Subclassing notify

[Benoit Boissinot]

On Fri, Apr 03, 2009 at 07:17:05PM +0200, Ulrich Pfeifer wrote:
> Hello,
> 
> I tried to subclass notify to overwrite the part displaying a single
> change set in the incoming hook:
> 
>       else:
>           count = 1
>           n.node(node)
> -        n.diff(node, node)
> +        n.diff_single(node)
>       data = ui.popbuffer()
>       n.send(node, count, data)
> 
> were diff_single differs from diff by this:
> 
> -    def diff(self, node, ref):
> +    def diff_single(self, node):
>           maxdiff = int(self.ui.config('notify', 'maxdiff', 300))
>           prev = self.repo.changelog.parents(node)[0]
> 
> -        chunks = patch.diff(self.repo, prev, ref, 
> opts=patch.diffopts(self.ui))
> +        chunks = patch.diff(self.repo, prev, node, 
> opts=patch.diffopts(self.ui))
>           difflines = ''.join(chunks).splitlines()
> 
> +        if len(self.repo.changelog.parents(node)) > 1:
> +            right = self.repo.changelog.parents(node)[1]
> +            chunks = patch.diff(self.repo, right, node, 
> opts=patch.diffopts(self.ui))
> +            difflines_right = ''.join(chunks).splitlines()
> +            self.ui.debug(_('\nleft=%s right=%s\nLeft %d lines, right 
> %d lines\n') % (short(prev), short(right), len(difflines), 
> len(difflines_right)))
> +            if len(difflines_right) < len(difflines):
> +                difflines = difflines_right
> +
>           if self.ui.configbool('notify', 'diffstat', True):
>               s = patch.diffstat(difflines)
>               # s may be nil, don't include the header if it is
>               if s:
> 
> Basically in the merge case this is emitting the shorter diff.  This
> is far from perfect - but a start of what I wanted to do.
> 
> My problem now is, that with a fairly big repository (manifest has
> about 65mb) the hg process bloats to above 750mb main memory.

The diff to the other parent can be really huge, can you check if this
is a leak (is the memory recovered when exiting the function).

Otherwise you could try:
- an heuristic based on the DAG (how far are the parents from the least
  common ancestor)
- use the size of the binary diff instead of building a "real" textual
  diff.

regards,

Benoit
 

-- 
:wq