Re: We need an internal keyword.
On Monday, 22 October 2018 at 11:06:42 UTC, Jonathan M Davis wrote: On Monday, October 22, 2018 2:30:21 AM MDT Basile B. via Digitalmars-d wrote: On Monday, 22 October 2018 at 08:25:17 UTC, Andrea Fontana wrote: > Moreover: you're the author of the module so you're supposed > to know how it works and which members you should call or > not. - team - maintainer of a module written by someone that works elsewhere now. that's two cases where strict privacy can be optionally a thing avoiding wrong usage of private members within the scope of a module. Part of the point is that if the module is large enough that the folks working on the code can't actually keep track of what's in it, then it's too large, and as such, if you need to protect your class or struct members from the rest of the module, then it really should be in a separate module for the code to be properly maintainable anyway. Yes, having multiple people involved makes the problem worse (especially when some of them join the team later), but it doesn't fundamentally change the issue. If it changes anything, it simply makes the argument stronger for preferring smaller modules so that they're easier to digest. Personally, I've found that larger modules have worked just fine for me without having to worry about these sort of encapsulation issues. It simply isn't a problem, and I don't recall ever seeing a bug because of it. But anyone who's worried about it always has the option of simply going for smaller modules, and the encapsulation problem is already solved without making the language any more complicated. Plenty of folks already think that it's best practice te prefer relatively small modules anyway, and if you need a way to protect your private member variables from the module when the module isn't large, then you're definitely doing something wrong. Given the D philosophy that the module is the primary unit of encapsulation and that you should at least roughly understand the entire module when working on it (or it's almost certainly too large), having an access level to protect member variables from the rest of the module simply makes no sense. Anyone who feels the need for such an access level think about what they're doing with their code and why they feel the need for it - whether it's simply because they're used to it from other languages, or because they're organizing their code in a manner which is detrimental to maintainability. - Jonathan M Davis Here is the intial dip draft: https://github.com/12345swordy/DIPs/tree/Encapsulation If Walter Bright insist that the module is the unit of encapsulation then I propose we get rid of the "one module per file" restriction, by introducing sub modules. -Alex
Re: We need an internal keyword.
On Monday, October 22, 2018 2:30:21 AM MDT Basile B. via Digitalmars-d wrote: > On Monday, 22 October 2018 at 08:25:17 UTC, Andrea Fontana wrote: > > Moreover: you're the author of the module so you're supposed to > > know how it works and which members you should call or not. > > - team > - maintainer of a module written by someone that works elsewhere > now. > > that's two cases where strict privacy can be optionally a thing > avoiding wrong usage of private members within the scope of a > module. Part of the point is that if the module is large enough that the folks working on the code can't actually keep track of what's in it, then it's too large, and as such, if you need to protect your class or struct members from the rest of the module, then it really should be in a separate module for the code to be properly maintainable anyway. Yes, having multiple people involved makes the problem worse (especially when some of them join the team later), but it doesn't fundamentally change the issue. If it changes anything, it simply makes the argument stronger for preferring smaller modules so that they're easier to digest. Personally, I've found that larger modules have worked just fine for me without having to worry about these sort of encapsulation issues. It simply isn't a problem, and I don't recall ever seeing a bug because of it. But anyone who's worried about it always has the option of simply going for smaller modules, and the encapsulation problem is already solved without making the language any more complicated. Plenty of folks already think that it's best practice te prefer relatively small modules anyway, and if you need a way to protect your private member variables from the module when the module isn't large, then you're definitely doing something wrong. Given the D philosophy that the module is the primary unit of encapsulation and that you should at least roughly understand the entire module when working on it (or it's almost certainly too large), having an access level to protect member variables from the rest of the module simply makes no sense. Anyone who feels the need for such an access level think about what they're doing with their code and why they feel the need for it - whether it's simply because they're used to it from other languages, or because they're organizing their code in a manner which is detrimental to maintainability. - Jonathan M Davis
Re: We need an internal keyword.
