[Difx-users] Handling lin-pol data with difx2fits

[harro verkouter]

Dear Leonid, DiFX users,

I would avoid applying this specific patch because it introduces a subtle
bug, which may result in a segmentation fault at best and producing
incorrect polarization label for polarization B without crashing at worst.

This specific piece of the code exposes a subtlety in the C/C++ languages.
The effect is that the second
+       if ( D->nPolar == 1 )
+       {
+             polTypeA = D->polPair[0];
+             polTypeB = '?'; /* Mark missing polarization */
+        }
Note the missing "else" here. The is syntactically valid C/C++ but the the
net effect is that, because there is no "else" keyword, the following piece
of code:
+        {
+             polTypeA = D->polPair[0];
+             polTypeB = D->polPair[1];
+        }

will be executed *unconditionally*. Syntactically speaking, the compiler
only sees the open/close braces as starting a new, temporary, context,
without any conditions. All statements inside the block are executed.
Thus, if D->nPolar is, in fact, just 1 (single polarization) then the
variable polTypeB will be unconditionally overwritten with whatever
D->polPair[1] contains.

If you're lucky D->polPair is length one in that case and you're addressing
outside array bounds and *maybe* get a segmentation violation.
But chances are better than average that that memory location is not
initialized or contains unknown content which is unconditionally
transferred to polTypeB.

Cheers,
h



On Thu, 11 Apr 2019 at 05:41, Leonid Petrov <Leonid.Petrov at lpetrov.net>
wrote:

> Dear colleagues,
>
>   DiFX has a bug and does not propagate polarization labels to the
> output FITS-IDI. The fix is rather trivial.
>
>   However when I discussed this issue with Rick Perley, he pointed out
> that another fix is needed as well. According to famous AIPS 114 memo,
> page 8, horizontal and vertical polarizations shall be labeled X and Y:
>
> X Horizontal linear
> Y Vertical linear
>
> i.e. fixed in the topo-centric, site-dependent coordinate system.
>
>   This is contrary to the IAU 1976 resolution that defines X polarization
> along direction to the Northern Pole and Y polarization along the
> right ascension axis (East), i.e. rotating in the topo-centric,
> site-dependent coordinate system.
>
>   I suggest to rename the polarizations recorded at the VLBI stations as
>
> H Horizontal linear
> V Vertical linear
>
>   and reserve X and Y labels for space telescopes.
>
>   In addition to just complying with the IAU nomenclature, it reduces
> chances
> of misunderstanding in communication with station personnel: who knows
> which
> polarization a station guy considers as "X", but unless he is operating an
> orbiting VLBI telescope, horizontal and vertical polarizations are
> unambiguous.
>
>   Any objections? Other thoughts?
>
>   Since we only started using FITS-IDI data with lin-pol, this change will
> be relatively painless.
>
> Leonid
> 2019.04.10_23:34:51
>
> ------------------------------------ Patch
> ---------------------------------
>
> --- applications/difx2fits/src/fitsAN.c~        2019-04-06
> 21:14:14.733263707 -0400
> +++ applications/difx2fits/src/fitsAN.c 2019-04-06 21:44:13.932896491 -0400
> @@ -106,8 +106,22 @@
>         stop  = D->mjdStop  - (int)D->mjdStart;
>
>         arrayId1 = 1;
> -       polTypeA = 'R';
> +       polTypeA = 'R'; /* Default */
>         polTypeB = 'L';
> +       if ( D->nPolar == 1 )
> +       {
> +             polTypeA = D->polPair[0];
> +             polTypeB = '?'; /* Mark missing polarization */
> +        }
> +        {
> +             polTypeA = D->polPair[0];
> +             polTypeB = D->polPair[1];
> +        }
> +        if ( polTypeA == 'X' ) { polTypeA = 'H'; } /* Rename linear
> polarizations */
> +        if ( polTypeA == 'Y' ) { polTypeA = 'V'; }
> +        if ( polTypeB == 'X' ) { polTypeB = 'H'; }
> +        if ( polTypeB == 'Y' ) { polTypeB = 'V'; }
> +
>         time = 0.5 * (stop + start);
>         timeInt = stop - start;
>         for(bandId = 0; bandId < nBand; ++bandId)
>
> _______________________________________________
> Difx-users mailing list
> Difx-users at listmgr.nrao.edu
> https://listmgr.nrao.edu/mailman/listinfo/difx-users
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listmgr.nrao.edu/pipermail/difx-users/attachments/20190411/73fd8199/attachment.html>