MDAY@XX.LCS.MIT.EDU (Moderator, Mark S. Day) (01/21/88)
Soft-Eng Digest Thu, 21 Jan 88 Volume 4 : Issue 4
Today's Topics:
Coordinating Software Development (2 msgs)
----------------------------------------------------------------------
Date: 11 Jan 88 17:06:59 GMT
From: mcvax!ukc!stl!stc!datlog!dlhpedg!cl@uunet.uu.net (Charles Lambert)
Subject: Coordinating Software Development
[ This is a reposting of an article I first posted two months ago. Two
correspondents finally got through to me to say my reply path was
garbage. ]
We have been developing tools and procedures (for use within our organisation)
to control those periods of intense development when many people are working
on a product at once, several of them perhaps altering the same source file
concurrently for different reasons. Now I have to produce documents and I
need good terms to make the concepts clear and easy to discuss. The
vocabulary that has grown up around the prototype is rather inconsistent
and misleading.
In my experience, well chosen terms improve people's comprehension of a
system dramatically, while fuzzy ones litter the mind with misconceptions.
So I hope that some of the technically literate out there will help me.
Let me outline what we have and ask anyone with constructive suggestions
to mail them to me.
ORGANISATION
------------
Partition
---------
The development community is divided into teams, each working on some
functional aspect of the same product. There's the first slippery term:
"product". It can mean either the final absolute object, or that object
and all the components (e.g. source files) from which it is built.
There is a hierarchy (tree, structure) of directories that will contain this
product (i.e. all the components and the final object).
There is one public copy of the above hierarchy that is fully populated;
it contains a copy of every component, each one being the correct version
to build the latest stable version of the product.
Each development team has a private instance of the directory hierarchy
in which they keep copies of the components they have altered. They do
not keep copies of components they have not altered; the building tool
is capable of finding those in the public copy. The private instance
is said to be sparsely populated.
Here we need a term for an instance of the directory hierarchy, so that we
can talk about "our ***," "your ****," or "the public ***". At the moment,
we call it a "build", after the mechanical engineering term referring to a
schedule of components. This is a bad term because
i) it is a verb doing service for a noun,
ii) we need the verb anyway ("which build did you build?"),
iii) it is not evocative - it doesn't portray what it means.
Association
-----------
Now an extra degree of complexity. Besides the public "build" and the
"builds" assigned to each functional task, a group of teams may share
an intermediate "build" where they deposit finished work that they all
want to incorporate, but is not for general release yet. So, when one
of the teams builds the product, the building tool must search a vector
of "builds" starting with the team's private one, then the shared one
and finally the public one. We have called this vector a "view" (with
acknowledgements to Erickson and Pellegrin of Bell Labs [1]).
Some of our developers have taken to referring to a "build" as a "view"
because:
i) the view is an invariant attribute of the "build" - in our
system, once you have declared the antecedence of the version
you're working on you stay with it;
ii) it rather neatly describes what a "build" is for - it is a place
from which you get an alternative view of the product.
However, we need separate terms to differentiate between an individual
"build" and the ordered summation of several "builds"; I think "view"
serves the latter purpose best.
[By the way: we rejected the term "node", as an alternative to "build",
because of its strong association with network topology and hence with
communications]
You will see that "views" converge from team "builds" through shared
"builds" to the public "build". We might use the term "branch" instead
of "build", but we would risk confusion with the terminology of the
Source Code Control System that maintains a history of changes for each
source file.
OPERATIONS
----------
Distribution
------------
When a development team needs to alter a component, they take a copy from
the first "build" in the view that contains a version of that component;
it may be copied from a shared "build" if a cooperating team has already
altered it, or it may be copied from the public build. After they have
started work on their copy, someone may alter the version from which
they started - their "antecedent" version. Eventually, the team must
combine that alteration with their own so that a unified version of the
file can be deposited in the shared or public "build".
We have called this operation "reconciliation"; clumsy - any offers?
Integration
-----------
When a team completes a functional development, it deposits the completed
work further down the view, where cooperating teams will see it. This
may involve combining their new version of a component with another version
that is already there. This is called "integration". It may appear to
be the same as reconciliation but there is an important distinction:
"reconciliation" takes place in the reconciler's "build" - several
"builds" may be "reconciled" with changes that have gone into a shared one
but the reconciled versions remain private and different from one another;
"integration" takes place further down the integrator's view where it
come within other views (and may cause other "builds" to need reconciliation).
