improved tests (still not cross-platform reproducible, though)

test/deploy_and_test.sh

+#!/usr/bin/env bash
+
+# Standalone script
+
+# on Maestro:
+#   SLURM_OPTS='-p dbc_pmo -q dbc -c 8' JULIA_THREADS=8 ./deploy_and_test.sh
+
+if [ -z "$(which julia)" ]; then
+  echo "\`julia\` not found"
+  exit 1
+fi
+
+if [ -n "$1" ] && [ "$1" = "-f" ] || [ "$1" = "--force" ]; then
+  rm -rf LarvaTagger
+elif [ -d LarvaTagger ]; then
+  echo "\`$0\` needs to make a local LarvaTagger directory"
+  echo "but this directory already exists."
+  echo "Run \`$0 --force\` or pick another location."
+  exit 1
+fi
+
+if [ -n "$LAUNCHER" ]; then
+  launcher="$LAUNCHER "
+elif [ -n "$(which srun)" ]; then
+  launcher="srun "
+  echo "Slurm detected"
+  if [ -z "$SLURM_OPTS" ]; then
+    echo "Options can be passed to \`srun\` with variable SLURM_OPTS"
+  else
+    launcher="$launcher$SLURM_OPTS "
+  fi
+fi
+
+CURDIR=$(pwd)
+
+mkdir -p LarvaTagger && cd $_
+
+# shunit2 is for tests specifically
+if [ -z "$(which shunit2)" ]; then
+  git clone -c advice.detachedHead=false -b v2.1.8 https://github.com/kward/shunit2
+fi
+
+# comment out the below line to build a development environment
+CLONE_OPTS=" --depth 1 --no-tags --single-branch -b dev"
+git clone$CLONE_OPTS https://gitlab.pasteur.fr/nyx/planarlarvae.jl PlanarLarvae
+git clone$CLONE_OPTS https://gitlab.pasteur.fr/nyx/LarvaTagger.jl
+ln -s $(realpath LarvaTagger.jl) LarvaTagger
+git clone$CLONE_OPTS https://gitlab.pasteur.fr/nyx/TaggingBackends
+git clone$CLONE_OPTS https://gitlab.pasteur.fr/nyx/MaggotUBA-adapter
+ln -s $(realpath MaggotUBA-adapter) MaggotUBA
+
+(cd TaggingBackends && julia --project=. -e 'using Pkg; Pkg.develop(path="../PlanarLarvae"); Pkg.instantiate()')
+
+(cd MaggotUBA && poetry install -v && poetry remove taggingbackends && poetry add ../TaggingBackends)
+
+cd LarvaTagger || exit 1
+julia --project=. -e 'using Pkg; Pkg.develop(path="../PlanarLarvae"); Pkg.instantiate()'
+
+# the remainder of this script is test-specific
+
+if [ -f ../../LarvaTagger_test_data.tgz ]; then
+  # To generate the test data, first run the tests with environment variable KEEP_MODEL_FILES=1
+  # copy the LarvaTagger/MaggotUBA/models/test_train_* directories into LarvaTagger/LarvaTagger.jl/test/data
+  # and build the archive from the LarvaTagger/LarvaTagger.jl directory.
+  # For example:
+  #   KEEP_MODEL_FILES=1 ./deploy_and_test.sh
+  #   cd LarvaTagger/LarvaTagger.jl
+  #   rm -rf test/data/test_train_*
+  #   cp -ra ../MaggotUBA/models/test_train_* test/data/
+  #   tar zcvf LarvaTagger_test_data.tgz test/data/*
+  #   mv LarvaTagger_test_data.tgz ../../
+  tar zxvf ../../LarvaTagger_test_data.tgz
+else
+  # Not recommended; reproducibility is not guarantee across hosts or architectures yet
+  wget -O- https://dl.pasteur.fr/fop/l3qruSQR/LarvaTagger_test_data.tgz | tar zxv
+fi
+
+if [ "$LOCAL_SCENARII" = "1" ]; then
+  if [ -f "$CURDIR/scenarii.sh" ]; then cp "$CURDIR/scenarii.sh" test/
+  elif [ -f "$CURDIR/test/scenarii.sh" ]; then cp "$CURDIR/test/scenarii.sh" test/
+  else echo "Cannot find a local scenarii.sh file"
+  fi
+fi
+
+if [ -z "$(which shunit2)" ]; then
+  ln -s "$CURDIR/LarvaTagger/shunit2/shunit2" test/
+fi
+
+echo "${launcher}test/scenarii.sh"
+${launcher}test/scenarii.sh

