[Monotone-devel] Re: monotone CVS import failed.

[Markus Schiltknecht]

Jon Smirl wrote:
> > The filename seems obvious, how else would I know what file a cvs_event
> > is bound to? And the CVS revision is kept for informative purpose: I
> > want users to be able to determine, what CVS revisions a certain
> > monotone revision consists of.
>
> You still have a C++ object for the file, just no need to store the file 
> name.

Huh? What do you mean by 'store'? On disk? What does that have to do 
with memory consumption?

> > I very much doubt it's more than 100MB, though. It's roughly:
> >
> > (different files + different version strings) * (avg str length) +
> > sizeof(int) * (total cvs events) * 2
>
> 110,000 files with 1.1M revs. 1,100 branches and 1,000 symbols.

Assuming 1.1 M revs is the amount of 'total cvs events', as I'm naming 
it, that's roughly:

(110'000 * 2) * (avg str length) +
4 * 1.1M * 2

So your avg. string length may be up to ca. 420 bytes and it still all 
fits into 100 MB of RAM.

> > Assuming the (avg str length) is 1K (!), you could theoretically store
> > more than 52K *distinct* filenames and over 13M pointers (thus over 6 M
> > total cvs events) in 100 MB of memory (assuming 50% - 50% sharing
> > between the two summands).
>
> The other systems identify the rev by file name + rev#, when you do
> that over 1M revs it really gets big.

Sure.

Look: I'm storing two pointers, one for the filename and one for the RCS 
version string. That's 8 bytes per cvs event (every commit in every 
file). Plus every filename and RCS version string exactly *once*, so 
these pointers can point somewhere meaningful. That avoids storing 
filenames and RCS version strings multiple times.

Regards

Markus