I have rna-seq data (18 samples) from which I produced the raw counts using htseq-count. Now I want to analyze the data using edgeR. I've done this dozens of times and had no issues. Not this time. When I execute

samples <- read.csv(file="metadata_composite.csv",header=TRUE,sep=",")
counts <- readDGE(samples$CountFiles)$counts

I get:

Error in taglist[[i]] : subscript out of bounds

I've analyzed the same data using Deseq2 and had no issues whatsoever. So it is something related to edgeR. Here is my data - it consists of 18 directories each containing a count file (accepted_hits.count) and a directory called edgeR which contains "metadata_composite.csv" which is the metadata file that I load in in the code snippet above. 

Can you tell what is the problem?

You've got a couple of duplicated entries in samples$CountFiles. Each entry of taglist is named according to the count file, so if you have the same count file for different samples, information from the later samples will overwrite that of earlier samples instead of forming a new entry in taglist. The offending samples seem to be WT-H2 and WT-H3 in your CSV file, both of which refer to the WT-H1 directory to extract the count file. Redirect these to get the count files from their appropriate directories, and you should be fine.

To add to Aaron's post, there are also a couple of tunes that you need to use to accommodate htseq output. First, htseq output has no headers, so

dge <- readDGE(samples$CountFiles, header=FALSE)

Second, the last 5 lines of htseq output are not real genes, so you need to remove them:

realgene <- grep("^ENS",rownames(dge))
dge <- dge[realgene,]