PR#7370: bump magic number (ocamldep segfault)

[damiendoligez]

The question is: are there any other magic numbers we need to change, and how do we find them?

[alainfrisch]

What's the argument against systematically bumping all magic numbers before a release?

[damiendoligez]

You mean a major release, right?

[alainfrisch]

I actually meant any release. I don't see a scenario where users would like to mix .cmi/.cmx/.cmxs files produced by different versions of the compiler. Even if the format doesn't change, the behavior of the compiler does. And even if it didn't, who would take the risk?

[trefis]

It's not about mixing files, it's about the tools processing this file, is it not? Why would you require some tool to be "updated" and built for as many (minor) versions of the compiler if there is no need to?
For instance merlin usually doesn't really need updating between minor releases and you need only build on merlin to work with say ocaml 4.02.2 and 4.02.3.

[alainfrisch]

Even for external tools, I would expect that the installation be usually managed by OPAM, and that the workflow would guarantee that Merlin has been compiled exactly with the same version as the one used for compiling the edited files.

[mshinwell]

I tend to think it's preferable to not bump the version numbers unless required for minor releases. Extra time spent managing deployments that are not using OPAM is a concern here. However I think all of the numbers could reasonably be bumped upon a major release. (Tools such as Merlin would probably need to be changed across such a release in any event.)

[damiendoligez]

We'll probably switch to bumping all the numbers on each major release, but not this time because I'd like to see a bit more discussion (and I'm feeling a bit nervous about breakage).