git cl try --luci, a set of hacks to demonstrate and iterate LUCI

luci_hacks/README.md

Index: luci_hacks/README.md
diff --git a/luci_hacks/README.md b/luci_hacks/README.md
new file mode 100644
index 0000000000000000000000000000000000000000..b2aa8c04a8cde411d1faec6fa1c620566378de05
--- /dev/null
+++ b/luci_hacks/README.md
@@ -0,0 +1,35 @@
+LUCI Hacks - A set of shims used to provide an iterable end-to-end demo.
+
+The main goal of Luci Hakcs is to be able to use iterate on Milo as if it was
+displaying real data.  These are a couple of hacks used to get LUCI running from
+"git cl try --luci" to displaying a page on Milo.  These include:
+
+luci_recipe_run.py:
+* Downloading a depot_tools tarball onto swarming from Google Storage to bootstrap gclient.
+** LUCI shouldn't require depot_tools or gclient.
+* Running gclient on a swarming slave to bootstrap a full build+infra checkout.
+** M1: This should check out the recipes repo instead.
+** M2: The recipes repo should have been already isolated.
+* Seeding properties by emitting annotation in stdout so that Milo can pick it
+  up
+* Running annotated_run.py from a fake build directory "build/slave/bot/build"
+
+trigger_luci_job.py:
+* Master/Builder -> Recipe + Platform mapping is hardcoded into this file.  This
+  is information that is otherwise encoded into master.cfg/slaves.cfg.
+** Actually I lied, we just assume linux right now.
+** M1: This information should be encoded into the recipe via luci.cfg
+* Swarming client is checked out via "git clone <swarming repo>"
+* Swarming server is hard coded into the file.  This info should also be pulled
+  out from luci.cfg
+* Triggering is done directly to swarming.  Once Swarming is able to pull from
+  DM we can send jobs to DM instead of swarming.
+
+
+Misc:
+* This just runs the full recipe on the bot.  Yes, including bot_update.
+** In the future this would be probably an isolated checkout?
+** This also includes having git_cache either set up a local cache, or download
+   the bootstrap zip file on every invocation.  In reality there isn't a huge
+   time penalty for doing this, but at scale it does incur a non-trival amount of
+   unnecessary bandwidth.