test/predict_and_retrain.sh

-#!/bin/bash
+#!/usr/bin/env bash
+
+# this script performs a few tests with the Docker image
+# TODO: implement more use cases as in scenarii.sh
 
 trap 'rm -rf "$TMPDIR"' EXIT
 

test/scenarii.sh

-#!/bin/bash
+#!/usr/bin/env bash
 
-if [ -z "$(which shunit2)" ]; then
-  echo "shunit2 is required to run the tests"
-  exit 1
-fi
+# This script performs tests for the scripts/larvatagger-toolkit.jl command.
+# See also the deploy_and_test.sh standalone script.
+
+project_root=$(dirname "$(dirname "$(realpath "${BASH_SOURCE[0]}")")")
 
-if [ -z "$(which h5diff)" ]; then
-  echo "h5diff (hdf5-tools) is required to run the tests"
+if [ -f "$project_root/test/shunit2" ]; then
+  SHUNIT2=$(realpath "$project_root/test/shunit2")
+elif [ -z "$(which shunit2)" ]; then
+  echo "shUnit2 is required to run the tests (see https://github.com/kward/shunit2)"
   exit 1
+else
+  SHUNIT2=shunit2
 fi
 
-project_root=$(dirname "$(dirname "$(realpath "${BASH_SOURCE[0]}")")")
+# Since larva_dataset.hdf5 files cannot be compared because of absolute paths
+# we no longer need `h5diff`
+
+# if [ -z "$(which h5diff)" ]; then
+#   echo "h5diff (hdf5-tools) is required to run the tests"
+#   exit 1
+# fi
 
 if [ -z "$JULIA_PROJECT" ]; then
   export JULIA_PROJECT=$(realpath "$project_root/../TaggingBackends")
@@ -28,8 +38,6 @@ fi
 
 larvataggerjl="scripts/larvatagger-toolkit.jl"
 
-# test data temporarily available at https://dl.pasteur.fr/fop/l3qruSQR/LarvaTagger_test_data.tgz
-
 datadir="test/data"
 datapath="$project_root/$datadir"
 
@@ -59,22 +67,26 @@ prepareTrainingData() {
 }
 
 testImportLabelFile() {
+  filename=result_imported.label
   # run
   cd "$project_root"
-  echo "\"$larvataggerjl\" import \"$datadir/partial_predictions.label\" imported.label --default-label=\"not back-up\""
-  "$larvataggerjl" import "$datadir/partial_predictions.label" result_imported.label --default-label="not back-up"
+  echo "\"$larvataggerjl\" import \"$datadir/partial_predictions.label\" $filename --default-label=\"not back-up\""
+  "$larvataggerjl" import "$datadir/partial_predictions.label" $filename --default-label="not back-up"
   # test
-  assertNull '\`import\` failed to reproduce the imported.label file' "$(cmp "$datapath/imported.label" "$datapath/result_imported.label")"
+  assertTrue '\`import\` failed to reproduce the imported.label file' '$(cmp "$datapath/imported.label" "$datapath/$filename")'
   # clean up
-  rm -f "$datapath/result_imported.label"
+  rm -f "$datapath/$filename"
 }
 
 testPredictDefault() {
   tagger="test_train_default$tagger_suffix"
   tmpdir=$(prepareTestData $tagger)
   # restore the model
+  mkdir -p "$maggotuba/models"
   rm -rf "$maggotuba/models/$tagger"
+  echo "cp -ra \"$datapath/$tagger\" \"$maggotuba/models/\""
   cp -ra "$datapath/$tagger" "$maggotuba/models/"
+  [ -f "$maggotuba/models/$tagger/clf_config.json" ] || exit 1
   # run
   cd "$project_root"
   echo "\"$larvataggerjl\" predict \"$maggotuba\" $tagger \"$tmpdir/sample.spine\""
@@ -84,7 +96,7 @@ testPredictDefault() {
   predictions="$tmpdir/$filename"
   expected_labels="$datapath/$tagger/$filename"
   assertFalse "\`predict\` failed to generate file $filename" '[ -z "$predictions" ]'
-  assertNull "\`predict\` failed to reproduce file $filename" "$(cmp "$expected_labels" "$predictions")"
+  assertTrue "\`predict\` failed to reproduce file $filename" '$(cmp "$expected_labels" "$predictions")'
 }
 
