Re: lifetime-9 test FAIL

Stefan Seefeld <[email protected]> Tue, 18 Jul 2006 11:03:36 -0400
Newsgroups gmane.comp.documentation.synopsis
Message-ID <[email protected]>
Gilles J. Seguin wrote:

> The testsuite is in bad state, means we have some work to do.
> Would like to have it in better shape to test improvement.
> The email was to raise the point.

OK. Those 'Cxx' tests aren't actually useful for the stable branch,
as they test code that isn't really used there anyway (symbol lookup,
mostly).
For the devel branch I'm considering how to change the overall testing
strategy as the current approach doesn't really work either.
Consider this example:

int i;
{
  int j = i;
  int i;
}

The current testing approach still suffers the multi-pass syndrome,
i.e. when the SymbolLookup test reports what 'i' refers to in the
assignment expression, it will report the local 'i', not the global
one, since at the point the test looks up the symbol, the symbol
table represents the state after parsing the whole thing has completed.

But, since I don't even have an internal representation right now
that remembers such references, there is no alternative. Ultimately the
AST might grow so it can represent arbitrary statements and expressions,
too. But we aren't there yet.

Let me know if you have ideas what to work on. And if you haven't,
I'm online again on IRC. :-)

Regards,
		Stefan

-- 

      ...ich hab' noch einen Koffer in Berlin...