REFERENCES
----------
[1] "Build - A Software Construction Tool" V. B. Erickson and J. F. Pellegrin;
AT&T Bell Laboratories Technical Journal, Vol. 63 No. 6 (July-August 1984)
pp. 1049-1059
-----------------------------------------------------------------------------
I will be glad to receive any pertinent criticism of the style, terminology
or clarity of the above.
Charles Lambert; cl@datlog.co.uk
------------------------------
Date: 18 Jan 88 05:18:15 GMT
From: tektronix!orca!stank@ucbvax.Berkeley.EDU (Stan Kalinowski)
Subject: Coordinating Software Development
Let me start off by making an observation. One can tell that the
science of software engineering is still in its infancy. It seems
that as much time is spent defining new software engineering terms for
existing concepts as is spent on solving the engineering problems.
Why not use the existing software engineering terms? Perhaps I'm
overly critical, but it seems that every new researcher sees the field
of software engineering as an opportunity and tries to glean a bit of
immortality by being the first to coin the term (substitute your
favorite buzzword here). I find it difficult to believe that so many
people are simply unaware that terms already exist to describe most
software activities.
By the way Mr. Lambert, I'm not necessarily picking on you personally,
but hoo-boy, you scored a direct hit on one of my hot buttons! Now
that I have dispensed with the knee-jerk reaction, perhaps it was just
a jerk reaction :-), let me describe the jargon my working group would
use for what you describe, perhaps it will help.
In article <368@dlhpedg.co.uk> cl@datlog.co.uk (Charles Lambert) writes:
.
. (text was trimmed and refilled for brevity but it's still long.)
.
>We have been developing tools and procedures (for use within our
>organisation) to control those periods of intense development when
>many people are working on a product at once, several of them
>perhaps altering the same source file concurrently for different
>reasons. Now I have to produce documents and I need good terms to
>make the concepts clear and easy to discuss. The vocabulary that
>has grown up around the prototype is rather inconsistent and
>misleading. In my experience, well chosen terms improve people's
>comprehension of a system dramatically, while fuzzy ones litter the
>mind with misconceptions. So I hope that some of the technically
>literate out there will help me.
It sounds like you are attempting to describe your "development
control" scheme to people that are not familiar with software jargon.
I think that inventing new terms would probably compound your problem.
I think it would be better to explain the existing terminology rather
than create an additional learning barrier. Even if your audience is
familiar with the commonly used software nomenclature, it probably
wouldn't hurt to define your terms in an appendix. (I think new terms
are invented by people too lazy to learn how to use the existing ones,
probably a fallout of the "software cannot be re-used mentality", but
that's just my opinion.)
>The development community is divided into teams, each working on
>some functional aspect of the same product. There's the first
>slippery term: "product". It can mean either the final absolute
>object, or that object and all the components (e.g. source files)
>from which it is built.
Around here, we call the thing we sell to customers a "product" and
the stuff that engineers produce is called software. If we want to
distinguish software that is used in our "product" as opposed to
software tools, we use the phrase "product software".
>There is a hierarchy (tree, structure) of directories that will
>contain this product (i.e. all the components and the final object).
We call that a "build tree" or, more specifically, a "build
directory-tree". (I don't know if Merriam Webster would approve of
using dual attributive nouns like "build directory", but
what-the-heck, it works.)
>There is one public copy of the above hierarchy that is fully
>populated; it contains a copy of every component, each one being the
>correct version to build the latest stable version of the product.
We would call that the "release" source directory.
>Each development team has a private instance of the directory
>hierarchy in which they keep copies of the components they have
>altered. They do not keep copies of components they have not
>altered; the building tool is capable of finding those in the public
>copy. The private instance is said to be sparsely populated.
We call the private instances "private build-trees" or "experimental
build-trees".
>Here we need a term for an instance of the directory hierarchy, so
>that we can talk about "our ***," "your ****," or "the public ***".
>At the moment, we call it a "build", after the mechanical
>engineering term referring to a schedule of components. This is a
>bad term because
>
>i) it is a verb doing service for a noun,
>ii) we need the verb anyway ("which build did you build?"),
>iii) it is not evocative - it doesn't portray what it means.
