[gbouras13]

Refs #92. This is the plassembler side. The root cause in the image is BusyBox's expr rejecting the expr -- that canu's job scripts use, and that is being fixed separately by adding coreutils to the bioconda recipe. So this PR doesn't close the issue.

Problem

Any non-zero exit from canu -correct was reported as:

Canu failed to correct any reads. This probably means there is low depth, don't be too concerned.

and plassembler went on with uncorrected reads. In #92 canu could not run at all: meryl-count failed on every attempt. Going on with uncorrected reads left Unicycler assembling 110 spurious contigs, which the depth filter then had to discard. Once canu worked there were none, and the run took 2.5× less time.

Change

canu ends every failed run with the same ABORT: Don't panic … banner, including the low-coverage stop that a plasmid-free sample is expected to hit. So the banner can't decide. The line before it can:

canu prints	meaning	now
-- <stage> jobs failed, tried N times, giving up.	a compute job never produced its output	fatal
a CRASH: banner	canu's own internal error	fatal
Read coverage (N) lower than allowed.	too few reads	carries on, saying so
anything else	not recognised	carries on, showing canu's stderr

This matches failure lines rather than whitelisting harmless stops. canu has ~200 caExit sites, and missing one harmless stop would turn plasmid-free samples into failed runs.

On the fatal path the canu directory is kept, because the run exits before remove_directory. The message includes the tail of canu's stderr, which names each job's own log. For #92 that is correction/0-mercounts/meryl-count.000001.out, which contains expr: syntax error.

canu now runs directly from run_canu.py, the same way qc.py handles chopper. The two canu special cases in ExternalTool.run_tool are removed.

Verification

I ran real canu 2.3 with plassembler's exact arguments on tests/test_data/end_to_end/abaumanii_plasmid.fastq.gz:

scenario	exit	job-failure line	low-coverage line
normal	0	–	–
expr wrapper rejecting -- like BusyBox	1	✓	–
50 reads (3.7×)	1	–	✓
empty input	1	–	✓
20 reads at the fallback genomeSize	0	–	–
30× non-overlapping random reads	0	–	–

None of the thin-data cases prints the job-failure line. I also drove the fixed run_canu_correct against real canu under the CLI's exiting ERROR sink. The BusyBox case exits 1, low coverage warns and carries on, and a normal run is silent. In canu's source the three lines are identical in 2.2 and 2.3.

tests/test_run_canu.py adds 12 tests that replay real canu 2.3 stderr (fixtures in tests/test_data/canu/) through a stub canu. 10 of them fail against main, and the other 2 are the success cases, which are meant to pass on both. The non-slow suite passes (102), and ruff format --check and ruff check are clean. I didn't run the slow tests locally, because this machine has no bioinformatics tool stack.

Also bumps the version to 1.8.6, with release notes crediting the reporter.