RPKM/FPKM, TPM and raw counts for RNA-seq

There was a hot discussion that RPKM should not be used in the RNA-seq analysis:
blahah404  Woohoo - death to FPKM and RPKM!t.co/CyR3ht1dqi

However @BioMickWatson has a different opinion, he thinks FPKM makes sense.
Why FPKM makes sense

what is RPKM or FPKM? see blog posts below:
update on 07/23/2015, see a youtube video  for great explanation of the three.
What is the FPKM? A review of RNA-Seq expression units
Counts vs. FPKMs in RNA-seq
Paper: RPKM measure is inconsistent among samples
FPKM/RPKM normalization caveat and upper quartile normalization
Finally a post from Lior Pachter
Estimating number of transcripts from RNA-Seq measurements (and why I believe in paywall)

I personally use raw counts and let DESeq2 to deal with normalization.

update on 07/08/2016.
Salmon and Kallisto outputs TPM, but if you want to convert counts to TPM, read this biostar post, and my post comparing salmon, kallisto and HTseq.

FPKMs and counts compared for RNA-seq data

A very nice article comparing  FPKMs and raw counts in RNA-seq data analysis

http://www.cureffi.org/2013/09/12/counts-vs-fpkms-in-rna-seq/
"

1. counts are simply the number of reads overlapping a given feature such as a gene.
2. FPKMs or Fragments Per Kilobase of exon per Million reads are much more complicated.  Fragment means fragment of DNA, so the two reads that comprise a paired-end read count as one.  Per kilobase of exon means the counts of fragments are then normalized by dividing by the total length of all exons in the gene (or transcript).  This bit of magic makes it possible to compare Gene A to Gene B even if they are of different lengths.  Per million reads means this value is then normalized against the library size.  This bit of magic makes it possible to compare Gene A in Sample 1 to Sample 2 even if Sample 1′s RNA-seq library has 60 million pairs of reads and Sample 2′s library has only 30 million pairs of reads. "