NTF Generic Attribute Reading: Offset at end of record treated as error instead of simply as end of data

[molnar]

This commit guards against overrunning the attribute record buffer, which is good: ​https://github.com/OSGeo/gdal/commit/9c3a4a49a3d95cf9d122dee4449308e7e57b98e9

However, I think this commit may be a bit overzealous in safety: ​https://github.com/OSGeo/gdal/commit/2b3f5f329d671e8d0a94a73ad6920f391cec1228

See line 953

iOffset = nEnd;
if( iOffset >= poRecord->GetLength() )
{
   bError = true;
   break;
}

I believe an offset that is equal to the record length should simply break the loop, and should not be an error. Greater offsets may still be an error.

I might suggest:

iOffset = nEnd;
const int nRecordLength = poRecord->GetLength();
if( iOffset >= nRecordLength )
{
   bError = (iOffset > nRecordLength);
   break;
}

or something similar. Attaching a repro dataset

[molnar]

Repro dataset

[Even Rouault]

In 41288:

NTF: fix regression introduced by fix for ​https://bugs.chromium.org/p/oss-fuzz/issues/detail?id=2166 causing some valid records to be skipped (patch by molnar, fixes #7204)

[Even Rouault]

In 41289:

NTF: fix regression introduced by fix for ​https://bugs.chromium.org/p/oss-fuzz/issues/detail?id=2166 causing some valid records to be skipped (patch by molnar, fixes #7204)

[Even Rouault]

In 41290:

NTF: fix regression introduced by fix for ​https://bugs.chromium.org/p/oss-fuzz/issues/detail?id=2166 causing some valid records to be skipped (patch by molnar, fixes #7204)