Alex Rousskov <rouss...@measurement-factory.com> writes: > On 01/31/2013 10:08 AM, Rainer Weikusat wrote:
[...] >> I spent some time thinking how a 'nicer' cancellation subsystem with all >> desirable properties (both 'simple to use' and 'efficient to >> implement' could look like). [...] > If you decide to work on this, please use the following plan: [...] There are certain limits regarding what I can and can't do. I could have fixed the original issues (connect timeout not working/ connect retries not working) within the limits of the exising ConnOpener because these were relatively simple changes I needed to get done in any case. I started to do this based on your outline of that from last September which included the suggestion to use eventAdd for the connect timeout, something I generally consider to be The Right Thing[tm] because the connect timeout isn't really associated with any particular file descriptor. This was essentially killed by you because of 'performance concerns' regarding the event.cc implementation. But I still need this working and preferably, with a (relatively) minimal set of changes from what will eventually become the squid 3.3 release. I could also have mitigated the linked-list 'performance concerns' because this is an area where I have some expertise and it basically required one hour of hacking and four hours of debugging. While I'm still planning to do a 'bug free' (w/o known bugs) release of that for the sake of completeness once I get out of the 'other problems eating me alive state' I'm presently in, it is quite evident that this is also going nowhere. I could conceivably even justify putting a nicer interface on top of that but something I can't do is go onto a (moderately) lengthy 'clean up the mess other people put into the project' spree, especially for code I'm neither using nor plan to use: That's of no conceivable use for the company employing me[*] and most of my so-called 'spare time' goes into rather mundane tasks such as cleaning, dish washing, cooking etc. So, this is (despite I'm not entirely uninterested in that, it really sounds like a nice, little fun project) not going to happen unless 'the external circumstances' change in some drastic way. [*] "use the quick-and-dirty solution" could be regarded as the middle name of the responsible descision person.
