Speeding up arch by Caching Revisions in Archives

This chapter will teach you one technique for speeding up access to an arch archive.

Consider an arch version that contains many revisions:

        mainline
        --------
        base-0
        patch-1
        ....
        patch-23
        patch-24
        patch-25
        ...
        patch-42

Suppose that a user (with no local pristine cache) wants to get the patch-42 revision. get proceeds by first getting and unpacking the base-0 revision, then getting each patch-<N> changeset, in order, and applying those to the tree.

If the list of changesets that need to be applied is long, or the sum of their sizes large in comparison to the tree side, then this implementation of get is needlessly inefficient.

One way to speed up get is by archive caching revisions -- storing "pre-built" copies of some revisions with the archive.

For example, the command:

        % tla cacherev -A lord@emf.net--2003-example \
                hello-world--mainline--0.1--patch-40

will build the patch-40 revision, package it up as a tar bundle, and store a copy of that tar bundle in the patch-40 directory of the archive.

Subsequently, a get of patch-42 will work by first fetching the cached copy of the patch-40 revision, then getting and applying the changesets for patch-41 and patch-42 : a savings of 40 changesets.

Usage Note: At this time, it's left up to you to decide which revisions to cache and which not. You might decide, for example, to automatically cache certain revisions from a cron job or to simply cache revisions by-hand whenever you notice that get is too slow. In the future, we hope to add better support for automatically caching revisions.