What advice would you give on the design of a large complex build process and makefile structure with ElectricAccelerator?
We are currently in the process of re-architecting our entire build process. This include larger machines, new revision control software, and will involve rewriting many makefiles. This project is currently in the design/brainstorm phase and I was hoping to get some advice from the community on what works well and what does not
Best Answer
There are many things to consider and the importance of each one depends on your specific situation:
- If you have a lot of "make" code to parse and you use lots of macros with $(call) and $(eval) then the startup time can begin to have an impact. It does depend a lot on your situation and in most cases this is fairly irrelevant but it has bitten me.
- Try not to have huge jobs - one job that takes a very long time can spoil your parallelism e.g.
- Calling out to some other build system to perform some big task like building java or gcc or whatever.
- Creating giant ZIP files of components that you just built. (try to split the components to create more small ZIPs)
- Try to generate packages/installers/CD images as part of the build so that they can happen in parallel to the maximum extent possible. What I have found sometimes is that the build shrinks thanks to Accelerator but every surrounding task in the release process gets bigger until build time is not the problem anymore but something else is.
- Give thought to how you create a clean build area to start a new build, how to get rid of an old build quickly and how to get builds from wherever they are made to the users. All these things can start to eat up time and they are often forgotten about because they don't happen inside the makefile.
The best way to make a build fast is to not build anything. This may seem like a joke, but it is true. Check out this article on implementing build avoidance using subbuilds:
http://www.electric-cloud.com/blog/2009/10/21/subbuilds-build-avoidance-done-right/