On Monday, 22 October 2018 at 08:25:17 UTC, Andrea Fontana wrote: Moreover: you're the author of the module so you're supposed to know how it works and which members you should call or not. - team - maintainer of a module written by someone that works elsewhere now. that's two cases where strict privacy can be optionally a thing avoiding wrong usage of private members within the scope of a module.
Re: We need an internal keyword.
On Sunday, 21 October 2018 at 03:17:23 UTC, 12345swordy wrote:
So that classes can share some of their variables but not
others in a module.
IE.
class A
{
internal int A; //This is shared in the module
private int B; // But not this.
}
No need to reintroduce the "Friend" feature from cpp.
As just said by others, if you need it probably your module is
too big and you have to split.
Moreover: you're the author of the module so you're supposed to
know how it works and which members you should call or not.
Anyway since they're private you can name them in particular way
to remember yourself not to use them eg: int internal_A;
Andrea
Re: We need an internal keyword.
On Sunday, 21 October 2018 at 23:50:57 UTC, 12345swordy wrote: If the cost out way the benefits then I simply introduce the "strict" keyword to avoid code breakage, or introduce the optional module scoping. -Alex Looking at the dlang.org page about visibility attributes, the `package` keyword can have an argument specifying which package has access to the symbol. What if `private` could have an argument dictating which other symbol could have access to that private symbol ? Setting this argument to the private symbol's own class would effectively make it strictly private. This wouldn't require a new keyword, and that syntax is not valid D code so it wouldn't break any existing code. Though it would look a bit weird, and it would also basically introduce a sort of "friend" feature. I don't know, that was just a random thought that crossed my mind just now.
Re: We need an internal keyword.
On Sunday, 21 October 2018 at 23:50:57 UTC, 12345swordy wrote: On Sunday, 21 October 2018 at 21:48:22 UTC, Laurent Tréguier wrote: On Sunday, 21 October 2018 at 17:09:05 UTC, 12345swordy wrote: [...] It's not "my" solution. It's D's solution. I perfectly understand why you'd want this and I would probably make use of a private/internal difference myself if it was available. If you already know about this solution however, I don't even know why you're starting this thread; since changing the behavior of private would be a major language change breaking tons of existing codebases, plus it would require adding yet another keyword. Given that this conversation has happened before and things haven't changed, I'm very doubtful that it could happen at any point in time, sadly. If the cost out way the benefits then I simply introduce the "strict" keyword to avoid code breakage, or introduce the optional module scoping. -Alex Technically, introducing any new keyword is also a potentially code breaking change. Any symbol named "strict" would have to be changed. This is why I'm doubtful that such a change would be accepted.
Re: We need an internal keyword.
On 10/20/18 11:17 PM, 12345swordy wrote:
So that classes can share some of their variables but not others in a
module.
IE.
class A
{
internal int A; //This is shared in the module
private int B; // But not this.
}
No need to reintroduce the "Friend" feature from cpp.
I've always felt the same.
I certainly don't intend this as a way to say "just accept it and forget
about it", but FWIW, I've learned to live with it: I just regard
accessing such privates from outside their class/struct to be bad style.
I mean, I agree it's not ideal, but at least it doesn't prevent me from
getting things done. Again, FWIW.
That said though, it *can* sometimes be helpful for tests and debugging
to be able to reach into a class/struct and muck about with the
privates. (tee hee). But FWIW, I do agree. If D were my own language,
I'd probably have done it a little differently: I would've taken the
current "private" behavior and called it something like "module", and
made "private" behave the way it does in other languages.
Re: We need an internal keyword.
