[MERGE][RFC] Compact origin information for *.knit (Draft)

[Dmitry Vasiliev]

John Arbash Meinel wrote:
> Dmitry Vasiliev wrote:
>> The changes are not so big and moreover some new *.knit files are even
>> bigger than the old ones (because gzip can't compress them better). So
>> currently I'm even not sure about the proposed idea. Maybe I just need
>> to leave only the tests/optimizations/fixes? What do you think?
> 
> I was expecting to not see a large post-compression change. Especially
> since the most common annotation revision id is going to be the version
> at the start of the hunk, gzip will already have a string that it can
> compress all of the lines.

Actually I've expected more benefits for lowering repository size, but the
result didn't meet my expectations.

> What I'm more curious about is to change the
> annotation lines to be:
[skipped some optimization ideas]

Moreover we can also mimic full encode/decode behavior inside the knit 
factories. I don't know if it's worth it yet.

> So I like having more tests, and I'm not opposed to allowing blank
> spaces in the annotations, but I don't think it gains us a lot. Unless
> you can show that the extraction performance for a large file is much
> faster.
> 
> So I would say +0.5. I don't think it is worth introducing a new
> repository format for a small gain, but the rest of your stuff looks
> pretty good.

Ok, I'll plan to forget about the new repository format and concentrate more on 
possible speed optimizations.

-- 
Dmitry Vasiliev (dima at hlabs.spb.ru)
     http://hlabs.spb.ru