• The original bug report contains example of the bug, do you think you can add a regression test to the testsuite?
• Custom block can normally live in the minor heap, why is it problematic for unmarshaling?

• The original bug report contains example of the bug, do you think you can add a regression test to the testsuite?

I will see if possible. Not easy to automatically track the 'C finalizer' execution.

• Custom block can normally live in the minor heap, why is it problematic for unmarshaling?

This only apply to custom block without finalizer. Custom block with
finalizer are never allocated in the minor heap, see
'caml_alloc_custom' in 'byterun/custom.c'; otherwise, the finalizer
would not be run when the block is dead at the first minor collection.

For the testsuite: perhaps we can add a primitive for computing the total number of memory taken by an ocaml program. It is not an easy thing to do in a portable way but it can be useful for tracking memory usage outside the ocaml heap, like for #6294.

Your fix is simple and not intrusive and I think it should be applied. For a future modification of the marshalling format, I just wonder if you can think of a way for allocating directly intern_dest in the major heap in the problematic case, perhaps with a special boolean in the marshalled data.

This was discussed in a development meeting, and we found that forcing a minor collection at each bigarray unmarshalling had an unreasonsable performance cost for some workflows. Alternative approaches have been discussed, including the one proposed by Pierre in #92 , which supersedes the present PR.