Happy Passover or Easter weekend to you all.

RJLRef: $CASE/bdelog2ks/BugClassification.2k0421

In response to a question from Jeff:

On Apr 20, no class was held - bde still needs testing, 
	and bde bugs need a classification scheme.
This classification is a tree with a top-level split 
into build (static/compile) and run (inspect/trace/execute) types:

	0: build 
	1: run 

Build-time bugs MAY have been fixed on alphas but still exist 
on decstations.  (But don't worry about the latter unless your 
familiariaty with OSF/Ultrix library version and environment
differences can help my remus build process. :-)

Myers: The Art of SW Testing, and Humphrey on testing recommend
inspection and bug-classifying  checklists which partition runtime 
bugs according to the semantic type of the source code error.


The root of the runtime branch of the bug classification tree 
can be divided into computation vs. data structure problems : 

    0: build (make/compile) (Precondition to run-time testing)
    1: run   (inspect/test)
        1.1 Computation (local)
            1.1.1.* compound block type (loop, conditional, sequence)
            1.1.2.* I/O 	(file/record/field level?, error type)
        1.2 Data Structures (non-local)
            1.2.1.* persistent data type  (pfkey relations?) 
            1.2.2.* internal data type    (pointers?)


Computation problems tend to be simpler to fix (e.g., they can often 
be localized to one assignment target (NULL Ptr?) or expression).
Compilers can report Warnings that deserve attention as run-time risks;
Code inspectons can find bugs and either fix them before testing;
compiler warnings can be exploited to focus tests and assertions 
on likely sources of problms; pay attention to the ways these warnings 
suggest potential risks.

Data/state problems are harder to fix, precisely because errors 
in static or persistent data can be introduced at anytime before 
they are first detected, hence the value of pre- and post-condition 
assertions, refactoring  and ghost variables to expedite detection.
Code inspections or enabled assertions may detect the insertion times
of these runtime errors without tracing backward from the crash.

So I expect you to classify bugs by describing the detection 
and repair processes you employed, not just about the type of 
syntax or semantic error that was responsible. 

Quantitative measurements, like identifiying
time-of-detection and method of diagnosing and labor to fix the bug,
are important metrics to improve quality because information 
on cost-benefit rations is needed to direct improvement effort.
 

Now that bde builds OK on alphas (but not on decstations) it's time for 
more test runs that crash bde and follow-up runs that trace the cause[s].
Each of you should become proficient at both; from your
reports we may be able to aggregate enough bugs and fixes to 
see how our bugs distribute over the classes above.  

Note that fixing the bug is a perequisite to classifying it by its cause.


>> Date: Thu, 20 Apr 2000 17:07:10 -0400 (EDT)
>> To: Bob Lechner <lechner@cs.uml.edu>
>> cc: 2ks523
>> Subject: Master build is fine.
>> 
>> I just did the master build. It looks fine.
>> make -e | tee result
>> 
>> The result of the build is redirected to
>> /nfs/jupiter/proj3/case/bdelog2ks/she/bde/result
>> 
>> I can not go to class today because I have monday's class to go.
>> 
>> Shuiqing He (Scott He)
>> 
>> 

 Scott's build result looks clean, but it didn't re-run genv11,
 He also used the default error/warnings reporting level.
 To aid in finding suspicious code, warnings should be turned on.
 Here is -Wall at man gcc line ~863:
---------------
  -Wall
       All of the above `-W' options combined.  These  are  all  the  options
       which  pertain to usage that we recommend avoiding and that we believe
       is easy to avoid, even in conjunction with macros.
---------------

 A  complete rebuild should run genv11 also. Here are the steps to do that:
 The pr_util/Makefile does this if the schema.h file changes.
------------- 
 [tern1](62)> cat ~lechner/bde2alpha_rjl/buildbde
 # do this in bde's parent directory: - RJL 2k0420
 xmkmf -a 
 cd bde
 # do this to run genv11 again:
 touch schema/*.sch
 # do this so g++ links instead of cxx:
 setenv CXX g++
 make -e
----------- 
 ==========================================================
 
 You should also do make clean before you build on a 
 different (decstation) platform and on return  to alphas.
 Right now I am stymied by a problem with Imake rules for
 decstations where printenv says: OSTYPE=ultrix, MACHTYPE=mipsel.
 The Imakefile contains #ifdef UltrixArchitecture dependent code.

 I tried xmkmf -a on remus:
 
 xmkmf -a on remus  first needed <standards.h>,
 whch I copied from jupiter. 
 Then xmkmf -a complained as follows 
 about bde/src/Imakefile (but not about pr_util/Imakefile)
 
 -------------
  cd src; imake -DUseInstalled -Dnostdinc      -I/usr/lib/X11/config
  -I/usr/lib/Xm             -DTOPDIR=../. -DCURDIR=./src;  make  Makefiles
  Make: Must be a separator on rules line 453.  Stop.
  *** Error code 1
 ------------