A possible future purpose for D1 [rant]

[Fawzi Mohamed]

On 2010-03-02 09:25:05 +0100, Walter Bright <newshound1 at digitalmars.com> said:

> Fawzi Mohamed wrote:
>> that s good, but maybe for a release one should also try to compile 
>> some of the largish projects that are done in D (even al older frozen 
>> version) to see if in larger codebases something comes up...
>> At least for D 1.0 on a fixed system the idea "if id did compile it 
>> should compile again" is something that could be considered.
>> You could ask people to make a script "setup my program" that does it 
>> with the dmd in the path.
>> This could be done for some releases (stable ones), yes bugs found this 
>> way are harder to isolate, but it could be worthwhile.
> 
> I generally don't do that because:
> 
> 1. I didn't write the code, and when it fails to compile it can be 
> rather time consuming because I don't know what it is intended to be 
> doing.
> 
> 2. They often don't come with a test suite, and simply compiling it 
> doesn't really say if it works or not. If it does come with a test 
> suite, and the test suite fails, debugging someone else's code can be 
> very hard.
> 
> 3. It makes running the test suite take considerably longer. I need it 
> to be fast, as I run it very often when developing things.
> 
> 4. Nearly all problems boil down to less than 10 lines of code. This 
> makes for a very compact and fast running test suite. If the test suite 
> fails, the problem is usually already isolated down.
> 
> 5. I've discovered over the years that programmers write in particular 
> "islands" of the language. No matter how large a code base they 
> produce, they never stray outside that island, so once the bugs they 
> initially encountered are fixed, they never run into compiler bugs 
> anymore. The coverage of a test suite is simply not a function of 
> number of lines of code thrown at it.
> 
> 
> In other words, large applications tend to make lousy test suites. A 
> test suite needs to be written to be a test suite.
> 
> 
> On the other hand, anyone can subscribe to the dmd-beta mailing list, 
> and have a chance to check for regressions on their own code before the 
> release. It's the point of the list.

I did not know about dmd-beta, and have now subscribed to it.
I must have missed its announcement, as I don't always read all the 
posts, on the NG, I just check it out from time to time.
It is definitely a good idea, a way for people not involved in compiler 
development to quickly, and possibly painlessly check if the new 
release breaks something.
I fully agree that large codebases aren't a good testsuite for compiler 
development, but they are a good test for beta releases.
As you say it is indeed better if the developer of the lib/app does the 
test, and it scales better, so dmd-beta is the right idea.

Fawzi