with diversions (and here, practically no other mechanism can be used), it's not reasonably possible to cover this edge case where the user *maybe* has also installed manpage translations, so that these are also moved away (but not "replaced", since the zutils has no translated manpages).
On my system, manpages-de 4.27.0-1 is installed. The mechanism via diversions is the implementation level. You don't want a typical user to care about which mechanism is used or how it performs the task; users want to use tools and get higher-level work done. As of now, this mechanism seems to be simply broken or, if you prefer politically correct language, inadequate for the purpose. From a user's viewpoint, if he/she can call two tools, say, `A` and `B`, then the user expects `man A` to deliver the documentation for A and `man B` to deliver the documentation for B. The situation in which `man A` delivers documentation for B or `man B` delivers documentation for A, regardless of the languages, is a bug, or, if you prefer politically correct parlance, ‘a very surprising feature’. If B were a full replacement for A at least in terms of the interface in the strict sense of the word, you could have at least a half of an argument, but it's not the case here (see below). In free software, you as a developer are free do whatever you wish, but from a user's viewpoint, the interface between commands and their manpages should be fixed. Don't ask me how; it sounds like it's going to be a difficult, nontrivial task.
given above situation and the fact, that I consider having no translated manpage at all to be worse than the (in this case compatible tool)
Thought I would agree that the tools are similar, they are incompatible in any direction in the strict sense of the word. If you rely on the current German manpage delivered by `man zgrep` (but running `zgrep` takes a version from zutils) and implement, say, catching errors (including tests for catching errors) based on that documentation (including testing that the switch ‘-r’ doesn't work), parts of your error-catching code remain untested. You will get a mess even if you use/test only positive features: the environment variable GREP is honoured by gzip's zgrep but not by zutils' zgrep. Off-topic, just in case you think that searching directories recursively is always a surplus: sometimes you wish to avoid that, e.g., if the caller is a script expecting no subdirectories, to be rightfully surprised if any is encountered.
