pts@faraday.clas.Virginia.EDU (Paul T. Shannon) (03/12/91)
I posted a question here last week about what I suspected was a compiler
bug in Borland's c++ 1.0. I received two replies which were *very*
helpful, one from Wayne Turner at Tektronix, and one from Sidney
Markowitz at Borland. I've been able to thank Sidney by email, but two
messages to Wayne have bounced. I hope you see this Wayne. Thanks!
Without your message I'd probably still be poring over assembly listings
of my code and tearing my hair.
In case there's any interest, I'll briefly describe the problem and the
fix. Last, if you're persuaded that this bug is the sort of thing that
programmer's ought not to have to sweat blood to uncover -- especially
when Borland has fixed the bug, created a new version of the compiler,
and mailed it to those who knew to ask for it -- I shall ask you to
write to the product manager at Borland, and maybe we can get the policy
changed.
THE BUG: I had a huge model c++ program, all contained in one file,
which worked fine. I tested it in every way imaginable. Satisfied
that the code was stable, I then split off one function from the
original source file, the main function, so that I could test the
code as a linked .obj file. I expected that this would work easily,
and I'd then be able to link the .obj file into a larger program.
Unfortunately, the stack overflow problem arose. I changed to the
small memory model, and the problem didn't go away. (I later -- much
later -- learned that separate compilation requires a larger stack:
when I increased the stack size to 10k, the small model separate-
compilation version worked fine.) I pored over assembly listings
of the program, and ran program under turbo debugger 2.0, and
couldn't figure out the problem.
Then Wayne and Stanley came through, and my code now works again,
and is incorporated into a larger program. It doesn't work in
huge model, but the 1.01 compiler should be here in a week or two.
I called Borland for the upgrade -- as Stanley suggested -- and when
I asked why I hadn't been notified of the c++ upgrade last August,
the gentleman told me politely that I should read the trade magazines
in order to find out about upgrades. I pointed out that Borland
sends me numerous letters every year, to try and get me to buy
their business software -- which I have no need for -- and yet
they find it too expensive to send me a postcard about the compiler
upgrade. The Borland representative agreed that was a little odd.
MY REQUEST: if you think this is worth your time, please send
a letter to
Paul Gross
Sr. Product Manager, C++
Borland International
1800 Green Hills Road
Scotts Valley, CA 95066-0001
and ask that they adopt a better upgrade notification policy. I should
add that the upgrade is shipped for free, which is a commendable
policy. I also wish to state that the two Borland employees I've
dealt with -- Sidney and the fellow on the phone -- could not have
been more helpful. Now, if they could only improve their upgrade policy...
- Paul Shannon
pts@virginia.edu