On Sunday, 21 October 2018 at 21:48:22 UTC, Laurent Tréguier wrote: On Sunday, 21 October 2018 at 17:09:05 UTC, 12345swordy wrote: [...] It's not "my" solution. It's D's solution. I perfectly understand why you'd want this and I would probably make use of a private/internal difference myself if it was available. If you already know about this solution however, I don't even know why you're starting this thread; since changing the behavior of private would be a major language change breaking tons of existing codebases, plus it would require adding yet another keyword. Given that this conversation has happened before and things haven't changed, I'm very doubtful that it could happen at any point in time, sadly. If the cost out way the benefits then I simply introduce the "strict" keyword to avoid code breakage, or introduce the optional module scoping. -Alex
Re: We need an internal keyword.
On Sunday, 21 October 2018 at 19:53:35 UTC, Jonathan M Davis
wrote:
On Saturday, October 20, 2018 9:17:23 PM MDT 12345swordy via
Digitalmars-d wrote:
So that classes can share some of their variables but not
others in a module.
IE.
class A
{
internal int A; //This is shared in the module
private int B; // But not this.
}
No need to reintroduce the "Friend" feature from cpp.
Well, if you feel strongly enough about it, you can always
create a DIP to try to change things, but I think that I can
quite safely say that you have pretty much no chance of it
getting accepted. Walter has been quite clear on this topic
when it has been discussed previously (and there was a major
discussion on it not very long ago). It is very much on purpose
that access is controlled at the module level, and if your
module is large enough that it is actually causing problems for
a class or struct to not be able to disallow access of its
members to other symbols in the module, then the module is too
large, and the class should be put in a separate module. We
already have full control of who can access the members of a
class or struct by controlling which module or package its in
and what else is in that module or package with it. If you want
greater control without splitting up your modules more (at
least as far as importing goes), you can always take advantage
of public imports to split the code itself into more modules
while making it possible to get all of the symbols with a
single import statement.
Certainly, if we had internal or friend or other mechanisms for
more fine-grained control without doing it at the module-level,
it would eliminate the need to put code into separate modules
in some cases, but it would also complicate the language, and
what we have works quite well in general.
If you don't like the fact that member access is done at the
module level, then I'm sorry, but overall, this really seems to
be a philosphical complaint and not a practical one, and Walter
has made his stance on it very clear. Based on what he's said
previously, you would need very strong, practical arguments for
why it's causing actual bugs in programs and why the current
approach is not cutting it to stand any chance of convincing
Walter. And honestly, I have never seen anyone come up with an
argument that was particularly good in that regard. It mostly
seems to come down to folks simply objecting to the idea that
anything inside a module would have access to a class' private
members rather than that it's actually caused bugs. In
practice, it's usually either actually useful, or it doesn't
matter.
Either way, simply making post stating that you think that we
should have the feature isn't going to get you anywhere. If you
want it, you'll need to find a way to convince Walter.
- Jonathan M Davis
I do plan to write a DIP on this, no question about it. However
walter have been shown to change his mind unexpectedly. (Remember
the "mangle symbol only" discussion that manu had argue in favor
of, which we all thought he sternly opposed to, only to changed
his mind at the last minute, when a pull request shown up?)
Yes I am aware of the previous discussion regarding this. How can
I forget about it as that guy who complain about it resort to
imposter other people including me.
https://forum.dlang.org/thread/vpxoidaqfbfrnnsep...@forum.dlang.org?page=1
However that guy didn't property explain why this was an issue.
The issue is encapsulation and the traditional get and set
functions for security checking and value checking, etc, before
setting the value, retrieving the value and after setting the
value. Programmers such my self sometimes find it easier to
maintain classes if they are put in the same file, rather than in
each own file.
The "package" solution brought up by another user risk the
possibility of making things more complicated and harder to
maintain than it should be, when working with code that is
considerably small spread across multiple files.
I have two solutions in mind:
A.)Introduce the keyword "strict" for the other keywords private
and protected only class and structs to prevent being accessed
from outside the class in the same module by accident.
Example:
class A
{
strict private int x;// This can't be shared in module
strict protected int y; //ditto
private int x; //This can be shared in module
protected int y; //ditto
}
B.)Get rid of the "one module per file" restriction by
introducing optional module scoping to allow multiple modules in
a file to allow easy package creation for code that is consider
to be small without having to maintain multiple files.
