Allow external TaskQueue implementations on Linux

webrtc/rtc_base/BUILD.gn

Index: webrtc/rtc_base/BUILD.gn
diff --git a/webrtc/rtc_base/BUILD.gn b/webrtc/rtc_base/BUILD.gn
index 60d1f57583cf9f86f0d74ac6c63484d30a5f39c6..9e015d155bd3f54bfae62072651a9d8fc165342a 100644
--- a/webrtc/rtc_base/BUILD.gn
+++ b/webrtc/rtc_base/BUILD.gn
@@ -294,11 +294,30 @@ config("enable_libevent_config") {
   defines = [ "WEBRTC_BUILD_LIBEVENT" ]
 }
 
-rtc_static_library("rtc_task_queue") {
+rtc_source_set("rtc_task_queue") {
   public_deps = [
     ":rtc_base_approved",
   ]
 
+  if (rtc_link_task_queue_impl) {
+    public_deps += [ ":rtc_task_queue_impl" ]

[nisse-webrtc, 2017/08/23 11:00:25]

I was thinking of adding the dependency on rtc_task_queue_impl in the top-level
library-like targets. But maybe it makes more sense here.

[perkj_webrtc, 2017/08/23 11:12:37]

That is what I did in a previous cl. But I think its nicer that targets does not
have to care if the implementation is linked or not and we don't need to change
the existing deps. If you want to have a tq you simply depend on this target and
do not have to care. 

[nisse-webrtc, 2017/08/23 11:24:46]

Ok. We could refine it to say that whoever instantiates a task queue needs this
target, while code that only get a task queue passed in can depend on an api
target only.

I'd suggest we move task_queue.h to api/ later on, and then we can revisit this.
 

+  } else {
+    if (build_with_chromium) {
+      sources = [
+        "../../webrtc_overrides/webrtc/rtc_base/task_queue.h",
+      ]
+    } else {
+      sources = [
+        "task_queue.h",
+      ]
+    }
+  }
+}
+
+rtc_source_set("rtc_task_queue_impl") {
+  deps = [
+    ":rtc_base_approved",
+  ]
   if (build_with_chromium) {
     sources = [
       "../../webrtc_overrides/webrtc/rtc_base/task_queue.cc",
@@ -307,18 +326,15 @@ rtc_static_library("rtc_task_queue") {
   } else {
     sources = [
       "task_queue.h",

[nisse-webrtc, 2017/08/23 11:00:25]

It's not so nice to task_queue.h in multiple targets (do we do that anywhere
else?). Would it make sense with a separate _api target for just the header
file?

[perkj_webrtc, 2017/08/23 11:12:37]

Sure

[mbonadei, 2017/08/23 12:12:29]

I agree, we should create a separate _api target. With that I think we will be
able to avoid to expose :rtc_task_queue_impl in the public_deps list of
rtc_task_queue. 

-      "task_queue_posix.h",
     ]
     if (rtc_build_libevent) {
-      deps = [
-        "//base/third_party/libevent",
-      ]
+      deps += [ "//base/third_party/libevent" ]
     }
-
     if (rtc_enable_libevent) {
       sources += [
         "task_queue_libevent.cc",
         "task_queue_posix.cc",
+        "task_queue_posix.h",
       ]
       all_dependent_configs = [ ":enable_libevent_config" ]
     } else {