Add Safe::cast<T> #898
Open
Add Safe::cast<T> #898
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Codecov Report
@@ Coverage Diff @@
## master #898 +/- ##
==========================================
+ Coverage 71.09% 71.12% +0.03%
==========================================
Files 148 148
Lines 19376 19400 +24
==========================================
+ Hits 13775 13799 +24
Misses 5601 5601
Continue to review full report at Codecov.
|
|
Gosh, Dan. That's amazing code and I have no idea what it does! I'm going to leave it to @piponazo to approve (although I will approve if you ask me). I got it to built and pass the test suite with msvc 2017/Release and 2019/Release (with a little code change). When I built, I got this error: My "fix" is to comment off some code in safe_op.hpp My reason to try thisYou have two definitions for the template: This might be ambiguous. Other ideas:I tried some other guesses and none of them compiled. For example, I changed My Suggestions
I suspect a lot of the template voodoo comes from the SDK and not the compiler. As CMake's announcing the SDK, it's likely that we could force him to use a different SDK with different consequences. |
| cast(U value) | ||
| { | ||
| // U signed, T unsigned => T_MAX < U_MAX | ||
| return (value <= std::numeric_limits<T>::max()) && (value >= std::numeric_limits<T>::min()) |
There was a problem hiding this comment.
I think it is possible that T_MAX > U_MAX here, because this code could be used for casts like int8_t to uint32_t. I think the code already works correctly, because value will get automatically promoted to type T in the comparison. But I would recommend swapping the order to first check that value isn't negative:
0 <= value && value <= std::numeric_limits<T>::max()
piponazo
previously approved these changes
Jun 18, 2019
|
Robin Mills <[email protected]> writes:
Gosh, Dan. That's amazing code and I have no idea what it does!
That's usually a bad sign.
I'm going to leave it to @piponazo to approve (although I will approve if you ask me).
I got it to built and pass the test suite with msvc 2017/Release and 2019/Release (with a little code change).
When I built, I got this error:
```
pngchunk_int.cpp
c:\msys64\home\rmills\gnu\github\exiv2\master\src\safe_op.hpp(347): error C2064: term does not evaluate to a function taking 0 arguments (compiling source file C:\msys64\home\rmills\gnu\github\exiv2\master\src\pngchunk_int.cpp) [C:\msys64\h
ome\rmills\gnu\github\exiv2\master\build\src\exiv2lib_int.vcxproj]
c:\msys64\home\rmills\gnu\github\exiv2\master\src\safe_op.hpp(348): error C2064: term does not evaluate to a function taking 0 arguments (compiling source file C:\msys64\home\rmills\gnu\github\exiv2\master\src\pngchunk_int.cpp) [C:\msys64\h
ome\rmills\gnu\github\exiv2\master\build\src\exiv2lib_int.vcxproj]
c:\msys64\home\rmills\gnu\github\exiv2\master\src\safe_op.hpp(348): error C2039: 'type': is not a member of 'std::enable_if<false,void>' (compiling source file C:\msys64\home\rmills\gnu\github\exiv2\master\src\pngchunk_int.cpp) [C:\msys64\h
ome\rmills\gnu\github\exiv2\master\build\src\exiv2lib_int.vcxproj]
C:\Program Files (x86)\Microsoft Visual Studio\2017\Community\VC\Tools\MSVC\14.16.27023\include\type_traits(227): note: see declaration of 'std::enable_if<false,void>' (compiling source file C:\msys64\home\rmills\gnu\github\exiv2\master\src\pngchunk_int.cpp)
c:\msys64\home\rmills\gnu\github\exiv2\master\src\safe_op.hpp(348): error C2146: syntax error: missing '>' before identifier 'type' (compiling source file C:\msys64\home\rmills\gnu\github\exiv2\master\src\pngchunk_int.cpp) [C:\msys64\home\r
mills\gnu\github\exiv2\master\build\src\exiv2lib_int.vcxproj]
```
My "fix" is to comment off some code in safe_op.hpp
```
namespace Internal
{
template <typename from, typename to, typename = void>
struct is_safely_convertible : std::false_type
{
static_assert(std::is_integral<from>::value&& std::is_integral<to>::value,
"from and to must both be integral types");
};
/*
template <typename from, typename to>
struct is_safely_convertible<
from, to,
typename std::enable_if<((std::numeric_limits<from>::max() <= std::numeric_limits<to>::max()) &&
(std::numeric_limits<from>::min() >= std::numeric_limits<to>::min()))>::type>
: std::true_type
{
};
*/
```
This is unfortunately not really a fix, as it just comments out the
problematic part. The compiler error goes away, but the intended
functionality is lost.