 testTrainDefault() {
@@ -127,12 +139,13 @@ postTrain() {
   interim_dir="$maggotuba/data/interim/$tagger"
 
   # interim files
-  fn=larva_dataset.hdf5
+  filename=larva_dataset.hdf5
   larva_dataset="$interim_dir/$(cd "$interim_dir"; ls larva_dataset_*.hdf5)"
-  assertFalse "\`train\` failed to generate file $fn" '[ -z "$larva_dataset" ]'
-  assertNull "\`train\` failed to reproduce file $fn" "$(h5diff "$expected_tagger_dir/$fn" "$larva_dataset")"
+  assertFalse "\`train\` failed to generate file $filename" '[ -z "$larva_dataset" ]'
+  # fails because of absolute paths
+  #assertTrue "\`train\` failed to reproduce file $filename" '$(h5diff "$expected_tagger_dir/$filename" "$larva_dataset")'
   if [ "$KEEP_MODEL_FILES" = "1" ]; then
-    cp "$larva_dataset" "$tagger_dir/$fn"
+    cp "$larva_dataset" "$tagger_dir/$filename"
   fi
 
   # test model files
@@ -148,17 +161,17 @@ postTrain() {
   assertFalse "\`train\` failed to generate file $enc_config" '[ -z "$enc_config_path" ]'
   assertFalse "\`train\` failed to generate file $clf_weights" '[ -z "$clf_weights_path" ]'
   assertFalse "\`train\` failed to generate file $clf_config" '[ -z "$clf_config_path" ]'
-  assertNull "\`train\` failed to reproduce file $enc_weights" "$(cmp "$expected_tagger_dir/$enc_weights" "$enc_weights_path")"
-  assertNull "\`train\` failed to reproduce file $enc_config" "$(cmp "$expected_tagger_dir/$enc_config" "$enc_config_path")"
-  assertNull "\`train\` failed to reproduce file $clf_weights" "$(cmp "$expected_tagger_dir/$clf_weights" "$clf_weights_path")"
-  assertNull "\`train\` failed to reproduce file $clf_config" "$(cmp "$expected_tagger_dir/$clf_config" "$clf_config_path")"
+  assertTrue "\`train\` failed to reproduce file $enc_weights" '$(cmp "$expected_tagger_dir/$enc_weights" "$enc_weights_path")'
+  assertTrue "\`train\` failed to reproduce file $enc_config" '$(cmp "$expected_tagger_dir/$enc_config" "$enc_config_path")'
+  assertTrue "\`train\` failed to reproduce file $clf_weights" '$(cmp "$expected_tagger_dir/$clf_weights" "$clf_weights_path")'
+  assertTrue "\`train\` failed to reproduce file $clf_config" '$(cmp "$expected_tagger_dir/$clf_config" "$clf_config_path")'
 
   # test the predictions
   echo "\"$larvataggerjl\" predict \"$maggotuba\" $tagger \"$tmpdir/sample.spine\""
   "$larvataggerjl" predict "$maggotuba" $tagger "$tmpdir/sample.spine"
   predictions="$tmpdir/predicted.label"
   expected_labels="$datapath/$tagger/predicted.label"
-  assertNull "\`predict\` failed to reproduce file predicted.label" "$(cmp "$expected_labels" "$predictions")"
+  assertTrue "\`predict\` failed to reproduce file predicted.label" '$(cmp "$expected_labels" "$predictions")'
 
   # clean up
   if [ "$KEEP_MODEL_FILES" = "1" ]; then
@@ -169,4 +182,4 @@ postTrain() {
   fi
 }
 
-. shunit2
+. $SHUNIT2