Add trailing nonce to null-flushed pipes (fixes #17)

[TinoDidriksen]

Works around issue #17 by adding a trailing nonce. The nonce is cut off from the text by nulls on both sides, just in case. Post-translation all nulls are removed, in case any extras got output by either previous or this run.

[coveralls]

Coverage decreased (-0.2%) to 47.764% when pulling f45d458 on nonce into 151e3e2 on master.

[coveralls]

Coverage decreased (-0.2%) to 47.764% when pulling f45d458 on nonce into 151e3e2 on master.

[coveralls]

Coverage decreased (-0.2%) to 47.764% when pulling f45d458 on nonce into 151e3e2 on master.

[unhammer]

Is there any chance the nonce could be eaten by a pipeline? In that case, the pipeline will hang and further requests are blocked. I tested this by making a script

#!/bin/bash
i=0
while read -r -d '' line; do
    if [[ $(( ++i )) -gt 3 ]]; then
        messedup=${line//NONCE/}
        printf "%s\0" "${messedup}"
    else
        printf "%s\0" "${line}"
    fi
done

and putting it in the middle of a pipeline and running servlet and curling against my modified pipeline – the second request then just hangs, and all further requests to that pipeline also hang.

This was probably already a problem c.f. comment

apertium-apy/apertium_apy/utils/translation.py

Lines 279 to 282 in 151e3e2

	# TODO: If the output has no \0, this hangs, locking the
	# pipeline. If there's no way to put a timeout right here, we
	# might need a timeout using Pipeline.use(), like servlet.py's
	# cleanable but called *before* trying to translate anew

(but pipelines tended to produce quite a lot of NUL's so instead of hanging they gave the wrong output?).

I added a timeout that seems to fix it here:

https://github.com/apertium/apertium-apy/compare/nonce...pr-171?expand=1

[TinoDidriksen]

If pipes start eating or modifying superblanks, we have larger issues. But yes, a timeout is great - I just didn't want to mess with more Python than absolute minimum.

[unhammer]

rebased and merged as 4188445 :)