Re: Working with incompetent adaists / unsafe typing war story

[Simon Wright <simon@pushface.org>]

<adaworks@sbcglobal.net> writes:

> At one point the person doing the demonstration said something such
> as, "Well let's get rid of these limited private types since they
> cannot be used for anything useful."  He then continued with his
> demonstration of the tool.

On my project, we have been using a code generator whose runtime makes
use of limited private abstract tagged types[1] where appropriate, and
doing unit tests with AUnit. AUnit ends up with a lot of unit test
code (seems to be about three times the operational code), so there's
an interest in finding a more cost-effective tool (I rather think that
what we want is a more cost-effective way of testing the code, which
is not quite the same thing, but that's another story).

At least one of the tools we looked at seemed to be incapable of
dealing with limited and/or indefinite types. I only got called in as
the eval was winding down, so I don't know the full story, but at
least one possibility is that the market for fancy test support
environments is one where limited and/or indefinite types are banned
anyway, for safety-related reasons perhaps, so it's not a problem. But
_I_ have a problem with an Ada test tool that can only handle a subset
of Ada, and I'm reluctant to modify the code generator to produce Ada
I'd be ashamed to write myself.

I _have_ been modifying the code generator to provide better support
for (AUnit-style) testing; for example, it can generate stub versions
of packages where you can store the values of in parameters, specify
the values returned via out parameters/function returns, and cause
exceptions to happen, all with indefinite values (but not limited);
implemented using streams. All easy enough if you're generating the
code in the first place, probably more work with ASIS.

[1] I think I got the order wrong there!