gnu-arch-users
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Gnu-arch-users] Re: patch: automatic cacherev and smarter get


From: Stefan Monnier
Subject: Re: [Gnu-arch-users] Re: patch: automatic cacherev and smarter get
Date: Wed, 07 Dec 2005 17:44:45 -0500
User-agent: Gnus/5.11 (Gnus v5.11) Emacs/22.0.50 (gnu/linux)

>> >> I really wish tla stopped supporting pristines and just forced the use of
>> >> a greedy revlib (which would be setup automatically in a default location,
>> >> if needed).
>> 
>> > This would be insane. You suggest that whenever someone requests a tree
>> > of 1000 files, 10Mb, tla should create million of files occupying 10Gb.
>> > [This is optimistic, since just nodes alone of typical 4Kb occupy 4Gb.]
>> 
>> Watch out: strawman argument!
>> Where did I say "dense"?

> The situation I described is likely to happen even with "sparse" revlibs,
> just not immediately. It is very easy to get a tree with 1000 files in
> any even small project (every new revision adds new files to {arch}).

I have misunderstood, sorry.  But now I don't see why this situation you're
describing is any different from the one we have now (except that now those
files are in a pristine rather than in a revlib).

> You claim (quoted) that manual setup of such massive structure as revlib
> is not needed and tla may perform it automatically.

Yes.  Even without any revlib or pristine, the checked out tree will have
those 1000 files as well.  Yes automatic setup of a revlib may choose a poor
default in some particular circumstances and you'll have to override it with
a manually setup revlib.  But the situation is just the same currently,
where for many people (see archive of this list for examples), tla seems
absurdly slow in its default setup and the answer is to setup a revlib.

So in either case some people will need to set things up manually.  My claim
is that the situation where a default revlib is a terrible choice will be
less frequent and that such situations already require manual tuning because
they are atypical (e.g. automated builder for regression testing).

> tla may expect that the tree directory is the most efficient place,
> otherwise a user would choose another filesystem to hack.

Huh?  Have you ever lived at a place where all HOMEs are on NFS?

> high speed of creation (replay, no need to rm).

I don't understand what you're referring to, here.

If you want other differences: revlibs can be placed on non-backed up
storage;  revlibs need to be flushed every once in a while.  This latter
point is the only serious drawback, really.  If revlibs are made the
default, tla will have to need with a handy way to handle it (e.g. a simple
"tla library-trim" which can be placed in a cron job or used by hand every
once in a while).


        Stefan




reply via email to

[Prev in Thread] Current Thread [Next in Thread]