> Regarding recursive clone, I packages the runtime dependencies > seperately but I left the mruby gems, since they seem to be needed > just as source (without needing to be compiled) as they are to be > configured by mruby for mruby-zest. [...] Bundling things that aren't compiled is still bundling. Also, now both mruby-zest and zynaddsubfx bundle rtosc, which is worse than a package bundling a dependent that isn't used anywhere else. I've looked at the latest patches and did some changes (see attached patches), though the bundling still remains. In nanovg, I changed the 'revision' to 0 -- its just a monotonically increasing number for each package update, it's not a 'how many commits have there been', see (guix)Version Numbers. I also changed "ar" to #$(ar-for-target) -- using the 'wrong' ar sometimes works, but not always, e.g. IIRC it failed when cross-compiling from some 64-bit arch to a 32-bit ach? Upstream appears to use 'premake4' as a build system, so I adjusted the package definition accordingly. I set the 'file-name' field for to be more informative. --- I corrected the revision in font-entypo as with nanovg. I expanded the description a little, and added a TODO on building the font from source (it seems to have been neglected for font packages because of node dependencies). The license combination seemed a bit odd to me so I added some more text there. --- mruby-zest: I changed 'TODO: package mruby gems separately' to 'TODO: unbundle mruby gems' as that seems easier to search for. Some things were actually packaged, so I removed those from 'deps' in a snippet. It turned out that the Makefile tried to build mruby, so I patched that out. As mruby isn't included anymore, you could give removing some of the bundled mruby gems a try, especially given that they aren't actually installed. I removed '#:tests? #false' because a test suite exists -- the makefile has a 'test' and 'rtest' target. I choose the 'rtest' target (because 'test' is for testing mruby, not mruby-zest). It turned out the test suite requires 'ruby-ruby-prof' so I added that as a native-input. You have done #$some-input in the 'install' phase -- package transformation don't know to adjust those, so I replaced it with a search-input-file equivalent. Putting the fonts in the output of mruby-zest doesn't lookk quite right to me, so I instead moved things to a post-unpack phase and used 'substitute*' instead of 'symlink'. I noticed that pcre was bundled (deps/mruby-regexp-pcre/pcre'), so I removed that. By looking at , I noticed it looks for libzest.so in /opt/zyn-fusion/libzest.so, which is incorrect in Guix (and other distros too). I removed the wrapping because: * no reason was listed for wrapping * presumably the incorrect reference was the thing that LD_LIBRARY_PATH was a work-around for, but that reference has now been corrected. * LD_LIBRARY_PATH is leaky and hence to be avoided (if zest spawns a subprocess, then that subprocess would get LD_LIBRARY_PATH too) As a bonus, this allowed removing 'bash-minimal' from the inputs. From , I noticed that apparently some references to schema/test.json and MainWindow.qml were incorrect. These are now patched. I simplified the unbundling substitutions a little. For example, it turned out that build_config.rb did not need any substitutions somehow. The makefile uses pkg-config instead of TARGET-pkg-config, which is incorrect when cross-compiling, so I patched that. On the fonts: I noticed that the fonts don't become part of the closure (with "guix gc --references"). As such, maybe the source files that use the fonts are actually unused. I propose to give installing the examples that use the fonts a try, or alternatively explicitly choose to not install the (font-using) examples. --- I changed the version of zynaddsubfx from (package-version mruby-zest) to "3.0.6", otherwise the package would break after a mruby-zest update. I re-added the comment of 'remove-sse-flags-from-generic-target' that was removed. I changed #$mruby-zest to a search-input-file equivalent, for the same reason as with previous patches. You forgot to mention the home page, so I adjusted the commit message. You are fixing a file name, not a path, so I adjusted the new phase name. By removing the 'ntk' input, zynaddsubfx now has one less interface, so I re-added it, with a comment. It appeared that doxygen documentation wasn't actually build (there is no .html in the output, and the configuraiton script complained about some missing component), so I removed the new doxygen (native-)input. bash completions weren't installed in the new version even though they were in the previous version, so I made some changes to support that. I adjusted the substitution of libzest.so to substitute all cases, in case somehow the first dlopen fails. I compared the diff between old and new source code, there doesn't appear to be anything 'suspicious' though its just a cursory look and it would be easy to hide things: guix shell diffoscope -- diffoscope /gnu/store/9j09zj472211nrrs5jmyxdsq2lpfd36q-zynaddsubfx-3.0.5.tar.bz2 /gnu/store/xyjiq3nmk372ap4vq7sl7n7f9rc5xshs-zynaddsubfx-3.0.6.tar.bz2