Memory Consumption for 'bzr co' [was Re: Work flow on large repositories]

[Andrew Bennetts]

John Arbash Meinel wrote:
[...]

> And finally, back to CHKInventory (excluding the chk_bytes repository
> objects)
> >>> summarize_refs(chki, excluding=[chk_bytes.address])
> Total 1231512 objects, 16 types, Total size = 107.6MiB (112783517 bytes)
>  Index   Count   %      Size   % Cum     Max Kind
>      0  654339  53  80479605  71  71     536 str
>      1   51514   4  15394232  13  85 1573004 dict
>      2   65581   5   4721832   4  89      72 InventoryFile
>      3   69639   5   3744212   3  92     208 unicode
>      4  186732  15   3266268   2  95      20 StaticTuple
>      5   47122   3   2827320   2  97      60 LeafNode
>      6  152186  12   1826232   1  99      12 int
>      7    4058   0    503192   0  99     124 CHKInventoryDirectory
> 
> So 65k files, and 4k directories, or roughly 70k total entries.

Coincidentally, I noticed tonight that various InventoryEntry subclasses had
much larger __slots__ than necessary.  I've just made a patch that (on 32-bit
systems) trims InventoryFile from 68 to 40 bytes, and CHKInventoryDirectory from
120 to 48 bytes.  That's only 2MB of that 107, but it's a cheap win.

The patch is
<https://code.launchpad.net/~spiv/bzr/smaller-inventory-entries/+merge/31636>.

-Andrew.