oschulz · Oct 12, 2026 · Oct 12, 2026 · Oct 18, 2026 · Oct 20, 2026 · Oct 20, 2026
diff --git a/Project.toml b/Project.toml
@@ -1,6 +1,6 @@
 name = "LegendDataManagement"
 uuid = "9feedd95-f0e0-423f-a8dc-de0970eae6b3"
-version = "0.6.26"
+version = "0.6.27"
 
 [deps]
 Dates = "ade2ca70-3891-5945-98fb-dc099432e06a"
@@ -55,7 +55,7 @@ IntervalSets = "0.6, 0.7"
 JSON = "1"
 LRUCache = "1.5"
 LegendDataTypes = "0.1.17"
-LegendHDF5IO = "0.1.14"
+LegendHDF5IO = "0.2"
 LinearAlgebra = "<0.0.1, 1"
 MIMEs = "0.1, 1"
 Markdown = "<0.0.1, 1"

diff --git a/ext/LegendDataManagementLegendHDF5IOExt.jl b/ext/LegendDataManagementLegendHDF5IOExt.jl
@@ -40,17 +40,6 @@ const RSelLike = Union{AbstractDataSelectorLike, FileKey, Unitful.Time, Abstract
 const PossibleDataSelectors = [DataTier, DataCategory, DataPeriod, DataRun, DataPartition, DetectorId]
 
 
-const dataselector_bytypes = Dict{Type, String}()
-
-LegendHDF5IO.datatype_to_string(::Type{<:T}) where {T <: LegendDataManagement.DataSelector} = 
-    dataselector_bytypes[T]
-
-function LegendHDF5IO._array_type(::Type{Array{T, N}}
-    ) where {T <: LegendDataManagement.DataSelector, N}
-
-    AbstractArray{T, N}
-end
-
 # write LegendDataManagement.DataSelector
 function LegendHDF5IO.create_entry(parent::LHDataStore, name::AbstractString, 
     data::T; kwargs...) where {T <:LegendDataManagement.DataSelector}
@@ -146,23 +135,14 @@ function _init_workers(wpool::WorkerPool)
 end
 
 function __init__()
-    function extend_datatype_dict(::Type{T}, key::String
-        ) where {T <: LegendDataManagement.DataSelector}
-
-        LegendHDF5IO._datatype_dict[key] = T
-        dataselector_bytypes[T] = key
+    for (name, T) in (
+        "expsetup" => ExpSetup, "datatier" => DataTier, "datarun" => DataRun,
+        "dataperiod" => DataPeriod, "datacategory" => DataCategory, "timestamp" => Timestamp,
+        "filekey" => FileKey, "channelid" => ChannelId, "detectorid" => DetectorId,
+        "datapartition" => DataPartition,
+    )
+        LegendHDF5IO.register_datatype!(name, T)
     end
-
-    (@isdefined ExpSetup) && extend_datatype_dict(ExpSetup, "expsetup")
-    (@isdefined DataTier) && extend_datatype_dict(DataTier, "datatier")
-    (@isdefined DataRun) && extend_datatype_dict(DataRun, "datarun")
-    (@isdefined DataPeriod) && extend_datatype_dict(DataPeriod, "dataperiod")
-    (@isdefined DataCategory) && extend_datatype_dict(DataCategory, "datacategory")
-    (@isdefined Timestamp) && extend_datatype_dict(Timestamp, "timestamp")
-    (@isdefined FileKey) && extend_datatype_dict(FileKey, "filekey")
-    (@isdefined ChannelId) && extend_datatype_dict(ChannelId, "channelid")
-    (@isdefined DetectorId) && extend_datatype_dict(DetectorId, "detectorid")
-    (@isdefined DataPartition) && extend_datatype_dict(DataPartition, "datapartition")
 end
 
 # Paths are resolved and stat'ed one at a time: file metadata lookups are expensive on
@@ -221,29 +201,12 @@ function _nest(paths, vals)
     end)
 end
 
