When invoked with multiple dictionaries, suggestions are only made using the dictionary in which a word was last found. This logic is totally unsuitable for vocabulary lists lacking effective language tagging, for the language is likely to alternate between the two languages. It also appears to be unsuitable for specialist dictionaries, for they are unlikely to be consulted for suggestions for misspelt technical vocabulary.
I attach a correction for the pipe invocation (hunspell -a) in function pipe_interface() in src/tools/hunspell.cxx. This includes, for this logic branch at least, corrections for Bugs 3468022 and 3178449. The starting point is the version in Release 1.3.2.