The is a C API for SAM/BAM as well as a Java API (see the simple example) reading the files and iterating over the records. I think that all the parameters you described are part of the SamRecord class.

I wouldn't assume everybody knows R/BioConductor, but if not its worth learning to know it. The Rsamtools package might be a little bit more high-level API. In connection with the GenomicRanges package you can read in a BAM file into a gapped alignment object. Haven't tried this myself, so I don't know if this gives you the statistics you need. If it does, it's likely that the function unique would work on a an alignment object. Worth giving it a look?

Edit:

I just became aware that I wasn't really answering your question, because despite it's name R>sam<tools:

The main purpose of the Rsamtools package is to import BAM files into R. Rsamtools also provides some facility for file access such as record counting, index file creation, and filtering to create new files containing subsets of the original.

Rsamtool does not read SAM files. However TAB-separated SAM files can be read in very efficiently using scan in R. The first entry of the list will then contain the names of the reads/paired reads. To find the unique/non-unique hits ratio shoul be therefore something like:

ncol <- 15 # adjust to number of cols in your file
# adapt the what part to fit your file:
mysam <- scan("myfile.sam", what = list(rep(character(), ncol)), fill=TRUE)
unique.ratio <- length(unique(mysam[[1]]))/ length(mysam[[1]])

To filter out spliced hits vs. non-spliced hits, one has to know how this would be encoded in the tophat output. I am not sure if and how this is marked, e.g. in the flags, in the SAM file.

Regarding the Spliced matches according to your example: looks as if it can be inferred from the CIGAR string in column 6?

40M1116N10M = 40 Matches, 1116 bases ignored, 10 Matches

That would then be an intron of length 1116. Therefor, we have to filter the CIGAR string, that must a least contain 1 N. That can be done with a regexp using grep:

cigar <- mysam[[6]]
intron.ind <- grep("\\d+M\\d+N\\d+M", cigar) # greps M N M patterns
n.intron.matches <- length(intron.ind)

Hope this works, and solves all of the questions (and shows that R rules on the way).

I wrote this simple python script. It works for accepted_hits.sam output from Tophat.

#!/usr/bin/env python

import sys

sam_fn = sys.argv[1]

#init dictionaries:
spliced   = {}
unspliced = {}

for i in range(0,7):
    spliced[i]    = 0
    unspliced[i]  = 0

for line in open(sam_fn):
    if line[0] == "@":
        pass
    else:
        sl = line.split()
        cigar, missmatch_str = sl[5], sl[11]
        num_mismatches = int(missmatch_str[-1])

        if len(cigar) != 3:
            spliced[num_mismatches]   += 1  
        else:
            unspliced[num_mismatches] += 1

print "#processed_file:", sam_fn
print "#unspliced, spliced"
for i in range(0,7):
    print unspliced[i], spliced[i]

print "###########################################################"