Rémi Zara has sent a couple of interesting ideas to the buildfarm members mailing list.
First, he proposes that the buildfarm client keeps a local git clone which it will point each branch repo at. We can do that, but it needs to be optional, not least because sometimes a local clone is used as the base for more than one buildfarm member (I have three on my Windows VM, all using the same git mirror), so this would increase disk usage in those cases rather than decrease it, which is the desired result. He accompanied his idea with a patch, which I will need to tweak a bit to make it a bit smarter.
Second, he proposes that the buildfarm implement a global lock - that is a lock that covers all its branches rather than just one branch, and decide among the various branches which one it will build. Again, he accompanied his suggestion with a patch.
The global lock on its own is worth doing, and won't be terribly intrusive. Holding an election among the branches, or, as an alternative I have sometimes considered, running all the branches, is rather a bit more difficult, not for any fundamental reason but just because of how the script works. Most of these difficulties could be overcome by using a wrapper script, and that's the way I'm currently inclined to go for this feature.
First, he proposes that the buildfarm client keeps a local git clone which it will point each branch repo at. We can do that, but it needs to be optional, not least because sometimes a local clone is used as the base for more than one buildfarm member (I have three on my Windows VM, all using the same git mirror), so this would increase disk usage in those cases rather than decrease it, which is the desired result. He accompanied his idea with a patch, which I will need to tweak a bit to make it a bit smarter.
Second, he proposes that the buildfarm implement a global lock - that is a lock that covers all its branches rather than just one branch, and decide among the various branches which one it will build. Again, he accompanied his suggestion with a patch.
The global lock on its own is worth doing, and won't be terribly intrusive. Holding an election among the branches, or, as an alternative I have sometimes considered, running all the branches, is rather a bit more difficult, not for any fundamental reason but just because of how the script works. Most of these difficulties could be overcome by using a wrapper script, and that's the way I'm currently inclined to go for this feature.