Why can't we detect at compile time module ctor/dtor cycles (instead of
runtime) ?
See example below.

It should be easy to detect the chain at CT:

algo:
build graph from import dependencies a=>b if module a imports module b.
mark as red the modules that have a ctor
collapse each non-red node into its parent (and update parent's edges)
detect cycles (eg BFS)

(same with dtor)


----
a1:
import a2;
void main(){}
static this(){}


a2:
import a3;
static this(){}

a3:
import a1;
----

rdmd --force --build-only -oftest a1 #works
./test #error:
object.Exception@src/rt/minfo.d(243): Aborting: Cycle detected between
modules with ctors/dtors:
a1 -> a2 -> a3 -> a1

On Saturday, 8 June 2013 at 06:31:14 UTC, Timothee Cour wrote:
> Why can't we detect at compile time module ctor/dtor cycles (instead of
> runtime) ?

No idea, but I've wondered this myself too. After all "imports" are a static feature and all are known at compile-time.

Andrej Mitrovic:

> No idea, but I've wondered this myself too. After all "imports" are a static feature and all are known at compile-time.

rdmd is used often in a situation where it knows all the modules of a program. So it must be able to detect those cycles. It seems an enhancement request for rdmd. Maybe a similar enhancement is possible for dmd, because there are plans to make the D compiler able to find all the modules it needs to compile a program.

Bye,
bearophile

On Mon, Jun 17, 2013 at 5:11 PM, bearophile <bearophileHUGS@lycos.com>wrote:

> Andrej Mitrovic:
>
>
>  No idea, but I've wondered this myself too. After all "imports" are a
>> static feature and all are known at compile-time.
>>
>
> rdmd is used often in a situation where it knows all the modules of a program. So it must be able to detect those cycles. It seems an enhancement request for rdmd. Maybe a similar enhancement is possible for dmd, because there are plans to make the D compiler able to find all the modules it needs to compile a program.
>
> Bye,
> bearophile
>

why do the email threads keep splitting up again?
Anyways, on the original thread 'can we detect at compile time module
ctor/dtor cycles ?', Steven Schveighoffer raised the concern that di files
may have certain private imports not written in the di file, to which I
answered we can at least attempt cycle detection at compile time (which
won't catch all cases) followed by cycle detection at runtime.

On 6/18/13, Timothee Cour <thelastmammoth@gmail.com> wrote:
> why do the email threads keep splitting up again?

I answered through DForum, but the email thread was already split-up
(it had 0 replies). See here:

http://forum.dlang.org/thread/mailman.961.1370673073.13711.digitalmars-d-learn@puremagic.com

Interestingly now the other thread showed up:

http://forum.dlang.org/thread/CANri+Ex8ZwYxtp5h+1FT-cve3SNbUze6CdUF10zHoC2=Rc4mPw@mail.gmail.com

Oh well.