Our current strategy for running many tests is to copy/symlink them to the objdir during the build and to run them out of the objdir. The development cycle thus looks like: edit -> copy/symlink -> test -> repeat As the number of tests has grown, that copy/symlink stage has become quite time intensive and extremely frustrating for the development experience, especially on Windows, where I/O is a bit slower. While there is room to optimize the syncing stage (which we may do for short term wins), the best way to fix this problem is to make it go away. We want to avoid having to copy tens of thousands of test files at build time and instead run tests from the source directory directly. This is a major change. It has implications for how tests are written and executed. It has implications for test packaging. For example, do we still need massive test archives after this change or should testers have a clone of the source archive to run tests from? This bug tracks everything around running tests more efficiently by running them from the source directory.

(In reply to Gregory Szorc [:gps] from comment #0) > This is a major change. It has implications for how tests are written and > executed. It has implications for test packaging. For example, do we still > need massive test archives after this change or should testers have a clone > of the source archive to run tests from? Note that we don't necessarily need to care about this directly. A transitional step is to generate test archives with the same tree structure as the source tree.

(In reply to Mike Hommey [:glandium] from comment #1) > Note that we don't necessarily need to care about this directly. A > transitional step is to generate test archives with the same tree structure > as the source tree. Yup. I believe the rough plan was to ignore the VCS/archives implications for now and continue generating test archives as they are today. That will hopefully postpone the headache of having to change automation too much.