Hello, Christopher Baines writes:
> Today I had a go at writing a dummy python library, and packaging it, > and then installing it within an environment, with the aim of > experimenting more with how Guix works, and to explore how Guix can be > used as a developer (with unpublished local packages). > > There were several slight issues with this, but I got some help on IRC > to work around these. I've definitely gone down the same route of wanting to use Guix in local development. At the moment I'm using guix environment -l project-guix.scm to spawn a shell with the full working environment for my project — this is awesome, but I think it's slightly orthogenal to what you describe. > The first, is that the hash is required, which I only had to compute > once, but if I wanted to change the package, I would have to update > this, which is prohibitive to local development. As an improvement to > this, could the hash be optional, and if it does not exist, be > calculated when the build is performed? >From my perspective, I think silently calculating a hash on the fly if it is not provided would be problematic: it might lead to laziness in completing the hash, which would undermine the security model of Guix (if I understand correctly). But maybe an explicit flag setting the declaration to "dev-mode", might be useful? > The second issue was getting the data to the store, as first I tried a > relative path, but that did not work, and ended up having to expose the > repository using a HTTP server, and access it over the loopback > interface. This again is rather prohibitive to local development. To > improve on this, could guix build do the insertion in to the store, such > that relative paths would work? This would allow just writing a > "guix.scm" file, which just uses (git-reference (url "./") (commit "HEAD")). This, I think, would be great. Specifically, accepting a "file:///" protocol for (uri)/(git-reference (url)) would be pretty useful… My 2¢… Alex
