref

which caused a speed regression:

 http://thread.gmane.org/gmane.comp.version-control.git/191626/focus=193228

This has been preventing some users from updating the git package to the version
in precise.

The fix is d21c463d558a "fetch/receive: remove over-pessimistic connectivity check".

I believe it falls in the category

 "Bugs which do not fit under above categories, but (1) have an obviously safe
  patch and (2) affect an application rather than critical infrastructure
  packages (like X.org or the kernel)."

as the impact of this regression fix would be isolated to git alone.

[Test Case]

"time git fetch". See the link above for details.

[Regression Potential]

Barring toolchain breakage, this fix should not produce a regression. It is a
simple patch that makes sense and has been well tested upstream.

With toolchain breakage, anything can happen, naturally. The git command could
stop working completely or subtly misbehave. Git has an extensive test suite so
this would be most likely to result in a build failure, but it's also possible
that the build and basic manual testing would not catch a problem, in which case
it would show up in user reports.

Git repositories are by the nature of how they are used almost always replicated,
limiting potential data loss in ordinary cases to changes that have not been
pushed yet.

[Other Info]

This is my first SRU request. I admit I'm a bit surprised at how quickly it seems
to have labelled as invalid.

[Phillip Susi (psusi)]

Not in precise, no. See http://wiki.ubuntu.com/StableReleaseUpdates.

[Jonathan Nieder (jrnieder)]

How about just

 d21c463d558a fetch/receive: remove over-pessimistic connectivity check

then? It is obviously safe and affects the git toolset, not critical infrastructure.

[Phillip Susi (psusi)]

Unless the performance is so bad as to make it practically unusable, then no, it doesn't justify an SRU.

[Jonathan Nieder (jrnieder)]

According to the description presented in the web UI, Invalid means "Not a bug. May be a support request or spam." Perhaps you misread the request or meant "won't fix".

I've updated the description to explain how it relates to the SRU guidelines.

Thanks for looking it over,
Jonathan

[Anders Kaseorg (andersk)]

Hey Phillip, in case this isn’t clear, Jonathan is the Debian maintainer of Git. He might or might not be familiar with the details of Ubuntu policy, but he generally knows what he’s talking about, so let’s not be so hasty with the “Invalid” hammer.