included update_warm_start functionality

[CLAassistant]

All committers have signed the CLA.

[raoul-herzog]

dear Ian,

the emosqp interface does indeed not offer the same functionality compared to the "class based" Matlab interface of OSQP.

best regards,

Raoul

De : Ian McInerney @.> Envoyé : vendredi 28 mai 2021 21:27 À : oxfordcontrol/osqp-matlab @.> Cc : Herzog Raoul @.>; Author @.> Objet : Re: [oxfordcontrol/osqp-matlab] included update_warm_start functionality (#37)

@imciner2 commented on this pull request.

---

In codegen/files_to_generate/emosqp_mex.chttps://github.com/oxfordcontrol/osqp-matlab/pull/37#discussion_r641773563:

@@ -299,6 +299,34 @@ void mexFunction(int nlhs, mxArray plhs[], int nrhs, const mxArray prhs[])

     return;

 }

• // update warm_start

• if (!strcmp("update_warm_start", cmd)) {

Actually, I may have gotten confused here because I didn't see this was the emosqp file, so this change probably isn't actually needed (although we need to document this now in https://github.com/oxfordcontrol/osqp/blob/master/docs/codegen/matlab.rst as well).

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHubhttps://github.com/oxfordcontrol/osqp-matlab/pull/37#discussion_r641773563, or unsubscribehttps://github.com/notifications/unsubscribe-auth/ATZN2G7NBHDTNKWBN53YWG3TP7U7JANCNFSM43MXTL3Q.

[imciner2]

Sorry about that confusion. My second comment on the if statement guarding the call to the actual OSQP update function still needs to be addressed though, since only checking for y non-empty seems to be incorrect. That function requires both x and y, so we should be making sure they both are non-empty for the call and throwing an error if either are empty I believe.

[raoul-herzog]

Hi Ian,

in principle a warm start with only x and without y is possible, but indeed the question is whether it makes sense ...

By the way, the same question applies to the if (!strcmp("update_bounds", cmd)) a couple of lines above. Does it make sense to provide only l without u ?

Do you plan a new release of osqp in the next future ?

best regards from Western Switzerland,

Raoul

De : Ian McInerney @.> Envoyé : dimanche 30 mai 2021 15:30 À : oxfordcontrol/osqp-matlab @.> Cc : Herzog Raoul @.>; Author @.> Objet : Re: [oxfordcontrol/osqp-matlab] included update_warm_start functionality (#37)

Sorry about that confusion. My second comment on the if statement guarding the call to the actual OSQP update function still needs to be addressed though, since only checking for y non-empty seems to be incorrect. That function requires both x and y, so we should be making sure they both are non-empty for the call and throwing an error if either are non-empty I believe.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHubhttps://github.com/oxfordcontrol/osqp-matlab/pull/37#issuecomment-851000671, or unsubscribehttps://github.com/notifications/unsubscribe-auth/ATZN2G74II4Z7HTXT7HH7XLTQI4VZANCNFSM43MXTL3Q.

[imciner2]

in principle a warm start with only x and without y is possible, but indeed the question is whether it makes sense ...

The function to call if you only want to warm start one of them is osqp_warm_start_x or osqp_warm_start_y - the function osqp_warm_start requires that both x and y be provided, so we should be checking that both x and y are non-empty vectors if we use that function.

[gbanjac]

We plan to remove osqp_warm_start_x, osqp_warm_start_y, as well as the function for updating lower and upper bounds separately. We have already done that in develop-1.0 branch of the main OSQP repository. See the commits below: https://github.com/oxfordcontrol/osqp/commit/fab90f8b138aa48c9be71c1d1ddbbe45267a1308 https://github.com/oxfordcontrol/osqp/commit/e49cf60b86f5762d02a61c14e859c052bc2c33c5

[imciner2]

Basically what I am saying we should do is have code before the call to osqp_warm_start like:

if ( mxIsEmpty(x) || mxIsEmpty(y) ) {
    mexErrMsgTxt("Both x and y must be given");
}

because otherwise we can get a segfault/random data if one isn't specified.