Files with inconsistent image slice sizes break loading

[tlnagy]

Currently TIFF images with embedded reduced resolution images cause problems since TiffImages assumes the images aren't ragged.

Take DAPI Myog Pax7_CH1.zip for example, which fails to load:

julia> TiffImages.load("DAPI Myog Pax7_CH1.tif")
ERROR: EOFError: read end of file
Stacktrace:
  [1] unsafe_read(s::IOStream, p::Ptr{UInt8}, nb::UInt64)
    @ Base ./iostream.jl:428
  [2] unsafe_read
    @ ./io.jl:761 [inlined]
  [3] read!
    @ ./io.jl:779 [inlined]
  [4] read!(file::TiffFile{UInt32, FileIO.Stream{FileIO.DataFormat{:TIFF}, IOStream, String}}, arr::Matrix{ColorTypes.RGB{FixedPointNumbers.N0f8}})

However, loading the image planes separately works:

tf = read(open("DAPI Myog Pax7_CH1.tif"), TiffFile)
ifds = collect(tf)
TiffImages.load!.(Ref(tf), ifds)
size(TiffImages.load(tf, ifds[1:1], nothing))

(720, 960)

and

size(TiffImages.load(tf, ifds[2:2], nothing))

(120, 160).

It appears as though we are handling everything just fine, it's just that we should detect when images are "ragged", i.e. don't have the same XY sizes and load those images as the currently unimplemented StridedTaggedImage

[chrstphrbrns]

I've come across GeoTIFFs with ragged data, so this is a must for 1.0 IMHO

[tlnagy]

Agreed, this would be nice to have in 1.0. AFAIK the TIFF spec doesn't record raggedness anywhere. There's some information in SubfileType and NewSubfileType tags, but it doesn't seem universal.

I think the best strategy is to assume consistent slice sizes and fail gracefully if we detect ragged images and build a StridedTaggedImage and let the user (or wrapping package e.g. OMETIFF.jl) deal with it.

[chrstphrbrns]

Thoughts on how to handle LazyBufferedTIFF, which assumes slices are homogeneous in size and color type?

[tlnagy]

LazyBufferedTIFF is currently a subtype of AbstractDenseTIFF so it should just error and not support ragged TIFFs. We could discuss whether that subtyping is the best, but just erroring could be fine for now.

EDIT: Discussion over in #154