Would it be possible instead to add a lib output to notmuch? In theory (I haven't tried) it should remove Emacs from the runtime closure for consumers that only use the library component itself.

This has been discussed in #84008, and should probably still be built, but a separate output of some sort. That way, it'd be outside the runtime closure.

I see, then allow me to take this 1 step even further. I hope you'll agree the next commits are better. Notably, there's no attribute rename now - notmuch is the derivation which has both the ruby script and the emacs stuff.

Well, this will not pull in emacs two times when I configure environment.systemPackages = [ pkgs.nemoutt pkgs.notmuch ];. I still think multiple outputs are the way to go here.

Well, this will not pull in emacs two times when I configure environment.systemPackages = [ pkgs.nemoutt pkgs.notmuch ];

No, but it will pull two different version of notmuch - the 1st notmuch is essentially notmuch-lib which has a size of 560Kb while the other notmuch will have a size of 1.4M.

I really don't know why the split outputs got broke during my build trials, as a maintainer of this package, if you feel this is a game changer for you I'd be ready to dive deeper into this. The issue was that I got rmdir: directory not empty errors for $out/share - no idea how that worked before.

My suspicion is that the emacs files weren't moved and as a consequence multiple-outputs.sh failed. That's why personally I like to avoid splitting outputs because Nixpkgs' multiple-outputs.sh is not that reliable.

BTW, just for the record, even with the current state of the derivation, the split outputs don't help removing the derivation's reference to emacs, unless you build it yourself with withEmacs = false; I just wanted that to be the default for neomutt because adding an overlay for that means you build both notmuch and neomutt in order to save 150Mb of Emacs.

@doronbehar I think multiple outputs were made exactly for these usecases. We should get notmuch.lib to work, instead of reinventing notmuch-lib.

I assume bin/notmuch-emacs-mua and share/emacs would both need to be moved to the emacs output - How did the mutli-output approach fail? In the case of circular references, it should show where.

Do you have the commits somewhere that tried to move emacs to a separate output? I could then take a look.

@doronbehar I think multiple outputs were made exactly for these usecases. We should get notmuch.lib to work, instead of reinventing notmuch-lib.

Technically I agree it's just that I have some opinions regarding the current implementation of this idea in Nixpkgs. This is out of scope for this PR.
This is what I got when I went back to master and added "lib" to outputs:

post-installation fixup
Moving /nix/store/9cjz23y381a66daq0m95749159p1h1bx-notmuch-0.29.3/share/man to /nix/store/ffbfvbh6p854887dx8yc8dwy066x4zmx-notmuch-0.29.3-man/share/man
rmdir: failed to remove '/nix/store/9cjz23y381a66daq0m95749159p1h1bx-notmuch-0.29.3/share': Directory not empty
shrinking RPATHs of ELF executables and libraries in /nix/store/9cjz23y381a66daq0m95749159p1h1bx-notmuch-0.29.3
shrinking /nix/store/9cjz23y381a66daq0m95749159p1h1bx-notmuch-0.29.3/lib/libnotmuch.so.5.2.0
shrinking /nix/store/9cjz23y381a66daq0m95749159p1h1bx-notmuch-0.29.3/bin/notmuch
strip is /nix/store/sq2b0dqlq243mqn4ql5h36xmpplyy20k-binutils-2.31.1/bin/strip
stripping (with command strip and flags -S) in /nix/store/9cjz23y381a66daq0m95749159p1h1bx-notmuch-0.29.3/lib  /nix/store/9cjz23y381a66daq0m95749159p1h1bx-notmuch-0.29.3/bin
patching script interpreter paths in /nix/store/9cjz23y381a66daq0m95749159p1h1bx-notmuch-0.29.3
checking for references to /build/ in /nix/store/9cjz23y381a66daq0m95749159p1h1bx-notmuch-0.29.3...
shrinking RPATHs of ELF executables and libraries in /nix/store/ffbfvbh6p854887dx8yc8dwy066x4zmx-notmuch-0.29.3-man
strip is /nix/store/sq2b0dqlq243mqn4ql5h36xmpplyy20k-binutils-2.31.1/bin/strip
patching script interpreter paths in /nix/store/ffbfvbh6p854887dx8yc8dwy066x4zmx-notmuch-0.29.3-man
checking for references to /build/ in /nix/store/ffbfvbh6p854887dx8yc8dwy066x4zmx-notmuch-0.29.3-man...
shrinking RPATHs of ELF executables and libraries in /nix/store/aqh84bqs0zrfmcm5fddybi7al57j7rv9-notmuch-0.29.3-info
strip is /nix/store/sq2b0dqlq243mqn4ql5h36xmpplyy20k-binutils-2.31.1/bin/strip
patching script interpreter paths in /nix/store/aqh84bqs0zrfmcm5fddybi7al57j7rv9-notmuch-0.29.3-info
checking for references to /build/ in /nix/store/aqh84bqs0zrfmcm5fddybi7al57j7rv9-notmuch-0.29.3-info...
find: '/nix/store/5qv7m2ma4pzi80zj3prmpfl0bhfblf4x-notmuch-0.29.3-lib': No such file or directory
strip is /nix/store/sq2b0dqlq243mqn4ql5h36xmpplyy20k-binutils-2.31.1/bin/strip
builder for '/nix/store/gsl65x0fax54v402b384asisfflxzyg6-notmuch-0.29.3.drv' failed to produce output path '/nix/store/5qv7m2ma4pzi80zj3prmpfl0bhfblf4x-notmuch-0.29.3-lib'
error: build of '/nix/store/gsl65x0fax54v402b384asisfflxzyg6-notmuch-0.29.3.drv' failed

Do you have the commits somewhere that tried to move emacs to a separate output? I could then take a look.

I didn't try to put emacs in it's own output, I only tried to keep the outputs attribute as is while I prepared the PR and at some point I started to get very similar errors to this one.

@doronbehar in your case, the build still failed because share/ wasn't empty after moving docs - as the emacs lisp stuff still was inside.

In fact, you can tell notmuch to put emacs lisp files into another location by the --emacslispdir configureFlag. If you then also move notmuch-emacs-mua, you end up eliminating all references to emacs in the out derivation.

I pushed a commit doing that, and cherry-picking the bash completion commit to https://github.com/flokli/nixpkgs/tree/notmuch-multiple-output.

I added some release notes and opened #84663, PTAL.

Closed in favor of #84663 .

This pull request has been mentioned on NixOS Discourse. There might be relevant details there:

https://discourse.nixos.org/t/could-weve-implemented-multi-output-packages-better/6597/1