it is made part of :standard

Just thinking about this again: why not make it part of the Linkage.t? It seems more like an option at the same level of -output-complete-exe to me. WDYT?

it is made part of :standard

Just thinking about this again: why not make it part of the Linkage.t? It seems more like an option at the same level of -output-complete-exe to me. WDYT?

Thinking about your suggestion actually made me realize that there is a slight problem with how we handle the js mode. Today, when the user writes js in the (modes ...) field of an executable, we secretely add also the byte mode (if it is not already present). Later when these "link modes" (Link_mode.t) are translated into Linking.t's, byte gets promoted to byte_complete if shared libraries are not supported. But when using -output-complete-exe the resulting bytecode executable cannot be read by js_of_ocaml! (this is just from reading the code, haven't actually done the experiment)

This means that (modes js) is basically broken on systems without shared libraries... The only way I see to fix it is to actually write down a specific rule for js which always builds a pure bytecode executable, regardless of what is built for byte mode. What do you think?

Anyway, perhaps all this is better addressed in a different PR.

This means that (modes js) is basically broken on systems without shared libraries... The only way I see to fix it is to actually write down a specific rule for js which always builds a pure bytecode executable, regardless of what is built for byte mode. What do you think?

That does seem like the most straightforward thing to do. It feels a bit sad to loose the sharing when everything lines up, but maybe we could try re-adding the sharing as an optimisation once we have a straightforward and working setup.

This has bitrotted and needs to be looked afresh.

@nojb, #5049 doesn't works as expected it seems. See #5282.

Attempting to fix it in #5297 breaks my test. It is possible that the fix is correct and the test is wrong (simulating no shared library support with disable_dynamically_linked_foreign_archives=true).

Would you be able to test the fix ( #5297 ) please ?