Module C;
Module A // Module A in the C package
{
}
module B // AKA B.A
{
}
-Alex
Re: We need an internal keyword.
On Sunday, 21 October 2018 at 17:48:08 UTC, Neia Neutuladh wrote: On Sun, 21 Oct 2018 08:40:36 +, Laurent Tréguier wrote: This is by design; the D way of dealing with this would be to split the module into a package with multiple modules. This is often a usable way of doing things, but sometimes not so much. If you're writing a script for use with dub --single or rdmd, you lose a lot of convenience if you use multiple files. It might also take a lot of time to split things up once static constructors get involved. As far as single file scripts go, I don't think something like 100% perfect encapsulation is always required. A single file script is usually a quick way to do a specific thing once or twice, and not a persistent project demanding the best code quality.
Re: We need an internal keyword.
On Sunday, 21 October 2018 at 17:09:05 UTC, 12345swordy wrote: I know what the current design is!! You have zero tools in regarding to allowing class to share certain variables but not others in the same module! Create a module for every class is taking all or nothing approach, when there is a reasonable middle ground. Your package solution just make things more unnecessarily complicated then warranted It's not "my" solution. It's D's solution. I perfectly understand why you'd want this and I would probably make use of a private/internal difference myself if it was available. If you already know about this solution however, I don't even know why you're starting this thread; since changing the behavior of private would be a major language change breaking tons of existing codebases, plus it would require adding yet another keyword. Given that this conversation has happened before and things haven't changed, I'm very doubtful that it could happen at any point in time, sadly.
Re: We need an internal keyword.
On Saturday, October 20, 2018 9:17:23 PM MDT 12345swordy via Digitalmars-d
wrote:
> So that classes can share some of their variables but not others
> in a module.
>
> IE.
>
> class A
> {
> internal int A; //This is shared in the module
> private int B; // But not this.
> }
>
> No need to reintroduce the "Friend" feature from cpp.
Well, if you feel strongly enough about it, you can always create a DIP to
try to change things, but I think that I can quite safely say that you have
pretty much no chance of it getting accepted. Walter has been quite clear on
this topic when it has been discussed previously (and there was a major
discussion on it not very long ago). It is very much on purpose that access
is controlled at the module level, and if your module is large enough that
it is actually causing problems for a class or struct to not be able to
disallow access of its members to other symbols in the module, then the
module is too large, and the class should be put in a separate module. We
already have full control of who can access the members of a class or struct
by controlling which module or package its in and what else is in that
module or package with it. If you want greater control without splitting up
your modules more (at least as far as importing goes), you can always take
advantage of public imports to split the code itself into more modules while
making it possible to get all of the symbols with a single import statement.
Certainly, if we had internal or friend or other mechanisms for more
fine-grained control without doing it at the module-level, it would
eliminate the need to put code into separate modules in some cases, but it
would also complicate the language, and what we have works quite well in
general.
If you don't like the fact that member access is done at the module level,
then I'm sorry, but overall, this really seems to be a philosphical
complaint and not a practical one, and Walter has made his stance on it very
clear. Based on what he's said previously, you would need very strong,
practical arguments for why it's causing actual bugs in programs and why the
current approach is not cutting it to stand any chance of convincing Walter.
And honestly, I have never seen anyone come up with an argument that was
particularly good in that regard. It mostly seems to come down to folks
simply objecting to the idea that anything inside a module would have access
to a class' private members rather than that it's actually caused bugs. In
practice, it's usually either actually useful, or it doesn't matter.
Either way, simply making post stating that you think that we should have
the feature isn't going to get you anywhere. If you want it, you'll need to
find a way to convince Walter.
- Jonathan M Davis
Re: We need an internal keyword.
