On Fri, 11 Apr 2008 15:40:50 +0200 Henk Langeveld wrote:
> What kind of version control is used inside at&t research?
> What do you use to keep track of bugs and changes in the AST code?

it varies by group/individual
for ast there are 3 main at&t contributors
I own the ast build / packaging
all changes filter through me

each leaf src dir controlled by a Makefile contains these files

    RELEASE
        one or two line description of changes/fixes in
        reverse chronological order, part of the src packages

    OFFICIAL
        a file containing diff -u for each file change, in
        chronological order, not part of the src packages

    local src tarballs
        local INIT and ast (internal ast that includes ast-open)
        src tarballs, roughly every 3 months, containing
        complete src snapshots; these snapshots are internal
        and do not contain copyright notice comment headers
        (those are generated by the external src tarball generator)

OFFICIAL is a byproduct of of a viewpath build
source changes are done in the top view, typically

    .../arch/$HOSTTYPE/src/...

an initial

    nmake install
    nmake test

for one group of changes on the master build machine is done
to verify compatibility with the master source

when master source compatibility is confirmed I check that the RELEASE
file says something about the change(s), and then I do

    nmake official

to do the ``diff -u ... >> OFFICIAL''

and then a build from the top src dir is unleashed on all machines
in our local build farm, ~30 builds, with one command, that builds
all ast on every machine

the first part of the farm build uses the ast ditto(1) script to
sync master and remote machine source -- this is controlled by
the package(1) script -- it handles groups of machines that
share the same physical src dir, and uses tw(1) and pax(1) vdelta
to minimize the remote data transfer

an up to date build of all ast on a ~3Ghz machine, NFS mounts, takes ~30s

``nmake test'', on ~20 machines (no patience for dinosaur machines that
take hours instead of minutes to test), is run to check for regressions

_______________________________________________
ast-users mailing list
[email protected]
https://mailman.research.att.com/mailman/listinfo/ast-users

Reply via email to