-# Every column of a group keeps the rows `sel` selects; a single column is unwrapped.
-# TODO: pass a mask to the read once `LH5Array` supports `AbstractVector{Bool}` indices.
+# Every column of a group keeps the rows `sel` selects; a single column is unwrapped. Rows
+# named by index are read as LegendHDF5IO reads index vectors: in one bounding read, a single
+# selection or one read per run of rows, whichever costs least for their spacing.
 _load_all_keys(nt::NamedTuple, sel=(:)) = length(nt) == 1 ? _load_all_keys(nt[first(keys(nt))], sel) :
     NamedTuple{keys(nt)}(map(k -> _load_all_keys(nt[k], sel), keys(nt)))
-_load_all_keys(x::Union{AbstractArray, Table}, sel=(:)) = sel === (:) ? x[:] : x[:][sel]
-# Rows named by index are read range by range, as an `LH5Array` reads ranges. Every read of a
-# column costs tens of microseconds however few rows it takes, so rows of scalars are read
-# along with gaps of up to `_bridged_rows` between them rather than one range per row; rows
-# holding arrays -- waveforms -- cost their bytes and are read in ranges of adjacent rows.
-const _bridged_rows = 256
-function _load_all_keys(x::Union{AbstractArray, Table}, sel::AbstractVector{Int})
-    isempty(sel) && return x[1:0]
-    maxgap = isbitstype(eltype(x)) ? _bridged_rows : 0
-    rows = sort(unique(sel))
-    ranges = [rows[1]:rows[1]]
-    for r in rows[2:end]
-        r - last(ranges[end]) - 1 <= maxgap ? (ranges[end] = first(ranges[end]):r) : push!(ranges, r:r)
-    end
-    read = fast_flatten([x[rg] for rg in ranges])
-    offsets = cumsum(length.(ranges)) .- length.(ranges)
-    pos = map(i -> (k = searchsortedlast(first.(ranges), i); offsets[k] + i - first(ranges[k]) + 1), sel)
-    pos == 1:length(read) ? read : read[pos]
-end
+_load_all_keys(x::Union{AbstractArray, Table}, sel=(:)) = x[sel]
 _load_all_keys(x, sel=(:)) = x
 
 _nrows(nt::NamedTuple) = _nrows(first(nt))

diff --git a/src/legend_data.jl b/src/legend_data.jl
@@ -157,6 +157,7 @@ tier_data[tier::DataTierLike, category::DataCategoryLike, period::DataPeriodLike
 tier_data[tier::DataTierLike, category::DataCategoryLike, period::DataPeriodLike, run::DataRunLike, det::DetectorIdLike]
 
