Experiments with redo

Makefile

-include ../../config

define clean-train
	cd $(<D) ; $(MOSES_SCRIPTS)/training/clean-corpus-n.perl $(notdir $(basename $<)) en mrl moses-corpus 1 100
	mv $(basename $@).en  $(basename $@).clean-en
	mv $(basename $@).mrl $(basename $@).clean-mrl
endef

ifdef DISABLE_MOSES_TUNING
%/moses-corpus.clean-en  : %/trainNdev-full-corpus.en %/trainNdev-full-corpus.mrl
	$(call clean-train)
else
%/moses-corpus.clean-en  : %/train-full-corpus.en %/train-full-corpus.mrl
	$(call clean-train)
endif

%/moses-corpus.clean-mrl : | %/moses-corpus.clean-en
	# moses-corpus.mrl and moses-corpus.en are generated together

N1-AFTER-a.sh

# common/moses-corpus.clean-en.do

. ../../../../lib

if [ $DISABLE_MOSES_TUNING -eq 0 ]; then
  # enable
  CORPUS=train-full-corpus
else
  CORPUS=trainNdev-full-corpus
fi
redo-ifchange $CORPUS.en $CORPUS.mrl

BNAME=`basename $1 .clean-en`
$MOSES_SCRIPTS/training/clean-corpus-n.perl $CORPUS en mrl moses-corpus 1 100
mv $BNAME.en $3

N1-AFTER-b.sh

# common/moses-corpus.clean-mrl.do
BNAME=`basename $1 .clean-mrl`
redo-ifchange $BNAME.clean-en
mv $BNAME.mrl $3

N2-COMMENTS.md

PRO

I think the AFTER is cleaner (but is that a result of hindsight or actually changing tools?).

• Only one syntax to deal with at a time (your shell language, not that PLUS make)
• Can interleave dependency information in script in more general ways than using things like $@

CON

• It seems tricky to express idea that foo/x.y and bar/x.y are built the same way. This is one thing that the makefile does better than my redo script (the % stands for foo or bar). With redo, I'm using a common/x.y.do and having a script copy them from common to foo/ and bar/. :-/
• Possibly tricky to have common information (note the . ../../../../lib)

Surprises

• default.x.do and foo.x.do provide different conventions for how $1 and $2 behave. There's justification in the manual, but do people really read manuals? Oh well, we'll have to see how djbredo behaves