RFC 68: C++11 Compilation Mode
Author: |
Kurt Schwehr |
Contact: |
|
Started: |
2017-Apr-11 |
Passed: |
2017-Sep-11 |
Status: |
Adopted, implemented in GDAL 2.3 |
This RFC is based on GEOS RFC 5 by Mateusz Łoskot.
Summary
The document proposes to switch to C++11 compilation mode as default throughout the whole C++ source code of GDAL.
The goal of the document is to request and achieve agreement on using C++11 as the minimum required version of the C++ programming language standard.
Motivation
C++11 is the first major update of the C++ standard since 1998. (C++03 was a bug fix release.)
Having fewer versions of C++ to support will reduce the load on developers and testing systems.
C++11 features aim to promote writing clean, compact, type-safe and fast code. It also delivers better feature-wise compatibility with the C language (C99).
The Wikipedia article at http://en.wikipedia.org/wiki/C++11 does a great job describing all changes in C++11 extensively.
The std::auto_ptr
smart pointer, together with a bunch of other
features, were deprecated and will be removed from C++17. Features like
std::unique_ptr
provide much stronger replacements.
Enabling C++11 compilation mode will improve the programming environment making it much friendlier than C++98.
A social factor: since (many) C++ programmers no longer enjoy C++98, allowing C++11 mode may increase potential for new contributions.
Compilers Landscape
Summary of compilers supported by GDAL with their minimal versions required to compile source code based on
http://en.cppreference.com/w/cpp/compiler_support C++11 features.
C++11
GCC |
4.8.1+ |
Debian 8 (stable), Ubuntu 15.04+, Ubuntu 14.04 |
|
---|---|---|---|
Clang |
3.3+ |
Debian 8 (stable), Ubuntu 14.04+, Fedora 19+, CentOS 6(?) |
|
MSVC |
14.0+ (2015) |
n/a |
C++14
The C++14 compilers are listed for comparison only:
GCC |
4.9+ |
---|---|
Clang |
3.4+ |
MSVC |
14.0+ (2015) |
Plan
This proposal only requests agreement for the C++11 compilation mode
switch in the current trunk
branch only.
This proposal does not suggest any detailed roadmap of large refactoring of the GDAL C++ codebase.
The GDAL codebase is > 1.3M LOC and given the available man-power to LOCs ratio, such one-step refactoring would not be feasible.
Instead, the task will be tackled with the baby step approach gradually transforming the codebase according to priorities set along the way. Any disruptive refactoring, changes in interfaces of C++ classes, breaking changes in C++ API must be announced and discussed on the mailing list or the bug tracker.
IMPORTANT: C++11 refactoring must not change the C API or break C API compatibility, unless agreed upon based on prior RFC proposed.
However, if the proposal is accepted, any new C++ code written for GDAL must be C++11-compliant.
Prior acceptance of this proposal is necessary in order to start any source code refactoring using C++11 features.
Once accepted, first step will be to update the build configurations to require C++11-compliant compiler.
Issues
This section outlines issues potentially caused by upgrade to C++11 language.
C++11 destructors, by default, have now the new exception specification of
nothrow(true)
. Destructors of GDAL classes should be reviewed and any that are allowed/expected to throw exceptions must be marked withnothrow(false)
. Otherwise, any user of the existing GDAL codebase would find the program terminating whenever GDAL destructor throws an exception. Such review would be beneficial anyway.
Release
First release of GDAL with C++11 compiler requirement could be 2.3.0.
C++14
This section clarifies status of C++14 support in GDAL.
Once C++11 is adopted as default compilation mode, GDAL developers and maintainers must ensure it also successfully compiles in C++14 and C++17 modes.
Are contributors allowed to add
ifdef
's for C++14 and C++17? No. Not at this time.Is there a plan to upgrade to C++14 or C++17 to allow use of the C++ latest features? No, there is no plan. It is, however, recognized, such motion may be put to the vote around 2020.
References
http://en.cppreference.com/w/cpp/compiler_support C++ compiler support
Self-assigned development constraints
The changes should have moderate impact on the existing GDAL/OGR code base, and particularly on most of its code, that lies in drivers. Existing users of the GDAL/OGR API should also be moderately impacted by the changes, if they do not need to use the new offered capabilities.
GDAL has been working well with C++11 builds for a number of years now, so there should be no externally visible changes for the initial removal of C++03 support.
Core changes: summary
Change configure.ac to remove
with_cpp11
flag, always use C++11, and fail ifAX_CXX_COMPILE_STDCXX_11
does not find C++11Remove @CXX11_SUPPORT@ in GDALmake.opt.in and anywhere else it occurs
Remove continuous build targets that do not support C++11 from Travis-CI and AppVeyor
Remove #if HAVE_CXX11 and leave only the C++11 code
find . -name \*.h -o -name \*.cpp | xargs egrep 'HAVE_CXX11'
CPL_STATIC_ASSERT ->
static_assert
NULL
->nullptr
(Only for C++ code)CPL_OVERRIDE -> override and remove redundant virtual
-MAX or -max() ->
std::numeric_limits<T>::lowest()
Changes in SWIG bindings
The switch does not impact the C
API used by the SWIG bindings.
However, there may be some code that can be removed.
Potential changes that are NOT included in this RFC
There are many C++11 features not address in this RFC. See https://en.wikipedia.org/wiki/C%2B%2B11
attributes
auto
consexpr
cstdint
delete
anddefault
for member functionsenum classes
initializer lists and
std::initializer_list
lambda
range based for loops
std::regex
rvalue references
smart pointers
std::unique_ptr
andstd::shared_ptr
New string literals: u8"An UTF-8 string", u"An UTF-16 str", U"An UTF-32 str", R"xml(raw content)xml"
std::thread
,thread_local
and relatedtuples
And more...
Backward compatibility
Any code using the C++ API must use C++11 or newer.
The C API should not be impacted.
GDAL 2.2.x and older will continue to have C++03 support.
Testing
The existing autotest suite should continue to pass.
Version numbering
Although the above describes changes should have very few impact on existing applications of the C API, some behavior changes, C++ level changes and the conceptual changes are thought to deserve a 2.3 version number.
Implementation
Implementation will be done by Kurt Schwehr. Others are welcome to pitch in.
Amendments
This RFC is superseded per RFC 98: Build requirements for GDAL 3.9.
Voting history
https://lists.osgeo.org/pipermail/gdal-dev/2017-September/047139.html
EvenR +1
JukkaR +1
DanielM +0
HowardB +1
KurtS +1