Re: How to get Ada to ?cross the chasm??

[Niklas Holsti <niklas.holsti@tidorum.invalid>]

On 18-05-04 01:27 , Randy Brukardt wrote:

   [snip]

> There is no realistic chance that Ada will
> ever be changed enough to make it amenable to GC.

Ouch and wow, when did that direction change?

As I remember it, for a long time the mantra was that Ada (83, 95, ...) 
was compatible with GC, but vendors did not implement GC because there 
was no immediate demand from the current users (chicken and egg situation).

I also remember that some people successfully used the "conservative" GC 
tools, such as the Boehm GC, with ordinary Ada programs and Ada 
compilers (one problem being that for arrays, some compilers used 
"offset pointers" pointing to the location of the hypothetical element 
at index zero, even if the actual first index was not zero). Were these 
people misguided? Perhaps this way before controlled types and Finalize 
operations were introduced? Seems to me that a vanilla Boehm GC would 
not be able to call the right Finalize before discarding an object.

The Ada implementations targeting the JVM or .NET were/are claimed to 
use the GC provided by the target. Were/are these somehow "non-Ada" 
implementations?

> I proposed such changes
> multiple times in the past and got nowhere with them. (The problem being
> that the location of finalization of most objects is too late for any useful
> GC to be possible -- it's generally associated with the access type, which
> is usually declared at the library level. Thus GC can't clean up most
> objects until the type goes away -- which is usually when the program
> exits.)

So the issue is possible side-effects of user-defined Finalize 
operations, and when those side-effects are allowed to happen? If so, 
I'm surprised that defining the timing of such side effects was 
considered more important than being compatible with an implementation 
using GC. Too bad.

Usually GC systems are criticized for finalizing some objects too late. 
Funny that an Ada GC would be damned for doing it too early...

-- 
Niklas Holsti
Tidorum Ltd
niklas holsti tidorum fi
       .      @       .