lclint-interest message 9

Date: Mon, 7 Nov 94 13:33:37 -0500
From: evs (David Evans)
To: lclint-interest@larch.lcs.mit.edu
In-Reply-To: David Evans's message of Mon, 7 Nov 94 13:09:27 -0500
Subject: [johnm@vlibs.com: experiences with lclint]


John,

Thanks very much for the message.  Your experience sounds very
encouraging.

> PS - the only problem which occured, and i believe there really is a
> solution at hand, has to do with object creation and deletion.  a
> classic example of the original code is ... 
>
> the error that i made was not recognizing the need of adding the first
> of the *-d statements.  i could swear that i can specify the need for
> numb_init() to be called but i haven't yet figured out how.

Currently, there is no way to specify this in LCL and get it checked by
LCLint.  LCL does include a special operator "_initMod()" which
clients of a module are supposed to call before using any of the type's
other operations (including creators).  We considered checking this with
LCLint, but don't currently.  Checking that a module is initialized
before any uses would require more general inter-procedural flow
analysis, but I don't think this is nearly as useful as checking
instances (as your example indicates).

A similar construct could be added to require initialization (and
descruction) of instances.  One possible extension would be to have type
qualifiers to declare functions as "modInitializer", "insInitializer",
"insFinalizer", "modFinalizer" and then if an abstract type has any
instance initializer functions check that one of them is called before
the new instance is used (and likewise for finalizer).  

Olender and Osterweil ("Interprocedural Static Analysis of Sequencing
Constraints", ACM Transactions on Software Engineering and Methodology,
Jan 1992) have done some work on allowing programmers to specify more
general sequencing constraints using regular expressions and detecting
violations of the constraints statically.  From my own experience, I
think it is very unusal to have code which depends on sequencing
constraints other than initialization before use.  

Does anyone have any conflicting experience suggesting more general
sequencing constraints are useful in real programs?

--- Dave