builder page allocation leak and segfault about mpack HOT 6 OPEN

Fixed, sorry about this. It's an unfortunate off-by-one bug which is triggered on the rare occasion that a build ends up exactly at the end of a 4096 page boundary. I'm a little surprised I never encountered this bug in my own projects; I guess I tend to keep my builds fairly small so my nested builds never roll over a page.

I'm having trouble crafting a unit test that will reproduce this. The tests currently set a very small page size to test various wrapping behaviors but this size makes it impossible for a build to be aligned at the end. I'll have to make the page size vary by unit test. I will keep this open until it's properly unit tested.

from mpack.

Thanks for the very quick fix!
can't we add the current example above to the unit tests as a regression test, at least until a more specific way to test this has been implemented?

from mpack.

Your example doesn't trigger the bug in the unit test suite because the unit tests use small page sizes to test various wrapping behaviors. I need to change the unit test runner to make the page size configurable to properly test this. It also needs to be optimized and simplified. The unit test suite is run under Valgrind so unit tests need to be as fast as possible.

from mpack.

Yes, that was quick. And now, everything works perfectly. Thanks a lot for your great work!

from mpack.

I think I stumbled over this, using v1.1. I haven't confirmed it, but the problem seems very similar and it was fixed by switching to the develop branch. It would be great, if this could be fixed with a new stable release.

from mpack.

Release v1.1.1 is out. Sorry for the delay. I'll leave this open because I still want to improve the unit tests for this.

from mpack.