HTTP push

HTTP push

[Johannes Schindelin]

Hi,

I played a little with git-http-push. There are some strange things going 
on: whenever I commented out the calls to curl_multi_remove_handle() and 
curl_multi_cleanup(), all was well. Whenever I did not, there was either a 
segfault, or an endless loop.

Also, I had to update my curl, because http-push *absolutely* needs 
CURLOPT_HTTPGET. I tried to fake it with any combination of 
CURLOPT_{NOBODY,POST,PUT,UPLOAD}, but failed. (Yes, that means that you 
won't have to deal any longer with my patches to support older curl 
versions...)

Note: all my tests are from an iBook (client) to an Intel Linux (server).

I will run some tests on another client with valgrind (that will wait 
until tomorrow; it's 3.30am local time, so I head to bed).

Ciao,
Dscho

Re: HTTP push

[Nick Hengeveld]

On Mon, Nov 07, 2005 at 03:35:33AM +0100, Johannes Schindelin wrote:

> I played a little with git-http-push. There are some strange things going 
> on: whenever I commented out the calls to curl_multi_remove_handle() and 
> curl_multi_cleanup(), all was well. Whenever I did not, there was either a 
> segfault, or an endless loop.

http-push is a bit more aggressive than http-fetch about freeing up memory,
perhaps too aggressive...  I ran into similar issues when memory was
freed more than once.

> I will run some tests on another client with valgrind (that will wait 
> until tomorrow; it's 3.30am local time, so I head to bed).

Thanks, I'll work on memory cleanup some more.

-- 
For a successful technology, reality must take precedence over public
relations, for nature cannot be fooled.