@zuiderkwast we didn't want to add python as a dependency for building redis, that's why we're packaging commands.c as part of the project source code.
i.e. people who just download the source package and build it.

Arguably, your suggestion doesn't break that, but i'm not sure i trust file system modification time.
i.e. there could be some cases that somehow the command json files will get a newer timestamp than commands.c.

@yossigo WDYT?

If not, then let's at least add the rule for commands.c so I can write make commands.c to update it, so I don't need to remember the name of the script.

that we can certainly do.. we actually had it, and i asked to remove it (together with the rule for building help.h).

i'm still not confident we can't take your PR as is.. maybe i'm just paranoid, and in practice there's no way that the json files will have a newer time than command.c for naive users that just wanna build redis from a tar file or a git clone.

p.s i do have some ideas about some GH action around commands.c generations or validations (e.g. it can generate it and compare to the copy in the branch).
and also some action to automatically make a PR to redis-doc when changes are merged to redis.

@oranagra I think the problem won't be the filesystem but inexperienced users who somehow get the files modified and don't have the Python dependency.

If don't want to depend on Python and we're going to use JSON more, perhaps it's time to include a minimal C JSON library? There are some in ~400 LOC. For commands.c, we can even use one without Unicode support.

Then we can rewrite the python script in C or we can let redis-server load the JSON files directly.

@zuiderkwast I don't feel very comfortable doing this in C (and certainly not at runtime) only because of toolchain issues.

Maybe we can adopt a middle way - include commands.c as part of the source tree, so you don't have to generate it. Add Make rules to re-generate it if necessary, but don't fail compilation if we lack Python to do that. Does this make sense?

This PR is already a middle way, leaving commands.c checked in. I can add some make ifdefs based on which python, skipping the rule if there's no python.