Out of bounds access in DoGFF()

[kettenis]

Hi Ivan,

Suma Murthy is hitting an out-of-bounds array access in DoGFF() when trying to PolConvert some EVN data. This happens at:

PolConvert/_PolGainSolve.cpp

Line 1533 in 040c093

AbsP = std::abs(Temp[m][time[m][l]*Nchan[i] + nu[m][k]]);

There is some guesswork involved, but I think time[n][1] is Nchan[i]-1 here because we hit:

PolConvert/_PolGainSolve.cpp

Line 1522 in 040c093

if (time[m][1]==0){time[m][0]=Nchan[i]-1;} else{time[m][0]=time[m][1]-1;};

This results in an array index of the order of Nchan[i]*Nchan[i] which is clearly larger than intended. So I think there is something wrong in the logic around L1522. The reason Suma hits this here may very well be because this data set has 16K channels. It may very well be that you don't hit this with data sets with less channels. But it would still do the wrong thing.

[marti-vidal-i]

Hi Mark! Indeed, Geoff and myself were working on the code in July and touched some parts related to array sizes. At a first glance, I'm afraid that something got broken on DoGFF, but I'd need to look at the changes to be sure. Next week, I'll be working on a regression test of EU-VGOS data (where the problem should also arise). I'll keep you informed. Anyway, I've been chatting with Benito in Tenerife about the issues you are finding with the conversion (e.g., data with high spectral resolution). I think all these cases and experiences are very useful to improve the code (so please tell me when you find them! :D ). It would be very nice if we could sit down together for a week, as soon as we'd have time, and process/assess completely some datasets!    Cheers!          Ivan El 8/9/22 a las 14:26, kettenis escribió:

…

Hi Ivan, Suma Murthy is hitting an out-of-bounds array access in DoGFF() when trying to PolConvert some EVN data. This happens at: https://github.com/marti-vidal-i/PolConvert/blob/040c09388cafe9b0a0df08b3e99b83d298c6d784/_PolGainSolve.cpp#L1533 There is some guesswork involved, but I think |time[n][1]| is |Nchan[i]-1| here because we hit: https://github.com/marti-vidal-i/PolConvert/blob/040c09388cafe9b0a0df08b3e99b83d298c6d784/_PolGainSolve.cpp#L1522 This results in an array index of the order of Nchan[i]*Nchan[i] which is clearly larger than intended. So I think there is something wrong in the logic around L1522. The reason Suma hits this here may very well be because this data set has 16K channels. It may very well be that you don't hit this with data sets with less channels. But it would still do the wrong thing. — Reply to this email directly, view it on GitHub <#7>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ATYO7CSKZQVSD3UBHIUCX4TV5HLQTANCNFSM6AAAAAAQHWD3AM>. You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

--

-------------------------------------------- || Dr. Ivan Marti-Vidal                   || || GenT Distinguished Researcher          || -------------------------------------------- || Departament d'Astronomia i Astrofísica || || Universitat de València                || || Office 3.08                            || || C/ Dr. Moliner 50                      || || 46100 Burjassot (València, Spain)      || -------------------------------------------- || E-mail: ***@***.***            || || Phone: +34 963 543 078                 || || Skype: marti-vidal                     || -------------------------------------------- ---------------------------------------- Webpage: http://mural.uv.es/imarvi ---------------------------------------- PGP Key ID:    0xC5B38704 ORCID:         0000-0003-3708-9611 Researcher ID: A-8799-2017 ---------------------------------------- -