[
https://issues.apache.org/jira/browse/THRIFT-2717?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14157504#comment-14157504
]
Konrad Grochowski commented on THRIFT-2717:
-------------------------------------------
1. I'll try to remember that - I've not added this conditional compilation yet,
as I believe that before I finish with generator we'll probably switch to cmake
and I'd have to add it once more ;) But I do remember to keep new generator as
separated as possible. It will probably end in own lib, which will be added
when system compiler is new enough. I'm not going to force everyone out of a
caves :)
2. I see it that way - because now generator will try to use old lib, generated
code will still depend on boost (shared_ptr), but when new library will be
developed, it could introduce thrift::optional and allow generated code to move
away from boost. In the mean time I try to keep 'optional fields generation'
code isolated and easy to substitute / configure.
> C++11 generator
> ---------------
>
> Key: THRIFT-2717
> URL: https://issues.apache.org/jira/browse/THRIFT-2717
> Project: Thrift
> Issue Type: New Feature
> Components: C++ - Compiler
> Reporter: Konrad Grochowski
>
> instead of adding another set of options to 'old' cpp generator I've started
> creating new one in:
> https://github.com/hcorg/thrift/tree/cpp11_generator
> using old as an reference
> main goals:
> * code compatible with old librart (at least for first tests, new lib and
> compiler switches can be added later)
> * no more ugly {{__isset}} structure -> boost::optional for optional values
> * as a result - no more {{__}} in names, which violates C++ standard
> * all generation code will have own unit tests (TDD used wherever possible)
> * generated types headers independent from Thrift header, to allow other
> layers of application using generated types without dependency leaks
> * each type will generate own header/cpp file - easier for user to include
> only used parts.
> * unordered map/sets
> * returning using move semantics, no more ugly 'return via output parameter'
> (still possible as option thou - sometimes it's needed for performance)
> * async client using boost::future
> * enum classes
> * initializer lists for constants (maybe)
> I'm aiming in C++11 subset available in gcc 4.8 and MSVC 2013
> currently I have only complete enum generation, but work is in progress
> all comments etc are very welcome :)
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
