Diamond relationship when using precomp modules causes packages to disappear

[p6rt]

Migrated from rt.perl.org#122773 (status was 'resolved')

Searchable as RT122773$

[p6rt]

From @hoelzro

If A and B both require C, and I load both A and B from main.pl, package definitions within A.pm disappear. If you load B before A in main.pl, B's disappear. Removing the "use C" from A or B prevents this problem.

I've attached a tarball and the raw files themselves.

I've only tried this on MoarVM.

[p6rt]

From @hoelzro

A.pm6

[p6rt]

From @hoelzro

B.pm6

[p6rt]

From @hoelzro

C.pm6

[p6rt]

From @hoelzro

Makefile

[p6rt]

From @hoelzro

example.tar.gz

[p6rt]

From @hoelzro

main.pl

[p6rt]

From @hoelzro

I also see this on the JVM, but not Parrot.

On Fri Sep 12 20​:10​:00 2014, rob@​hoelz.ro wrote​:

If A and B both require C, and I load both A and B from main.pl,
package definitions within A.pm disappear. If you load B before A in
main.pl, B's disappear. Removing the "use C" from A or B prevents
this problem.

I've attached a tarball and the raw files themselves.

I've only tried this on MoarVM.

[p6rt]

From @hoelzro

git bisect points to f2842a5 as the culprit.

On Fri Sep 12 21​:09​:56 2014, rob@​hoelz.ro wrote​:

I also see this on the JVM, but not Parrot.

On Fri Sep 12 20​:10​:00 2014, rob@​hoelz.ro wrote​:

If A and B both require C, and I load both A and B from main.pl,
package definitions within A.pm disappear. If you load B before A in
main.pl, B's disappear. Removing the "use C" from A or B prevents
this problem.

I've attached a tarball and the raw files themselves.

I've only tried this on MoarVM.

[p6rt]

From @hoelzro

At least it is for MoarVM; I went back to 2014.04 on the JVM and the bug was still present there.

On Sat Sep 13 09​:46​:41 2014, rob@​hoelz.ro wrote​:

git bisect points to f2842a5 as the culprit.

On Fri Sep 12 21​:09​:56 2014, rob@​hoelz.ro wrote​:

I also see this on the JVM, but not Parrot.

On Fri Sep 12 20​:10​:00 2014, rob@​hoelz.ro wrote​:

If A and B both require C, and I load both A and B from main.pl,
package definitions within A.pm disappear. If you load B before A in
main.pl, B's disappear. Removing the "use C" from A or B prevents
this problem.

I've attached a tarball and the raw files themselves.

I've only tried this on MoarVM.

[p6rt]

From @hoelzro

The problem seems to be with MoarVM commit 7a722dc4018ce1b99bbd4abea198e242d7da0157.

On Mon Sep 15 06​:13​:58 2014, rob@​hoelz.ro wrote​:

At least it is for MoarVM; I went back to 2014.04 on the JVM and the
bug was still present there.

On Sat Sep 13 09​:46​:41 2014, rob@​hoelz.ro wrote​:

git bisect points to f2842a5 as the culprit.

On Fri Sep 12 21​:09​:56 2014, rob@​hoelz.ro wrote​:

I also see this on the JVM, but not Parrot.

On Fri Sep 12 20​:10​:00 2014, rob@​hoelz.ro wrote​:

If A and B both require C, and I load both A and B from main.pl,
package definitions within A.pm disappear. If you load B before
A in
main.pl, B's disappear. Removing the "use C" from A or B
prevents
this problem.

I've attached a tarball and the raw files themselves.

I've only tried this on MoarVM.

[p6rt]

From @dwarring

The attached test cases have been added to roast S10-packages/precompilation.t. Fudged for jvm and moar.

On Mon Sep 15 06​:48​:06 2014, rob@​hoelz.ro wrote​:

The problem seems to be with MoarVM commit
7a722dc4018ce1b99bbd4abea198e242d7da0157.

On Mon Sep 15 06​:13​:58 2014, rob@​hoelz.ro wrote​:

At least it is for MoarVM; I went back to 2014.04 on the JVM and the
bug was still present there.

On Sat Sep 13 09​:46​:41 2014, rob@​hoelz.ro wrote​:

git bisect points to f2842a5 as the culprit.

On Fri Sep 12 21​:09​:56 2014, rob@​hoelz.ro wrote​:

I also see this on the JVM, but not Parrot.

On Fri Sep 12 20​:10​:00 2014, rob@​hoelz.ro wrote​:

If A and B both require C, and I load both A and B from
main.pl,
package definitions within A.pm disappear. If you load B
before
A in
main.pl, B's disappear. Removing the "use C" from A or B
prevents
this problem.

I've attached a tarball and the raw files themselves.

I've only tried this on MoarVM.

[p6rt]

The RT System itself - Status changed from 'new' to 'open'

[p6rt]

From @hoelzro

This has been fixed on MoarVM; the issue is still outstanding on JVM.

On Fri Sep 12 20​:10​:00 2014, rob@​hoelz.ro wrote​:

If A and B both require C, and I load both A and B from main.pl,
package definitions within A.pm disappear. If you load B before A in
main.pl, B's disappear. Removing the "use C" from A or B prevents
this problem.

I've attached a tarball and the raw files themselves.

I've only tried this on MoarVM.

[p6rt]

From @hoelzro

My hacky "make it eager" fix has been reverted in MoarVM, but the underlying issue has been fixed in commit c6e8df8. The JVM issue is still outstanding, but maybe that commit will offer hints on how to fix it.

On Tue Nov 04 18​:47​:33 2014, rob@​hoelz.ro wrote​:

This has been fixed on MoarVM; the issue is still outstanding on JVM.

On Fri Sep 12 20​:10​:00 2014, rob@​hoelz.ro wrote​:

If A and B both require C, and I load both A and B from main.pl,
package definitions within A.pm disappear. If you load B before A in
main.pl, B's disappear. Removing the "use C" from A or B prevents
this problem.

I've attached a tarball and the raw files themselves.

I've only tried this on MoarVM.

[p6rt]

From @hoelzro

According to jnthn, the MoarVM changes were integrated into nqp-j in NQP commit cbef59357.

On Sun Mar 22 14​:52​:00 2015, rob@​hoelz.ro wrote​:

My hacky "make it eager" fix has been reverted in MoarVM, but the
underlying issue has been fixed in commit c6e8df8. The JVM issue is
still outstanding, but maybe that commit will offer hints on how to
fix it.

On Tue Nov 04 18​:47​:33 2014, rob@​hoelz.ro wrote​:

This has been fixed on MoarVM; the issue is still outstanding on JVM.

On Fri Sep 12 20​:10​:00 2014, rob@​hoelz.ro wrote​:

If A and B both require C, and I load both A and B from main.pl,
package definitions within A.pm disappear. If you load B before A
in
main.pl, B's disappear. Removing the "use C" from A or B prevents
this problem.

I've attached a tarball and the raw files themselves.

I've only tried this on MoarVM.

[p6rt]

From @usev6

The test in S10-packages/precompilation.t now passes on rakudo.JVM, also.

I'm closing this ticket as 'resolved'.

[p6rt]

From @usev6

The test in S10-packages/precompilation.t now passes on rakudo.JVM, also.

I'm closing this ticket as 'resolved'.

[p6rt]

@usev6 - Status changed from 'open' to 'resolved'