I discovered a couple of existing shortcomings while implementing this
client that mean this is *not yet ready for inclusion in CI*. I
nonetheless want to get an early review and integrate this to the master
branch, in the interest of small easily reviewable code changes.
The first problem was that the bazel build has apparently never pulled
roots.pem in as a data dependency. This appears not to have been a
problem until XDS enters into the mix. This has been added into the
Bazel build with a small change to the pyx_library rule.
The larger problem is that there is currently no public Python API to
get a peer's identity from the client side. This is crucial to
determining the correctness of an interaction of a client-server pair
under XDS. I intend to add such an method to the RpcContext interface in
an upcoming PR and use it in the interop client. For the moment, I
simply fake the peer details.
Finally, I add a knob to run_xds_tests.py allowing multiple instances of
this test to exist at once. Multiple instances forwarding to the same
port cannot exist on GCE, so this enables multiple people to run tests
at the same time.
d8f3d6f9d4 Merge pull request #233 from annarev/add_port_dependencies
a27429f265 Add port_def.inc and port_undef.inc to generated code hdrs
3797ad2217 port should not be public
578dc32208 Set port visibility to public.
fc4941183f Add :port dependencies to libraries that include it
7d1abedbdd Merge pull request #232 from annarev/forward_assert_changes
1b131ca544 Merge pull request #231 from annarev/fix_build_for_sync
d2389ba261 Add correctness asserts to upb_addmsg
de1bc11663 Split port into its own target, add msg.h to :legacy_msg_reflection
abdf14937d Merge pull request #216 from XrXr/use-memcpy-to-do-unaligned-read
47ab661706 Merge pull request #227 from lizan/absl_workspace_name
0b63c75164 Do not strip :descriptor_upbreflection and make some headers public that are used outside of the their module
bcdfe6b1b0 bazel: use canonical repository name for absl
a73fd86c13 Use memcpy to perform unaligned reads
git-subtree-dir: third_party/upb
git-subtree-split: d8f3d6f9d415b31f3ce56d46791706c38fa311bc
8a3ae1ef3e Merge pull request #229 from veblush/zlib
34e55c1c56 Use github.com as a zlib repo
f32f2fdb25 Merge pull request #221 from haberman/endsubmsg
ea99941f92 Merge pull request #223 from haberman/timestamp
88b1026d26 Fixed leap year handling by reworking upb_mktime() -> upb_timegm().
715718d5a5 Changed endsubmsg handler to return the closure of the submessage.
56589c7276 Merge pull request #219 from haberman/asan
1ed64db8c2 Fixed compiler error "error: control reaches end of non-void function".
32bed8562a Merge pull request #217 from haberman/fix-bazel
5768e68ed7 Fixed for Bazel >= 1.0.
git-subtree-dir: third_party/upb
git-subtree-split: 8a3ae1ef3e3e3f26b45dec735c5776737fc7247f
Turns out that putting this in the main build definition starklark file
means that our users will also incur a dependency on the repo rule.
That's no bueno.
This reverts commit 78e443b4f6.