 tier_data[tier::DataTierLike, filekey::FileKeyLike]
+tier_data[tier::DataTierLike, filekeys::AbstractVector{<:FileKey}]
 tier_data[tier::DataTierLike, filekey::FileKey, ch::ChannelIdLike]
 tier_data[tier::DataTierLike, filekey::FileKey, det::DetectorIdLike]
 ```
@@ -252,6 +253,8 @@ function _getindex_impl(tier_data::LegendTierData, tier::DataTierLike, filekey::
     )
 end
 
+_getindex_impl(tier_data::LegendTierData, tier::DataTierLike, filekeys::AbstractVector{<:FileKey}) = [tier_data[tier, filekey] for filekey in filekeys]
+
 # Disambiguation between DataCategory and FileKey:
 function _getindex_impl(tier_data::LegendTierData, tier::DataTierLike, s::AbstractString)
     if _can_convert_to(DataCategory, s)

diff --git a/src/utils/data_utils.jl b/src/utils/data_utils.jl
@@ -199,10 +199,10 @@ function load_partition_ch(open_func::Function, flatten_func::Function, data::Le
 
     function get_key_evt_tbl(ds, period, run, k)
         if select_random
-            t = Table(ds[ch, tier, k])[:]
+            t = ds[ch, tier, k][:]
             t[evt_range[(period, run)][k]]
         else
-            Table(ds[ch, tier, k])[evt_range[(period, run)][k]]
+            ds[ch, tier, k][evt_range[(period, run)][k]]
         end
     end
     # load data with given range

diff --git a/test/Project.toml b/test/Project.toml
@@ -1,5 +1,6 @@
 [deps]
 Aqua = "4c88cf16-eb10-579e-8560-4a9242c79595"
+ArraysOfArrays = "65a8f2f4-9b39-5baf-92e2-a9cc46fdf018"
 Dates = "ade2ca70-3891-5945-98fb-dc099432e06a"
 Distributed = "8ba89e20-285c-5b6f-9357-94700520ee1b"
 Documenter = "e30172f5-a6a5-5a46-863b-614d45cd2de4"
@@ -12,6 +13,7 @@ ParallelProcessingTools = "8e8a01fc-6193-5ca1-a2f1-20776dae4199"
 Plots = "91a5bcdd-55d7-5caf-9e0b-520d859cae80"
 PropDicts = "4dc08600-4268-439e-8673-d706fafbb426"
 PropertyFunctions = "09e99361-2bb8-48a2-a80f-de58f0739eb4"
+RadiationDetectorSignals = "bf2c0563-65cf-5db2-a620-ceb7de82658c"
 SolidStateDetectors = "71e43887-2bd9-5f77-aebd-47f656f0a3f0"
 StructArrays = "09ab397b-f2b6-538f-b94a-2f83cf4a842a"
 Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
@@ -23,3 +25,4 @@ YAML = "ddb6d928-2868-570f-bddf-ab3f9cf99eb6"
 [compat]
 Documenter = "1"
 LegendTestData = "0.3.7"
+RadiationDetectorSignals = "0.3.12"
diff --git a/test/test_ext_legendhdf5io.jl b/test/test_ext_legendhdf5io.jl
@@ -5,7 +5,10 @@ using Test
 
 using LegendHDF5IO
 using LegendTestData
+using ArraysOfArrays
 using PropertyFunctions
+using RadiationDetectorSignals
+using StructArrays
 using TypedTables
 using Unitful
 
@@ -527,4 +530,70 @@ using Distributed
             end
         end
     end
+
+    @testset "_load_all_keys" begin
+        Ext = Base.get_extension(LegendDataManagement, :LegendDataManagementLegendHDF5IOExt)
+        mktempdir() do dir
+            fn = joinpath(dir, "load_all_keys.lh5")
+            n = 200
+            a = rand(Float32, n)
+            wf = VectorOfVectors([rand(Float32, 8) for _ in 1:n])
+            wfs = ArrayOfRDWaveforms((fill(range(0.0u"μs", 1.0u"μs"; length = 8), n), VectorOfSimilarVectors(rand(Float32, 8, n))))
+            lh5open(fn, "w") do h
+                h["tbl"] = StructArray((a = a, wf = wf))
+                h["nested"] = StructArray((t = a, sub = StructArray((a = a, wf = wf))))
+                h["tbl_rdwf"] = StructArray((a = a, wf = wfs))
+            end
+            lh5open(fn, "r") do h
+                tbl = h["tbl"]
+                @test tbl isa LegendHDF5IO.LH5LazyTable
+
+                # a lazily read table is loaded, not passed through as it is
+                full = Ext._load_all_keys(tbl)
+                @test full isa StructArray
+                @test full.a isa Vector{Float32}
+                @test full.a == a
+                @test full.wf == wf
+
+                rows = [3, 7, 7, 150, 2]
+                byrow = Ext._load_all_keys(tbl, rows)
+                @test byrow.a == a[rows]
+                @test byrow.wf == wf[rows]
+
+                # tables concatenated lazily load like the tables they are made of
+                twice = vcat(tbl, tbl)
+                @test Ext._load_all_keys(twice, [3, n + 7, 2]).a == a[[3, 7, 2]]
+                @test Ext._load_all_keys(twice).wf == vcat(wf, wf)
+
+                mask = a .> 0.5f0
+                bymask = Ext._load_all_keys(tbl, mask)
+                @test bymask.a == a[mask]
+                @test bymask.wf == wf[mask]
+
+                col = Ext._load_all_keys(tbl.a, rows)
+                @test col isa Vector{Float32}
+                @test col == a[rows]
+
+                # rows of a table with waveform columns are usable as rows
+                tbl_rdwf = h["tbl_rdwf"]
+                rows_wf = Ext._load_all_keys(tbl_rdwf, rows)
+                @test rows_wf.a == a[rows]
+                @test rows_wf[1].wf == wfs[rows[1]]
+                @test Ext._load_all_keys(tbl_rdwf)[1].wf == wfs[1]
+                @test Ext._load_all_keys(tbl_rdwf, mask)[1].wf == wfs[findfirst(mask)]
+
+                # a table nested in a table is loaded along with it
+                nested = Ext._load_all_keys(h["nested"])
+                @test nested.sub.a isa Vector{Float32}
+                @test nested.sub.wf == wf
+
+                nt = Ext._load_all_keys((x = tbl.a, y = tbl.wf))
+                @test nt.x == a
+                @test nt.y == wf
+                @test Ext._load_all_keys((x = tbl.a,)) == a
+
+                @test Ext._load_all_keys("not an array") == "not an array"
+            end
+        end
+    end
 end
diff --git a/test/test_legend_data.jl b/test/test_legend_data.jl
@@ -18,6 +18,9 @@ include("testing_utils.jl")
 
     @test getproperty(l200, :tier) isa LegendDataManagement.LegendTierData
     @test normalize_path(@inferred(l200.tier[:raw, filekey])) == "/some/other/storage/raw_lh5/cal/p02/r006/l200-p02-r006-cal-20221226T200846Z-tier_raw.lh5"
+    other_filekey = FileKey("l200-p02-r006-cal-20221226T210000Z")
+    @test @inferred(l200.tier[:raw, [filekey, other_filekey]]) isa Vector{String}
+    @test l200.tier[:raw, [filekey, other_filekey]] == [l200.tier[:raw, filekey], l200.tier[:raw, other_filekey]]
     @test normalize_path(@inferred(l200.tier[:raw, "l200-p02-r006-cal-20221226T200846Z"])) == "/some/other/storage/raw_lh5/cal/p02/r006/l200-p02-r006-cal-20221226T200846Z-tier_raw.lh5"
     @test normalize_path(@inferred(l200.tier[:dsp, "l200-p02-r006-cal-20221226T200846Z"])) == normalize_path(joinpath(testdata_dir, "generated", "tier", "dsp", "cal", "p02", "r006", "l200-p02-r006-cal-20221226T200846Z-tier_dsp.lh5"))