Re: hg clone from anonhg issues

[Joerg Sonnenberger <joerg%bec.de@localhost>]

On Tue, Jul 28, 2020 at 09:50:45AM -0700, Jeff Rizzo wrote:
> This is NOT intended to kindle another firestorm of complaints; I'm trying
> to provide info that's hopefully useful in resolving the underlying
> usability problems.  Anyone actively working on solving this is welcome to
> contact me directly if I can provide any useful help, but I'm not interested
> in a flame war or general gripe session.

This is a combination of two issues. The bundle contains a couple of
sections that are processed in order. After the file data section, hg
recomputes the roots of all phases and then continues processing the
remaining sections. Now the phase computation itself is essentially
linear time, but made much more expensive for stupid reasons. For hg
5.5, it will be something like four times faster. Now, why is this a
problem? The delay hits a timeout from fastly ("Between bytes send") and
the server disconnects. I'm a bit puzzled why we still hit a timeout
here as it was supposed to have been raised. Testing with
"curl $url; sleep 480; fg" and hitting C-z after a few seconds, I can
trigger the disconnect reliable here. I'm still contemplating whether I
need to hack around this by essentially doing a steady background fetch,
but then I would prefer to just fix the real problems first.

Joerg