Fix marshalling of bool flags in MF #3210
Conversation
Codecov Report
@@ Coverage Diff @@
## master #3210 +/- ##
==========================================
+ Coverage 72.61% 72.62% +0.01%
==========================================
Files 807 807
Lines 145077 145080 +3
Branches 16213 16213
==========================================
+ Hits 105345 105365 +20
+ Misses 35316 35297 -19
- Partials 4416 4418 +2
|
| @@ -133,19 +133,19 @@ private struct MFParameter | |||
| /// Specify if the factor matrices should be non-negative. | |||
| /// </summary> | |||
| [FieldOffset(48)] | |||
| public int DoNmf; | |||
| public byte DoNmf; | |||
There was a problem hiding this comment.
While this is better than what we had before, this is still not correct.
https://docs.microsoft.com/en-us/cpp/cpp/fundamental-types-cpp?view=vs-2019
bool Type bool is an integral type that can have one of the two values true or false. Its size is unspecified.
https://en.cppreference.com/w/cpp/language/types
bool - type, capable of holding one of the two values: true or false. The value of sizeof(bool) is implementation defined and might differ from 1.
Instead, to make interop/marshalling work correctly on all computers - the technique we use in .NET is to ensure the ABI (application binary interface) is stable no matter the compiler/processor/etc. This means only stable-sized types can be used in the ABI that C# code P/Invokes into.
So to fix this, the libmf C code should change to have a stable ABI. I see it uses float, int, and bool. For sure int and bool can change sizes depending on the compiler/processor. Instead, we should be using int32_t and int8_t on the C side. (Note: you can just have a wrapper export function in the C library that exposes the stable ABI and calls into your current code. So the main C/C++ code can remain unchanged.) #Closed
There was a problem hiding this comment.
Thank you for pointing this out. Another iteration is pushed. Hopefully, it can address this comment.
| @@ -133,19 +133,19 @@ private struct MFParameter | |||
| /// Specify if the factor matrices should be non-negative. | |||
| /// </summary> | |||
| [FieldOffset(48)] | |||
| public int DoNmf; | |||
| public sbyte DoNmf; | |||
There was a problem hiding this comment.
Make it uint8_t on the C side then?
Signed-ness doesn't really matter here. So using byte is more appropriate.
There was a problem hiding this comment.
Ok. I will change other place to match this decision.
| @@ -9,25 +9,48 @@ | |||
|
|
|||
| using namespace mf; | |||
|
|
|||
| EXPORT_API(mf_parameter) make_param(const mf_parameter_bridge *param_bridge) | |||
There was a problem hiding this comment.
I don't think this needs EXPORT_API does it?
| param.eta = param_bridge->eta; | ||
| param.alpha = param_bridge->alpha; | ||
| param.c = param_bridge->c; | ||
| param.do_nmf = static_cast<bool>(param_bridge->do_nmf); |
There was a problem hiding this comment.
Instead of static cast, would it be better to use param_bridge->do_nmf != 0 ? true : false?
|
I will make change to |
ghost
locked as resolved and limited conversation to collaborators
Fix #3003. The marshalled Boolean size from C# to C should be 1 byte, not 4. Also, the change from LIBMF is for supporting quiet option in new implicit-feedback MF.