Fix wrapping of %define and #define expressions containing chars

[nunoguedelha]

This PR merges the fix swig#781, after the discussion on the issue regarding the wrapping of global defines and enums having char type elements, or being expanded with char type inputs. The chars would lose the quotes and become variables.

[nunoguedelha]

Implementation

@traversaro , this PR is a result of 2 steps:

1. I first pushed to this repo the branch update/matlab2jaeandersson aligned with the latest commit on the https://github.com/jaeandersson/swig.git matlab branch. I suggest we later just do a fast-forward of our own matlab branch, as soon as tests are complete.
2. I then "rebased" the fix Fix enum and #define error when value contains char in compound expression swig/swig#781 (9 commits, from e1afb6f to 9b85318) on to the branch update/matlab2jaeandersson.

Tests

The tests on simple examples work fine:

%define VOCAB(x1,x2,x3,x4) x4*16777216+x3*65536+x2*256+x1
#define VOCAB_FRAMEGRABBER_IMAGE        VOCAB('a','b','c','d')

generates:

SWIGINTERN int swigConstant(int SWIGUNUSEDPARM(resc), mxArray *resv[], int argc, mxArray *argv[]) {
...
  case 6: *resv = SWIG_Matlab_SetConstant(module_ns,"VOCAB_FRAMEGRABBER_IMAGE",SWIG_From_int(static_cast< int >('d'*16777216+'c'*65536+'b'*256+'a')));; break;

But it's failing with the yarp-matlab-bindings files. Checking again now it's not an issue with the install.

[traversaro]

But it's failing with the yarp-matlab-bindings files.

I think we can also just try to merge the mastlab branch directly with the latest version of SWIG, if it is easier.

[nunoguedelha]

The bindings generation is working fine, the problems I had earlier were probably due to leftover temporary objects involved in the SWIG processing. After rebuilding yarp-matlab-bindings the VOCABS enums and global defines were properly generated with the correcct values. I've also checked on Matlab, by comparing:

• the yarp.VOCAB_XXXXXXX constant value
• the formula x4*16777216+x3*65536+x2*256+x1 result
• the result of yarp.Vocab.encode(<string>)
  For instance :

>> yarp.Vocab.encode('vel')
ans =
  int64
   7103862
>> 0*16777216+'l'*65536+'e'*256+'v'
ans =
     7103862
>> yarp.VOCAB_CM_VELOCITY
ans =
  int64
   7103862

[nunoguedelha]

Last updates on the PR:

"Our" matlab branch was pointing to the commit we've been using for a while now (260ed47).
I've fast-forwarded that branch to update/matlab2jaeandersson and updated the PR accordingly, which just changed the PR base name, not the diff obviously.

So we just need to merge the PR to matlab.

[nunoguedelha]

I had an issue with a missing SwigClear() method in the SwigRef, the superclass of every class wrapped by SWIG.
The SWIG commit cd3f6c5 - Using method for clearing swigPtr introduces the method SwigClear() definition in the SwigRef creation, but for some reason it's not working as expected.
Taking a closer look now...

[nunoguedelha]

@traversaro , the issue with the missing SwiClear() was due to SwigRef.m being overwritten at configuration time by these lines. I noticed that you did those changes (robotology/yarp-matlab-bindings#11) for restoring the missing subsref and subsasgn (robotology/idyntree#188), which was a requirement for iDynTree.

Since we now have and will maintain our own fork of SWIG, I'll apply the changes directly to the SWIG source files MATLAB::createSwigRef().

[nunoguedelha]

I will make the change in a different PR, in case anyone needs the fix in this current PR just by doing a single commit cherry-pick. So I'm merging this PR now.

[traversaro]

@traversaro , the issue with the missing SwiClear() was due to SwigRef.m being overwritten at configuration time by these lines. I noticed that you did those changes (robotology/idyntree#305) for restoring the missing subsref and subsasgn (robotology/idyntree#188), which was a requirement for iDynTree.

Oh fak, my fault in accumulating technical debt.

Let's proceed with your solution of patching our branch, but this will be problematic once the matlab branch will be merged in SWIG upstream. I think we should open an issue discussing robotology/idyntree#188 (comment) in either https://github.com/jaeandersson/swig or https://github.com/swig/swig .

[nunoguedelha]

I suggest we open it in https://github.com/jaeandersson/swig so that we avoid diverging from his matlab branch as much as possible. He commented on the issue linked to this PR, he would be interested in this fix, so there's a chance we don't diverge.

[traversaro]

@nunoguedelha I agree.

[traversaro]

Ok, for now opened an iDynTree bug to track the problem on iDynTree's side: robotology/idyntree#448 .