Re: re:gcc test results question

[Phil Edwards <phil at codesourcery dot com>]

On Thu, Jul 08, 2004 at 07:48:54PM -0400, David P. Riedel wrote:
> let me rephrase my question then.
> 
> > Depends on what those tests are testing, and when they started failing.

I'll restate my answer then, because it hasn't changed.  :-)


> > Depends on what those tests are testing,

If the tests in question are testing oddball corner cases of the language,
or GNU extensions, or for that matter anything which you don't use or
don't care about, then the fact that the tests fail shouldn't affect your
decision whether or not to use the compiler.

My car does a horrible job when underwater.  Since I don't ever use my
car underwater, I don't care, and this failure of the car's design did
not enter into my decision to purchase and drive the car.

Joe's just pointed out that most of the tests are for a GNU extension
which many projects don't use.  If you're in the majority, you don't care.
If you're in the minority which use them, then it's a fatal bug.


> > and when they started failing.

If the bugs being tested have always been there, then you probably don't
care about those either; you've probably devised workarounds by now.
(Not whether the test cases have always failed, but whether the bug being
tested has always been there.  Testcases usually aren't added until the
bug is found.)


Phil

-- 
AI.cpp:33241: warning: You wrote 'neurons.merge(solution1, solution2)",
AI.cpp:33241: you probably MEANT "neurons->merge(solution1, solution2)",
AI.cpp:33241: but there is a MUCH better way to implement this whole
AI.cpp:33241: function; doing that instead.