Use of glob(**) in top-level BUILD.bazel breaks builds in dirty workspaces

[jmillikin-stripe]

(copied/moved from google/subpar#29)

BUILD.bazel contains this target:

py_library(
   name = "containerregistry",
   srcs = glob(["**/*.py"]),
   deps = [ ... ]   
)

In a clean workspace, that glob picks up only the intended Python sources. But the build creates bazel-* symlinks, which are then detected by the glob and break later builds:

$ bazel clean
$ bazel build :pusher.par
INFO: Found 1 target...
Target //:pusher.par up-to-date:
  bazel-bin/pusher.par
INFO: Elapsed time: 1.992s, Critical Path: 0.54s
$ bazel build :pusher.par
INFO: Found 1 target...
ERROR: /Users/jmillikin/src/containerregistry/BUILD.bazel:36:1: Building par file //:pusher.par failed: I/O exception during sandboxed execution: /private/var/tmp/_bazel_jmillikin/165d82f9b8e48612b4a588f395fdde11/execroot/containerregistry/bazel-bin (No such file or directory).
Target //:pusher.par failed to build
Use --verbose_failures to see the command lines of failed build steps.
INFO: Elapsed time: 0.503s, Critical Path: 0.06s

This can be fixed by using a more verbose glob over the directories you expect to exist:

py_library(
   name = "containerregistry",
   srcs = glob([
     "__init__.py",
     "client/**/*.py",
     "tools/**/*.py",
     "transform/**/*.py",
     "transport/**/*.py",
   ]), 

Another solution is to move the containerregistry target to a subdir, so it won't pick up stray build artifacts.

[mattmoor]

We could probably add exclude=["bazel-*/**/*.py"] to the glob, but I'll go for the slightly less lazy enumeration you suggest. thanks again for spotting this.

[duggelz]

Can you also file a bug against Bazel itself? It's supposed to have logic to ignore these links for globbing, I believe.

[mattmoor]

Yeah, I will open one.