I think I still don't understand dependencies

[davidchisnall]

I have a target that does:

add_deps(...)

This target has a rule that implements on_linkcmd to define the link job. This batchcmd is sometimes (deterministically, 100% of the time, on first build after clean) run before some of the dependencies listed in add_deps are built, which then causes the linker to fail because some files don't exist.

I am using batchcmd:add_depfiles in my rule to specify some additional files, but I'd assumed that the dependencies are also built. I've also tried explicitly using add_depfiles on the set of files that are passed to the linker. The files being passed to the linker are collected with:

for name, dep in pairs(target:deps())
  -- filter out some and use dep:targetfile() to get the files for the rest
end

I don't understand why a target can be built before its dependencies are built.

[waruqi]

I'm not sure what you are talking about, you could provide a reproducible demo project and then provide some error output.

[davidchisnall]

Sorry, when the firmware target is created it uses (description scope) add_deps to add the dependencies. This is why they then show up in the target:deps() iteration in the rule target.

[waruqi]

But right now I'm not sure what target dependencies you've configured, or what errors you've encountered, and I don't have any error logs that I can analyse or an example project that I can debug.

If you think there's a problem with the build order, you'll still need to provide a minimal example project that can reproduce the problem so I can continue to analyse the cause.

[davidchisnall]

That may have to wait until our open source release, which unfortunately means that we'll have to keep using the CMake build system and reconsider xmake in a year or so. I can run whatever debugging commands you want but I can't share the code yet. Thanks anyway.

[waruqi]

I don't need your complete code, huge code and environments are hard to debug. Provide minimal reproducible example projects that just contain empty code and configuration scripts that reproduce build order errors, and then tell me what errors I encounter.

[davidchisnall]

I can't produce a reduced test case without some idea of what kinds of thing might be causing this. It seems to have gone away after some changes that I'd expect to be completely unrelated. It also seems not to happen at all when I run xmake from the source directory.