Should rm_work be moved after do_deploy, or is ${WORKDIR} not available in do_deploy?

[Rod Whitby <rod@whitby.id.au>]

[Follow-ups to openembedded-devel@lists.openembedded.org only, please.]

A question for the OE core team representatives on the list:  Is
do_deploy prohibited from being able to get at things in ${WORKDIR}, or
should rm_work happen after do_deploy?

Details follow:

There is this neat rm_work.bbclass in OpenEmbedded which removes each
individual build/tmp/work/... directory as each package is successfully
built (this saves a *lot* of space, especially with the openmoko svnnow
paradigm which will quickly fill your disk with multiple build copies of
openmoko packages).

Unfortunately (at least with bitbake 1.6 head), rm_work is called before
do_deploy, so the do_deploy() function in OpenMoko's
sjf2410-linux-native.bb fails:

do_deploy() {
        install -d ${DEPLOY_DIR_IMAGE}
        install -m 0755 sjf2410 ${DEPLOY_DIR_IMAGE}/sjf2410
}

The sjf2410 executable no longer exists at that point, cause rm_work has
deleted it.

Should rm_work be moved to after do_deploy, or should
sjf2410-linux-native.bb (and all other packages that assume that the
contents of ${WORKDIR} are still available in do_deploy) be fixed?

(For example, the same problem occurs in ixp4xx-kernel ...)

-- Rod