#### My reason to try this
You have two definitions for the template:
```
template <typename from, typename to, typename = void>
and
template <typename from, typename to>
```
This might be ambiguous.
That's kind of the point. What I'm doing here is called SFINAE
(Substitution Failure Is Not An Error), which means that the compiler
does not error out immediately when fails to insert a type in a template
overload.
Let's say you have two template overloads of the function foo:
template <typename T, typename = void>
void foo()
and
template <typename T>
void foo<T, typename something_that_depends_on_T<T>::type>()
When you instantiate foo, the compiler will start inserting the template
parameters starting with the most specialized overload (here the second
one). It will insert T and then try to find out what
`something_that_depends_on_T<T>::type` is. If the resolution of this
type results in an error, then this is **not** fatal. The compiler
simply tries the next less specialized template overload.
You can use this in conjunction with "dummy" template parameters
(template parameters that have a unused default value and are not
actually used in the template) to have multiple specializations of the
same function. You then provide a conditional type
(`something_that_depends_on_T<T>::type` in the above example) that
resolves only in the case when you want the overload and doesn't either.
#### Other ideas:
I tried some other guesses and none of them compiled. For example, I changed `::type` to `::typename` because it complained:
```
safe_op.hpp(348): error C2146: syntax error: missing '>' before identifier 'type'
```
Yes, that cannot work, because type is a public member of
std::enable_if.
#### My Suggestions
1) is `type` correct on line 348? For sure, I don't know what it means and apparently the compiler doesn't either!
It is definitely correct. GCC and clang don't complain about this and
work without problems.
2) Only compile `template <typename from, typename to> .... };` for `#ifndef _MSC_VER`. You'll loose safety for MSVC builds. It could be a compiler bug that goes away in the future. However it's there with both 2017 and 2019.
Not including the overload for MSVC seriously cripples the overall
usefulness of the whole code. I'd prefer not to do that unless MSVC
really doesn't cooperate.
3) It could be an SDK issue. For both 2017 and 2019, CMake reports:
```
Selecting Windows SDK version 10.0.17763.0 to target Windows 10.0.17134.
```
I suspect a lot of the template voodoo comes from the SDK and not the compiler. As CMake's announcing the SDK, it's likely that we could force him to use a different SDK with different consequences.
Unlikely, as this uses only std::enable_if and std::numeric_limits (the
former can be implemented in ~5 lines of code and the latter is also
very simple, albeit requiring a lot of boilerplate code).
… --
You are receiving this because you authored the thread.
Reply to this email directly or view it on GitHub:
#898 (comment)
|
Merged
Open
tbeu
reviewed
Oct 13, 2019
Comment on lines
+36
to
+41
|
|
||
| // MSVC is stupid and pollutes the global namespace with max() and min() macros | ||
| // that break std::numeric_limits<T>::max() and min() | ||
| #undef max | ||
| #undef min | ||
|
|
There was a problem hiding this comment.
We should define NOMINMAX, see https://stackoverflow.com/questions/4913922/possible-problems-with-nominmax-on-visual-c.
There was a problem hiding this comment.
It is actually already defined in cmake/compilerFlags.cmake. Although it seems it is only defined for MSVC. @D4N maybe you had issues with MinGW ?
|
Let's see if we can get this magic into v1.00. |
clanmills
added
the
refactoring
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR adds a new function into the
Safenamespace:cast<T>. It is intended to convert a integer from one type to the other without causing over- or underflows.Thanks to some ternary operator abuse it is even completely
constexpreven with C++11, although readability suffers a little.Please review this PR with more care, I've created it in a pretty sleep deprived state.