rose arch: files with identical checksums get missed

[dpmatthews]

If you have multiple files with identical checksums then rose arch will only archive one of them. The following suite.rc illustrates the problem:

[scheduling]
    [[dependencies]]
        graph = setup => rose_arch_test
[runtime]
    [[setup]]
        script = """
                 mkdir -p ${CYLC_SUITE_RUN_DIR}/app/rose_arch_test
cat >${CYLC_SUITE_RUN_DIR}/app/rose_arch_test/rose-app.conf  <<EOF
[arch]
command-format=echo Source: %(sources)s Target: %(target)s
source-prefix=${CYLC_SUITE_SHARE_DIR}/
[arch:data1.tar.gz]
source=data/*
[arch:data2.tar.gz]
update-check=mtime+size
source=data/*
EOF
                 mkdir -p ${CYLC_SUITE_SHARE_DIR}/data
                 touch ${CYLC_SUITE_SHARE_DIR}/data/empty{1,2}.txt
                 echo Hello >${CYLC_SUITE_SHARE_DIR}/data/hello1.txt
                 echo Hello >${CYLC_SUITE_SHARE_DIR}/data/hello2.txt
                 """
    [[rose_arch_test]]
        script = "rose task-run -v"

If you run this workflow (with either Cylc 7 or 8) you find that only empty2.txt and hello2.txt get added to data1.tar.gz whereas data2.tar.gz contains all 4 files.