hg runs out of ram, fails

[Bryan O'Sullivan]

On Apr 16, 2006, at 7:00 PM, TK Soh wrote:

> On 4/17/06, Matt Mackall <mpm at selenic.com> wrote:
>>  Mercurial currently assumes that single files, indices, and  
>> manifests
>>  can fit in memory for efficiency.
>
> Just wondering... is it possible for hg to detect automatical or let
> user manually switch to other algorithm that may be less efficient,
> but still allow the work to get done?

Not in any easy way.  The assumption that files fit comfortably into  
memory is wired in all over the place.  There's not one "algorithm"  
or "the work"; there are dozens, possibly hundreds.

With some judicious use of iterators and generators, it should be  
possible to fix this up over the course of time without making a huge  
difference to code size or performance, but it's certainly not a  
small amount of work.  A nice goal would be to have streaming I/O  
used wherever possible, but this will often be at odds with clarity  
of expression.

	<b