Transport w/ extra tests

Aaron Bentley at
Mon Jul 18 22:51:37 BST 2005

Hash: SHA1

John A Meinel wrote:
> I also started playing around with pipelining http access using
> But in my testing, it wasn't much faster (1m20 instead of 1m50).

In my tests, parallel downloads were noticeably faster:
9m3.351s vs. 14m13.682.  I would expect pipelined to be even faster than

> I may not have been doing it correctly. I also had a problem because of
> the "GzipConsumer" class. I think your Http server recognizes if a file
> is already gzipped, and you would unzip it as you pulled it across. But
> since a ImmutableStore (CompressedTextStore) expects things to be
> GZip'd, it would get very confused.

Yes.  The web server is misconfigured-- it reports
content-coding: gzip for gzipped files, but content-coding is supposed
to mean 'compressed for transport', not 'this file happens to be a
compressed file'.  I monkey-patched GzipFile into oblivion to fix that
in my effbot version.

> I was actually thinking more along the lines of:
> bzr selftest RsyncTransportTest
> Which would skip all tests that did not match the string RsyncTransportTest.
> That way you don't have to modify any tests, just change how you invoke
> "bzr selftest".

Yes, something like that would be very nice.

Version: GnuPG v1.4.1 (GNU/Linux)
Comment: Using GnuPG with Thunderbird -


More information about the bazaar mailing list