I am using GATK 3.5 to call SNPs/indels.
For some variants, the DP argument is ".", but the GQ is not ".". For example in the second sample below, the "1/1" genotype is called with a depth of zero and a high confidence.
chr1 13859273 . C T 21107.89 . AC=18;AF=0.818;AN=22;BaseQRankSum=1.113;ClippingRankSum=0.429;DP=398;FS=0.000;InbreedingCoeff=0.9591;MQ=60.00;MQ0=0;MQRankSum=-0.139;QD=35.55;ReadPosRankSum=5.678;SOR=3.213 GT:AD:DP:GQ:PL 1/1:0,28:28:84:1260,84,0 1/1:.:.:99:2004,138,0
You must have cut off more columns, because AC=18 and AF=82% imply a lot of samples show the T allele.
The genotyper you're using is designed for population studies. If someone has no data, it could be called the most common genotype. Or maybe you're using a family study and told the software of the pedigree, so it can be reasonably certain the child has the same genotype as the parents. The GATK is complicated like that.
If you wanted to go by read-depths alone, use a different tool. Samtools mpileup works on read counts and would give the expected NOCALL allele.
There could be a command line parameter to GATK that is set incorrectly. Please give the command used so we can see the parameters.