From fc1eebae2dd0bfecbdf2fd0a499a34e253386638 Mon Sep 17 00:00:00 2001
From: LTLA <infinite.monkeys.with.keyboards@gmail.com>
Date: Sun, 26 Nov 2023 23:50:52 -0800
Subject: [PATCH] More coordinated optimization of storage layouts for HDF5.

---
 .gitignore                                             |  2 ++
 DESCRIPTION                                            |  1 +
 R/{optimizeStorage.R => optimize_storage.R}            |  7 ++++++-
 .../collect_float_attributes.cpp                       | 10 +++++-----
 4 files changed, 14 insertions(+), 6 deletions(-)
 rename R/{optimizeStorage.R => optimize_storage.R} (96%)
 rename collect_float_attributes.cpp => src/collect_float_attributes.cpp (92%)

diff --git a/.gitignore b/.gitignore
index 1c56872..d291ee9 100644
--- a/.gitignore
+++ b/.gitignore
@@ -1,3 +1,5 @@
 *.swp
 *.rds
 *.html
+*.so
+*.o
diff --git a/DESCRIPTION b/DESCRIPTION
index 67d4b5a..ac7bd8a 100644
--- a/DESCRIPTION
+++ b/DESCRIPTION
@@ -24,6 +24,7 @@ Suggests:
     knitr,
     BiocStyle,
     chihaya
+LinkingTo: Rcpp
 VignetteBuilder: knitr
 RoxygenNote: 7.2.3
 biocViews:
diff --git a/R/optimizeStorage.R b/R/optimize_storage.R
similarity index 96%
rename from R/optimizeStorage.R
rename to R/optimize_storage.R
index 713121b..1c63f87 100644
--- a/R/optimizeStorage.R
+++ b/R/optimize_storage.R
@@ -19,6 +19,7 @@ setMethod("collect_integer_attributes", "array", function(x) {
 })
 
 setMethod("collect_integer_attributes", "ANY", function(x) {
+
     collated <- blockApply(x, collect_integer_attributes)
     list(
         range=aggregate_range(collated, "range"),
@@ -133,9 +134,13 @@ optimize_float_storage <- function(x) {
         } else if (!attr$has_max) {
             placeholder <- max_double()
         }
+
+        # Fallback that just goes through and pulls out all unique values.
         if (is.null(placeholder)) {
-            warning("cannot guess a suitable missing value placeholder, treating all NAs as NaNs")
+            u <- Reduce(union, blockApply(x, function(y) unique(as.vector(y))))
+            placeholder <- chooseMissingPlaceholderForHdf5(u)
         }
+
         return(list(type="H5T_NATIVE_DOUBLE", placeholder=placeholder))
 
     } else {
diff --git a/collect_float_attributes.cpp b/src/collect_float_attributes.cpp
similarity index 92%
rename from collect_float_attributes.cpp
rename to src/collect_float_attributes.cpp
index 82a4e9a..9023ac1 100644
--- a/collect_float_attributes.cpp
+++ b/src/collect_float_attributes.cpp
@@ -62,8 +62,8 @@ Rcpp::List collect_double_attributes(Rcpp::NumericVector x) {
         if (!non_integer) {
             for (auto y : x) {
                 if (!ISNA(y)) {
-                    minv = std::min(y, v);
-                    maxv = std::max(y, v);
+                    minv = std::min(y, minv);
+                    maxv = std::max(y, maxv);
                 }
             }
         }
@@ -71,7 +71,7 @@ Rcpp::List collect_double_attributes(Rcpp::NumericVector x) {
 
     bool has_lowest = false, has_highest = false;
     {
-        constexpr double lowest = lowest_double();
+        double lowest = lowest_double();
         for (auto y : x) {
             if (!ISNA(y) && y == lowest) {
                 has_lowest = true;
@@ -79,7 +79,7 @@ Rcpp::List collect_double_attributes(Rcpp::NumericVector x) {
             }
         }
 
-        constexpr double highest = max_double();
+        double highest = max_double();
         for (auto y : x) {
             if (!ISNA(y) && y == highest) {
                 has_highest = true;
@@ -88,7 +88,7 @@ Rcpp::List collect_double_attributes(Rcpp::NumericVector x) {
         }
     }
 
-    return Rcpp::List(
+    return Rcpp::List::create(
         Rcpp::Named("range") = Rcpp::NumericVector::create(minv, maxv),
         Rcpp::Named("missing") = Rcpp::LogicalVector::create(has_missing),
         Rcpp::Named("non_integer") = Rcpp::LogicalVector::create(non_integer),