Select between introspection_c and introspection_cpp #133
Conversation
| find_package(rosidl_typesupport_introspection_cpp) | ||
| if(NOT rosidl_typesupport_introspection_cpp_FOUND) | ||
| if(NOT rosidl_typesupport_introspection_cpp_FOUND AND NOT rosidl_typesupport_introspection_c) |
|
lol |
|
That's how we roll. |
|
I'm starting CI for this to see that it doesn't break anything. Locally testing with test_rclcpp worked. http://ci.ros2.org/job/ci_linux/978 I haven't tested this with C yet though. I'll make a differently named branch to enable building rcl for Connext Dynamic. |
jacquelinekay
force-pushed
the
introspection_cpp
branch
from
4285dc6
to
f976b85
Compare
|
Alright, I fixed some stuff with strings and started a branch in rcl that enables the dynamic test again. |
jacquelinekay
force-pushed
the
introspection_cpp
branch
from
06e4282
to
024aa4b
Compare
|
I made some style fixes, relaunched with only Connext Dynamic enabled (hope that works). |
jacquelinekay
force-pushed
the
introspection_cpp
branch
from
c0b80dd
to
7260481
Compare
jacquelinekay
force-pushed
the
introspection_cpp
branch
from
f2edc65
to
27440bb
Compare
|
Fixed a warning on OSX and Windows. http://ci.ros2.org/job/ci_osx/827/ I think the failed tests on Windows also appear on the nightly build. I think we should fix those failures before merging this, but we should start review on this now. I separated the new macros into a macros.hpp file so that the diff isn't impossible to view in the browser. |
jacquelinekay
added
in review
| if (using_introspection_c_typesupport(typesupport)) { | ||
| CREATE_TYPENAME_PREFIX(INTROSPECTION_C_TYPE) | ||
| } else if (using_introspection_cpp_typesupport(typesupport)) { | ||
| CREATE_TYPENAME_PREFIX(INTROSPECTION_CPP_TYPE) |
There was a problem hiding this comment.
Since both macros have the code to access the fields of members in common I would suggest to let the macro only define the members variable and continue to handle accessing the fields package_name_ and message_name_ outside.
There was a problem hiding this comment.
Then the macro could also be renamed to something like GET_TYPED_MEMBERS.
tfoote
added
in progress
|
After some thought I created a branch with a different approach that is more concise than this one (although I'm still not sure if it's the best): https://github.com/ros2/rmw_connext/compare/introspection_template Unfortunately I'm running into a compile error that I can't quite figure out, due to 7e8232e. Looks like feature isn't going to make it into this release, if the pull request in the current state is stylistically unacceptable. |
|
@jacquelinekay thanks for trying to get it in. If you need someone to review it, then I can lend a hand after the alpha. We can try to get it merged asap after the release. |
jacquelinekay
force-pushed
the
introspection_cpp
branch
from
7861a0a
to
7ae7e7f
Compare
jacquelinekay
added
in review
|
alright, I took a pretty different approach that is more concise and I reset this branch to reflect it. (the old approach is still around, in https://github.com/ros2/rmw_connext/compare/introspection_cpp_old) http://ci.ros2.org/job/ci_linux/1020/ |
|
|
||
| bool using_introspection_c_typesupport(const char * typesupport_identifier) | ||
| { | ||
| return strcmp(typesupport_identifier, rosidl_typesupport_introspection_c__identifier) == 0; |
There was a problem hiding this comment.
The should not use string comparison since it is a pretty expensive operation. See the previously used macro: https://github.com/ros2/rmw/blob/55254b66ddb0f7c5ff8f05fb8c0c60378fb3b2da/rmw/include/rmw/impl/cpp/macros.hpp#L70
Same below.
There was a problem hiding this comment.
everything here is a const char *, so using the equality operator should be okay.
| return _create_type_name<rosidl_typesupport_introspection_cpp::MessageMembers>(untyped_members, | ||
| sep); | ||
| } | ||
| RMW_SET_ERROR_MSG("service members handle is null"); |
There was a problem hiding this comment.
The error message doesn't match the error case.
Same below.
|
To simplify the code further I think there should be only one list of field types (in C) and C++ should use the same constant. The C++ part could still define "nicer" constants in a namespace (based on the C constants) but the rmw impl. would only need to deal with the C constants. |
|
Overall this looks much nicer then the previous branch. |
|
I agree that the C/C++ introspection APIs should share some types; I see no reason why they need separate definitions--that was the most frustrating thing about making this change... I suppose the current scheme is better if we wanted to avoid a dependency of rosidl_typesupport_introspection_cpp on rosidl_typesupport_introspection_c, but I think most dynamic RMW implementations will want to support both C and C++ introspection and will therefore depend on both packages. So introducing a dependency between the two introspection packages is not a big deal. |
|
Oh, I understand now, changes to the introspection packages aren't necessarily needed, I believe the field type enums from either package can be used in the switch statements that I currently have macros for. |
|
I realized why sharing the enum definitions would be nice to have. ros2/rosidl#94 |
|
CI for the new approach, with only connext dynamic enabled. http://ci.ros2.org/job/ci_linux/1029 |
|
+1 |
|
Since @esteve gave his +1 I'll merge these PRs today unless there are any objections. |
jacquelinekay
force-pushed
the
introspection_cpp
branch
2 times, most recently
from
5672492
to
23c6170
Compare
jacquelinekay
force-pushed
the
introspection_cpp
branch
from
23c6170
to
b76d1f2
Compare
Select between introspection_c and introspection_cpp
|
🎆 🎉 🎈 |
These changes are necessary to use rcl with Connext Dynamic.