I can see where the confusion could arise. We also use the term
"build" to refer to the compiled/assembled/linked system, but we call
the "schedule of materials" a "bill of materials" or, more briefly,
"BOM". This expression comes from manufacturing terminology. To me,
it evokes the image of software modules specified in the BOM being
pulled together on an assembly line, to create a given "build" of the
product. We do not think of the directory tree itself as being the
bill of materials, we maintain a separate file that lists what's in
the build.
The term "build" is used often around here and people seem to have no
trouble gathering the meaning from the context in which its used. I
avoid cases like number ii) above, I would probably be more specific
and say "Which configuration did you make?" When I what to know what
sources or compilation switches were used in the build, people usually
respond with a product codename as an answer.
>Now an extra degree of complexity. Besides the public "build" and
>the "builds" assigned to each functional task, a group of teams may
>share an intermediate "build" where they deposit finished work that
>they all want to incorporate, but is not for general release yet.
We use the word "release" to indicate a build that must be coordinated
with other groups. If we need to be more descriptive we can tack on
something like "evaluation" or "beta test" to help describe the
intended use of the build. We also assign a single revision/version
identifier to identify the build as a whole, this makes tracking the
builds simple.
>So, when one of the teams builds the product, the building tool must
>search a vector of "builds" starting with the team's private one,
>then the shared one and finally the public one. We have called this
>vector a "view" (with acknowledgements to Erickson and Pellegrin of
>Bell Labs [1]).
It seems to me that the term "view" is just as bad (if not worse) than
the term "build". The word "view" also has several meanings. It can
be what one sees from the top of a mountain, it can be the transformed
image found on a graphics display, or it can be a projection of an
object on an engineering drawing. At least with the term "build" you
have the manufacturing analogy, which I think *IS* evocative.
>Some of our developers have taken to referring to a "build" as a "view"
>because:
>i) the view is an invariant attribute of the "build" - in our
> system, once you have declared the antecedence of the version
> you're working on you stay with it;
>ii) it rather neatly describes what a "build" is for - it is a place
> from which you get an alternative view of the product.
I don't see the analogy supporting i) at all, and the support for ii)
is weak at best. By the way, we avoid maintaining multiple, actively
used, versions of a given module's codee whenever possible. We have
found that it is very difficult to propagate bug fixes common to
several product lines to each product without having to "un-fix" some
bugs that are unique to a product. Also, maintaining functional
compatibility is made more difficult by parallel versions of source
code. We choose, instead, to maintain one code base for the products
(where possible) and do product version control via conditional
compilation. (If you have a better answer to this problem, I'd like
to hear it.)
>When a development team needs to alter a component, they take a copy
>from the first "build" in the view that contains a version of that
>component; it may be copied from a shared "build" if a cooperating
>team has already altered it, or it may be copied from the public
>build. After they have started work on their copy, someone may
>alter the version from which they started - their "antecedent"
>version. Eventually, the team must combine that alteration with
>their own so that a unified version of the file can be deposited in
>the shared or public "build".
>
>We have called this operation "reconciliation"; clumsy - any offers?
Yeah, call it "programmer's hell". In my experience, communication
generally breaks down between between parallel code developers. This
is usually due to schedule demands placed on the engineer. Even when
communication is mandated by management through the use of
inter-project design reviews, the reviewer often does not realize the
impact the proposed changes will have. This frequently results in
expensive redesign and rework when the "reconciliation" happens.
Until I find a better answer, I think a single designer should
maintain a given module for all products. This also forces the
designer to consider inter-product functional
compatibility/consistancy, which is VERY important these days.
Well, I've probably insulted just about everybody on the net by now,
but I feel better. :-) I hope what I had to say helps someone out
there avoid what I learned the hard way. (It probably won't though,
most people would have hit "N" as soon as they saw "long" in the
subject line. They probably said something like "Oh God, not another
one of those software engineering nut's ranting again.".
--
US Mail: Stan Kalinowski, Tektronix, Inc.
Information Display Group, Graphics Workstations Division
PO Box 1000, MS 61-028, Wilsonville OR 97070 Phone:(503)-685-2458
uucp: {ucbvax,decvax,ihnp4,allegra,uw-beaver}!tektronix!orca!stank
------------------------------
End of Soft-Eng Digest
******************************
-------