Abnormal exit after upgrading to OCaml 5

[cometkim]

Abnormal exit after upgrading to OCaml 5

[gasche]

This is very likely to be an issue to fix in the program/library, rather than an issue to fix in the OCaml distribution. Please open an issue there.

It looks like the GC is encountered a value that is malformed -- it should be a block but accessing its header segfaults.

There are several ways this could happen, but one possibility would be that some FFI code uses "naked pointers" (OCaml values that are raw pointers to non-OCaml values outside the OCaml heap), which used to accepted for OCaml 4, were deprecated in OCaml 4.11 (2020), and are now illegal in OCaml 5. One would need to look at the FFI code to tell if it is using naked pointers to rule this possibility out. See the reference manual on this topic.

[Octachron]

Here is an example of naked pointer in the odiff code for reading tiff files: https://github.com/dmtrKovalenko/odiff/blob/c79541749f0debaf34b698e592dce6da9fb87e9d/io/tiff/ReadTiff.c#L52 . This is most likely the source of the illegal naked pointer in the test.

[gasche]

There you go:

https://github.com/dmtrKovalenko/odiff/blob/c79541749f0debaf34b698e592dce6da9fb87e9d/io/jpg/ReadJpg.c#L75

I think that you could change the code (also in ReadPng) to use a "managed" bigarray where the memory is malloced by the OCaml runtime when creating the bigarray, and deallocated when it gets collected. Alternatively, you can just set the last bit of the pointer when storing it as an OCaml value, this is the "valid representation as an OCaml tagged integer" approach mentioned last in the reference manual.

[cometkim]

Oh I see, thank for explanation :)