[fpc-devel] RIP NoGlobals

[Hans-Peter Diettrich]

Jonas Maebe schrieb:

>>> As far as allocated memory is concerned: yes. It does free a bunch of 
>>> stuff when an error occurs, but not everything, and what is not freed 
>>> depends on the error.
>>
>> Ok. Thanks.
>> And I guess there are currently no plans to fix this, right?
> 
> No, because it would be lots of work (writing test programs that trigger 
> all possible error conditions and testing/debugging them one by one, or 
> implementing some kind of mark/release system) with no real payoff 
> except if you're an IDE developer that wants to integrate the compiler.

IMO you simply don't understand the reason for my NoGlobals refactoring, 
or deny its use for (assumed) performance degradation. Right?


>> So the proper way to integrate FPC is to run it as separate process or 
>> in a dyn lib with its own memory manager. Correct, or ?
>> This means, in order to share some caches an IDE must use some 
>> IPC/shared memory. Right?
> 
> Probably, I haven't really thought about that.

And this may be were a (possible) performance gain in the compiler can 
result in an even higher performance degradation, when the compiler is 
used by other applications. Overall performance may be increased as 
well, when parts of Make are added to the compiler itself. Then e.g. a 
simple flag could indicate which files or units can not be used without 
recompilation, instead of a long-running chain of file removes in 
advance (make clean).

DoDi