I splitted the patch in 5. If it is not possible to update pcre2, we only need to disable a couple of tests. zimoun writes: > I am not convinced by this move of patches from ’source’ to ’phases’. > My understanding about the usual way is to let the patch in the source > field. Is this move motivated by something special? Yes. Some of the patches need to be applied with the "-p1" flag some with "-p2". In the source field it is not possible to do so. So I would need to modify beforehand the "p2" patches for example, and I could not think of a nice way to do that without including the patches directly in gnu/packages/patches which does not seem like a nice option. And since all these patches are in julia source, I thought it was nicer this way. > I have not read the Julia ChangeLog. Do they remove Arpack? This > should be mentioned in the commit message. Yes they did, I updated the commit message. Same with the other new dependencies. >> + (link "zlib" "usr/lib/julia/" "libz\\.so") > Does this fix > > > ? If yes, cool and thank you! :-) So it should be mentioned in the > commit message, something like: No it does not. Reason for this is that the install target of the makefile copies a bunch of libraries from the build directory to the install directory. I guess "normally" you rely on the makefile to also build these libraries for you. In our case many libraries are missing in this folder at the end of the build target so I create the symlinks so the install target works and also julia can later on find the libraries. For some reason utf8proc and libLLVM-11 need to be symlinked directly in the install directory, this is done in the next phase. >> + ("nss-certs" ,nss-certs) ; required to precompile > Hum? Is it really necessary? I think it is necessary for some of the tests. So I moved it to native-inputs is that correct ? >> + ("glibc-locales" ,glibc-locales) > Idem. Is it really necessary? Because it is a “big“ packages which > drastically increases the closure size of the Julia package. It's not! Good find, this was committed inadvertently when I was looking for the reason why some tests were complaining of not finding locales. Turns out that these tests were happening in a julia process started without the GUIX_LOCPATH environment variable. Nicolò Balzarotti wrote: > julia -e 'using Pkg; Pkg.add("GZip"); using GZip' > Same error > ┌ Error: curl_easy_setopt: 48 > └ @ Downloads.Curl /tmp/guix-build-julia-1.6.1.drv-0/julia-1.6.1/usr/share/julia/stdlib/v1.6/Downloads/src/Curl/utils.jl:36 I have the same "bug" here though everything works correctly. I suspect it is due to a mismatch in the libcurl version. Cheers, JB.