I'm trying to optimize the semantic analysis pass in Descent. The idea is to not load things up if they are not needed. For example:

---
module one;

import two;

void someMethod(Foo foo) {
  foo.x = 1;
}
---
module two;

import three;

class Foo {
  int x;
  Bar bar;
}
---
module three;

class Bar { }
---

Now, if I want to help the user by giving autocompletion, go to definition, etc., in module *one*, you can see that property "bar" is not even mentioned there, so module "three" would not be loaded, not even looked up.

I've managed to do this (DMD's compiler doesn't care about this because it needs to create an object file for the whole thing). The problem is that this optimization gives a semantic error when you use a fully qualified name for something that is privately imported in an imported module (a very known bug, I think). Although reporting this as error is good, because it is "how it should be", there are a lot of existing modules that depend on this functionality, so seeing an error there may be annoying.

I can't "fix" it because if I do it, I loose the optimization (module "three" would have to be loaded in this example, at least to preserve DMD's semantic). So I think I'll leave it like that (with the optimization on), but I wanted to know what did the community think about this.

I also wonder if it helps if I report these erorrs now marked by Descent into bugzilla...

Ary Borenszweig wrote:
> I'm trying to optimize the semantic analysis pass in Descent. The idea is to not load things up if they are not needed. For example:
> 
> ---
> module one;
> 
> import two;
> 
> void someMethod(Foo foo) {
>   foo.x = 1;
> }
> ---
> module two;
> 
> import three;
> 
> class Foo {
>   int x;
>   Bar bar;
> }
> ---
> module three;
> 
> class Bar { }
> ---
> 
> Now, if I want to help the user by giving autocompletion, go to definition, etc., in module *one*, you can see that property "bar" is not even mentioned there, so module "three" would not be loaded, not even looked up.
> 
> I've managed to do this (DMD's compiler doesn't care about this because it needs to create an object file for the whole thing). The problem is that this optimization gives a semantic error when you use a fully qualified name for something that is privately imported in an imported module (a very known bug, I think). Although reporting this as error is good, because it is "how it should be", there are a lot of existing modules that depend on this functionality, so seeing an error there may be annoying.
> 
> I can't "fix" it because if I do it, I loose the optimization (module "three" would have to be loaded in this example, at least to preserve DMD's semantic). So I think I'll leave it like that (with the optimization on), but I wanted to know what did the community think about this.
> 
> I also wonder if it helps if I report these erorrs now marked by Descent into bugzilla...

Perhaps if possible you could show these errors in some other way than the usual way, some other color or something like that. Perhaps it could be an option you could turn off if possible.

Ary Borenszweig wrote:
> I'm trying to optimize the semantic analysis pass in Descent. The idea is to not load things up if they are not needed. For example:
> 
> ---
> module one;
> 
> import two;
> 
> void someMethod(Foo foo) {
>   foo.x = 1;
> }
> ---
> module two;
> 
> import three;
> 
> class Foo {
>   int x;
>   Bar bar;
> }
> ---
> module three;
> 
> class Bar { }
> ---
> 
> Now, if I want to help the user by giving autocompletion, go to definition, etc., in module *one*, you can see that property "bar" is not even mentioned there, so module "three" would not be loaded, not even looked up.
> 
> I've managed to do this (DMD's compiler doesn't care about this because it needs to create an object file for the whole thing). The problem is that this optimization gives a semantic error when you use a fully qualified name for something that is privately imported in an imported module (a very known bug, I think). Although reporting this as error is good, because it is "how it should be", there are a lot of existing modules that depend on this functionality, so seeing an error there may be annoying.
> 
> I can't "fix" it because if I do it, I loose the optimization (module "three" would have to be loaded in this example, at least to preserve DMD's semantic). So I think I'll leave it like that (with the optimization on), but I wanted to know what did the community think about this.
> 
> I also wonder if it helps if I report these erorrs now marked by Descent into bugzilla...

You're talking about bug #313, right? (http://d.puremagic.com/issues/show_bug.cgi?id=313)
I think it's clearly a bug, there are even testcases for it, so the IDE should threat it as a bug, and any such illegal uses should be marked as errors in the IDE.

You said "there are a lot of existing modules that depend on this functionality". Such as? I would find that odd, modules depending on that bug.

-- 
Bruno Medeiros - Software Developer, MSc. in CS/E graduate
http://www.prowiki.org/wiki4d/wiki.cgi?BrunoMedeiros#D

Bruno Medeiros a écrit :
> Ary Borenszweig wrote:
>> I've managed to do this (DMD's compiler doesn't care about this because it needs to create an object file for the whole thing). The problem is that this optimization gives a semantic error when you use a fully qualified name for something that is privately imported in an imported module (a very known bug, I think). Although reporting this as error is good, because it is "how it should be", there are a lot of existing modules that depend on this functionality, so seeing an error there may be annoying.
> 
> You're talking about bug #313, right? (http://d.puremagic.com/issues/show_bug.cgi?id=313)
> I think it's clearly a bug, there are even testcases for it, so the IDE should threat it as a bug, and any such illegal uses should be marked as errors in the IDE.
> 
> You said "there are a lot of existing modules that depend on this functionality". Such as? I would find that odd, modules depending on that bug.
> 

Like... a lot of modules in phobos. But I think I got rid of that problem when I tweaked a little the way I optimized things (so the bug is back :-P).

Ary Borenszweig wrote:

> Bruno Medeiros a écrit :
>> Ary Borenszweig wrote:
>>> I've managed to do this (DMD's compiler doesn't care about this because it needs to create an object file for the whole thing). The problem is that this optimization gives a semantic error when you use a fully qualified name for something that is privately imported in an imported module (a very known bug, I think). Although reporting this as error is good, because it is "how it should be", there are a lot of existing modules that depend on this functionality, so seeing an error there may be annoying.
>> 
>> You're talking about bug #313, right?
>> (http://d.puremagic.com/issues/show_bug.cgi?id=313)
>> I think it's clearly a bug, there are even testcases for it, so the IDE
>> should threat it as a bug, and any such illegal uses should be marked as
>> errors in the IDE.
>> 
>> You said "there are a lot of existing modules that depend on this functionality". Such as? I would find that odd, modules depending on that bug.
>> 
> 
> Like... a lot of modules in phobos. But I think I got rid of that problem when I tweaked a little the way I optimized things (so the bug is back :-P).

Please mark it as a coding error/bug if I do something like that while coding in Descent :)