Wed, 14 Feb 2001 07:41:24 -0500
On Tue, 13 Feb 2001, Michael Feldman wrote:
> >And of course there was no validation suite for C++, so these design
> >bugs slipped through more easily than they could have with Ada.
> Why is our industry so far behind in their view of something so obviously
> beneficial as a compiler validation process? This confounds me.
Well, we saw Phil Breashear's note about this - no C++ compiler could
pass the validation suite.
Only in Ada do we care seriously about this, [..]"
Even now it is deemed to be too early to expect any C++ compiler to be
conformant to the Standard. Those who care about C++ Standard conformance
aren't necessarily won over by the idea of a test suite but that is not to
say that they won't insist on being independent of their vendors.
Standard C++, which has vast standard libraries and extensions to the
language not found in traditional C++ (such as the ARM (Annotated C++
Reference Manual)), is being approached by popular implementations but
getting the new features in has left users with sludgy tools with the hope
of efficient implementations in the future.
Shortly after the Standard came to be, there were implentations which
scored conformance percentages in roughly the mid-ninties. Microsoft
Visual C++ came out as one of the worst offenders but even so scored about
92 or 93%.
"Two more ancient-history anecdotes:
(1) In the late 80s, Philippe Kahn, founder and then-CEO of Borland,
was quoted _on the record_ (I don't remember where) that Borland
would never do Ada because he would never subject his products
to someone else's validation suite."
Thank you for sharing that with us.
"(2) At about the same time, I attended an Ada BoF at SIGCSE. One of
the guys there was an Englishman who worked for BSI (their ANSI) and
said he was responsible for the ISO Pascal validation suite. He said
he respected the Ada community for their serious view of validation,
and that Borland Pascal had failed his suite miserably. And nobody
seemed to care."
Indeed Borland didn't care that Turbo Pascal wasn't ISO compliant, and
neither did its users. But do not make it out as a woefully failed attempt
by a vendor to make an ISO Pascal compiler: it never tried and the market
was completely happy. How many vendors make complete OSI stacks? None. How
many make complete TCP/IP stacks? Many. (Not a perfect complete analogy
since there were ISO Pascals around too. Strangely Free Pascal aims to be
ISO compliant AND compatible with Delphi (an object-oriented descendent
by Inprise/Borland of Turbo Pascal which is marketed as a rival to Visual
Basic, and its Linux based cousin under development is called Kylix -- yet
how many people use the multiplatform (not just for two OSes) RAD tool
Omnis Studio? Very few by comparsion I think).)
Colin Paul Gloster