In current apertium-fin-eng https://github.com/flammie/apertium-fin-eng using apertium-fin from github https://github.com/flammie/apertium-fin and apertium-eng_feil from apertium, the memory keeps running out for long translations.
Do e.g.,
apertium -d $HOME/Koodit/apertium-fin-eng eng-fin-debug < europarl-v8.fi-en.norm.tok.en > europarl-v8.fi-en.norm.tok.apertium.en-fi.translated.fi.debug ^Z bg top
and see:
12345 tpirinen 20 0 3866m 3,8g 1640 R 99,4 11,9 27:39.49 lt-proc
Surely it should be relatively stable for lt-proc. This is likely apertium's English analysis, since generation uses hfst-proc.
Any easy way to debug further? I'm running this on quite limited cluster node with little chance to control much, though I can test elsewhere if it don't need as much memory.
(sorry if sf.net makes the usual mess out of this message despite it looking neat in the preview again...)
Ok, here's something I found with valgrind and Tino's help:
The listing of leaks is quite long, I've attached the log, the biggest memory leaks are:
It's noteworthy that leak count is all 0's for all the previous lines of the file, I believe a specially formatted lines will bump the memory requirement up by that some 5 megs per line...