On Sun, 21 Oct 2018 08:40:36 +, Laurent Tréguier wrote: > This is by design; the D way of dealing with this would be to split the > module into a package with multiple modules. This is often a usable way of doing things, but sometimes not so much. If you're writing a script for use with dub --single or rdmd, you lose a lot of convenience if you use multiple files. It might also take a lot of time to split things up once static constructors get involved.
Re: We need an internal keyword.
On Sunday, 21 October 2018 at 08:40:36 UTC, Laurent Tréguier
wrote:
On Sunday, 21 October 2018 at 03:17:23 UTC, 12345swordy wrote:
So that classes can share some of their variables but not
others in a module.
IE.
class A
{
internal int A; //This is shared in the module
private int B; // But not this.
}
No need to reintroduce the "Friend" feature from cpp.
This is by design; the D way of dealing with this would be to
split the module into a package with multiple modules. The A
class would be in its own module, and use `package` where you
used `internal` so that other modules in the same package can
have access to it.
Using a package.d package module
(https://dlang.org/spec/module.html#package-module), you can
still use the multiple modules just as if they were a single
module.
Instead of a source tree like this:
source
|
+-some
|
+-thing.d
You'd end up with a source tree like this:
source
|
+-some
|
+-thing
|
+-package.d
|
+-a.d
|
+-rest_of_the_stuff.d
Where package.d publicly imports a.d and rest_of_the_stuff.d,
so `import some.thing` would still work.
I know what the current design is!! You have zero tools in
regarding to allowing class to share certain variables but not
others in the same module! Create a module for every class is
taking all or nothing approach, when there is a reasonable middle
ground.
Your package solution just make things more unnecessarily
complicated then warranted
Re: We need an internal keyword.
On Sunday, 21 October 2018 at 03:17:23 UTC, 12345swordy wrote:
So that classes can share some of their variables but not
others in a module.
IE.
class A
{
internal int A; //This is shared in the module
private int B; // But not this.
}
No need to reintroduce the "Friend" feature from cpp.
I feel like if this is a problem your Module has too much in it
and you should refactor it into more than 1 module. If you need
to protect a module from itself then I feel like there is too
much in it, essentially.
Re: We need an internal keyword.
On Sunday, 21 October 2018 at 03:17:23 UTC, 12345swordy wrote:
So that classes can share some of their variables but not
others in a module.
IE.
class A
{
internal int A; //This is shared in the module
private int B; // But not this.
}
No need to reintroduce the "Friend" feature from cpp.
I partially agree because Object Pascal uses `strict private` for
this. IIRC they needed at some point to make this for their
Objective-C compatibility. I find it occasionally useful for
example to enforce the usage of the getters method over their
matching fields (if they do lazy update, lazy init, or such
things). So far in D the way to do this is to put the stuff that
has to be strictly private (or internal if you prefer) in a
dedicated module.
This suggestion comes semi-regularly on the forum and last time
it did, the whole topic got very unfriendly so while i partially
agree i have to say that i don't really care. It's a small
sympathetic feature, but i can live without.
Baz.
Re: We need an internal keyword.
On Sunday, 21 October 2018 at 03:17:23 UTC, 12345swordy wrote:
So that classes can share some of their variables but not
others in a module.
IE.
class A
{
internal int A; //This is shared in the module
private int B; // But not this.
}
No need to reintroduce the "Friend" feature from cpp.
This is by design; the D way of dealing with this would be to
split the module into a package with multiple modules. The A
class would be in its own module, and use `package` where you
used `internal` so that other modules in the same package can
have access to it.
Using a package.d package module
(https://dlang.org/spec/module.html#package-module), you can
still use the multiple modules just as if they were a single
module.
Instead of a source tree like this:
source
|
+-some
|
+-thing.d
You'd end up with a source tree like this:
source
|
+-some
|
+-thing
|
+-package.d
|
+-a.d
|
+-rest_of_the_stuff.d
Where package.d publicly imports a.d and rest_of_the_stuff.d, so
`import some.thing` would still work.
