[c++] Allow RayConfig to have items other then integer

[guoyuhong]

What do these changes do?

1. Current ray config only has integer items. Some other developers wonder whether we can use other types in ray config and they have to work around to use them. This PR allows other types to be used in ray config.
2. For id_def.h and ray_config_def.h, some compiler will raise some warning message showed below.

In file included from XXX.cc:12:0:
ray/src/ray/id.h:128:27: warning: extra ‘;’ [-Wpedantic]
 DEFINE_UNIQUE_ID(UniqueID);
                           ^

Related issue number

N/A

Linter

• I've run scripts/format.sh to lint the changes in this PR.

Test

Since there is no other types used in ray config now, I test the case of using double, bool, std::string by myself. The code works fine.

[AmplabJenkins]

Test PASSed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Ray-Perf-Integration-PRB/605/
Test PASSed.

[AmplabJenkins]

Test FAILed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Ray-PRB/13978/
Test FAILed.

[jovany-wang]

Nice change!

[jovany-wang]

Another approach to reserve the ; for every RAY_CONFIG () statement is to add a RAY_IGNORE() at the end of this macro like:

#define RAY_CONFIG(type, name, default_value) \
 if (pair.first == #name) {                  \
    std::istringstream stream(pair.second);   \
    stream >> name##_;                        \
    continue;                                 \
  }                                           \
  RAY_IGNORE(nothing)

But I don't know if this way is good enough.

[guoyuhong]

That way looks a bit wired. We can wait for other developers' comments.

[jovany-wang]

I don't think this is wired since it's similar to RAY_LOG().

[robertnishihara]

I have a slight preference for requiring a ; after the macro also. If you add RAY_IGNORE at the end, then I'd suggest including a comment to explain its purpose.

[guoyuhong]

OK, will do.

[raulchen]

Adding RAY_IGNORE looks fine to me as well.

[guoyuhong]

RAY_IGNORE_EXPR(0) can only be used in functions. It is not allowed to use RAY_IGNORE_EXPR(0) when a class is defined. I used another way to may the ; required.

[AmplabJenkins]

Test PASSed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Ray-Perf-Integration-PRB/726/
Test PASSed.

[AmplabJenkins]

Test PASSed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Ray-Perf-Integration-PRB/727/
Test PASSed.

[AmplabJenkins]

Test FAILed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Ray-PRB/14135/
Test FAILed.

[raulchen]

This way seems more weird. I slightly prefer to just remove the semicolons.

[guoyuhong]

Yes. Since the Macro is only used in id_def.h and ray_config_def.h, it won't break the traditions in most c++ files.

[AmplabJenkins]

Test PASSed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Ray-Perf-Integration-PRB/735/
Test PASSed.

[AmplabJenkins]

Test FAILed.
Refer to this link for build results (access rights to CI server needed):
https://amplab.cs.berkeley.edu/jenkins//job/Ray-PRB/14142/
Test FAILed.

[raulchen]

LGTM