Error catching

[ileppe]

• Some general error-catching functionality that pops up a window to let the user know the input was not provided correctly.
• Please see issue Error checking #156 for cases currently handled and try to break it! ;)

[tanguyduval]

load_nii handles single slice perfectly. Issue can happen if the nifti (or .mat data) were badly saved. We should add this to the doc and explain how to repair the nifti in this case.
Maybe also clarify the error in the SanityCheck if we suspect this error (e.g. if the first 2 dimensions match but not the 3rd one).

[tanguyduval]

Travis is fixed, @ileppe I'm done with this branch. I let you clarify the SanityCheck warnings for singleton dimensions if you want (see previous message). Thanks!

[mathieuboudreau]

@tanguyduval Even if load_nii currently works as expected, we should also anticipate what happens if it doesn't, like it did to @agahkarakuzu .

It should be simple to ensure that the number of dimensions matches the qMRLab expected number of dimensions right when the data is loaded, by comparing the size of the loaded data and adding a singleton dimension if missing (e.g. if size of data is 3 ([x,y,t]), add a singleton dimension ([x,y,1,t])). That way if in the future we expand i/o compatibility to other formats like MAT files/MINC/DICOM, or if a future version of load_nii breaks single slice cases for NIFTI files, we'll be prepared for it.

But we can continue that discussion on the other issue page, let's not bloat this current branch with too many unrelated changes like we did for new_doc.

[ileppe]

@tanguyduval, why did you change to break instead of return out of SanityCheck? Right now it does not exit once it finds an error, it goes on to all checks. For e.g. I would like the following code to stop execution and throw an error (i.e.missing a necessary input).

Model = vfa_t1; 
FlipAngle = [3.0000; 20.0000];
TR = [0.0150; 0.0150];
Model.Prot.VFAData.Mat = [ FlipAngle TR];
data = struct();
data.B1map=double(load_nii_data('qMRLab/Data/vfa_t1_demo/vfa_t1_data/B1map.nii.gz'));
FitResults = FitData(data,Model,0);

Instead, it goes on and crashes in the next error check. I hope that's clear.

[tanguyduval]

@mathieuboudreau if load_nii does not load correctly a single slice, we have to solve this at the source. If we handle bad input data, that might be a bit of a headache in some cases. For instance, we cannot differentiate between these two inputs: single 3D volume, and multiple 2D slices stack in the 3rd dimension. Just a feeling, I'm ok with any solution

@ileppe If an error appears, it DOES NOT go on to all checks, but breaks and throw an errordlg if no output to SanityCheck.
You are right, that does not stop the code... but that's necessary until I find a solution to the get_optional_inputson Octave (currently it believes that everything is mandatory)...

[mathieuboudreau]

@tanguyduval I agree, that's fine with me then. However, we should still investigate deeper into why @agahkarakuzu encountered this in the first place. We were lucky he's a dev that encountered this problem, if he was another user, he might not have reached out to us. We need to try to ensure that whatever lead to him having this issue is not repeated by other people after the ISMRM release, whether that be changes in the code or changes in the documentation on how to load/save data.

@agahkarakuzu let's continue this discussion on the other issue. Could you give a more detailed explanation as to how we could reproduce this error you encountered regarding singleton dimensions?

[ileppe]

@tanguyduval ok, sorry I was not clear, but the problem is that it breaks out of the for loop but does not throw an errordlg and continues to crash later on. i.e. Picks up an error here (Line 72):
https://github.com/neuropoly/qMRLab/blob/30e7603464a39e795aa8e5c14e7d16d1a3603956/Common/AbstractModel.m#L67-L76

Does not throw an errordlg and crashes here (line 79):

https://github.com/neuropoly/qMRLab/blob/30e7603464a39e795aa8e5c14e7d16d1a3603956/Common/AbstractModel.m#L79

Giving the user no info about the missing input problem.

[tanguyduval]

@ileppe argh... I see the issue, it's because of the nested loops. It does not break the parent loop, only the first loop for ii=1:length(optionalInputs) . Return was nice but we had to create the errordlg afterward which was not perfect neither. I solved this in 3a8f821

[ileppe]

@tanguyduval please check and approve this one too pls!

[agahkarakuzu]

Hooray! We are sane.

Regarding dimension issue discussed in this PR, please see this suggestion to confirm if it is also sane :)

[tanguyduval]

@ileppe looked good to me too, good job! Thanks Agah for reviewing it :)