[PATCH 2 of 5] verify the store

[Adrian Buehlmann]

On 31.07.2008 00:25, Matt Mackall wrote:
> On Wed, 2008-07-30 at 20:48 +0200, Adrian Buehlmann wrote:
>> +    class checkstore:
>> +        def err(self, text):
>> +            ui.warn(_(" store: %s\n") % text)
>> +            errors[0] += 1
>> +
>> +        def __init__(self, store):
>> +            self.datafiles = []
>> +            for f, size in store.datafiles(self.err):
>> +                if size > 0:
>> +                    self.datafiles.append(f)
>> +
>> +        def check(self, fl, f):
>> +            for ff in fl.files():
>> +                if not ff in self.datafiles:
>> +                    self.err(_("missing entry '%s'") % f)
> 
> This is a good example of when making something a class is overkill.

Full ack. Mostly a leftover from the variant I sent earlier this week,
which still had an orphan check, which I removed for this series.

> This happens so often that I'm tempted to say that people can't use the
> class keyword without asking me first.

That's true. But at least I got the store classes in without asking first :-)

> In this case, we've got a class that gets used exactly once to create an
> object that has a single attribute and a single method (I'm discounting
> err as it doesn't even use self!). The following is less code, is easier
> to read, and does more:
> 
> storefiles = {} 
> for f, size in store.datafiles(lambda m: err(None, m)):
>     if size > 0:
>         storefiles[f] = True
> 
> for f in files:
> ...
>     for ff in fl.files():    
> 	try:
>             del storefiles[ff]
>         except KeyError:
>             err(linkrev, "missing revlog!", ff)
> 
> for f in storefiles:
>     err(None, "orphan revlog!", f)
> 
> And we've actually already got (most of) the missing check in
> checklog(), so it's mostly the new orphan check that's interesting.

I will gladly use your code sketch for the next round.