clarifies stream seeking for querying and detection functions

README.md

@@ -25,7 +25,7 @@ using FileIO
 obj = load(filename)
 ```
 to read data from a formatted file.  Likewise, saving might be as simple as
-```
+```jl
 save(filename, obj)
 ```
 
@@ -53,9 +53,11 @@ add_format(format"PNG", [0x89,0x50,0x4e,0x47,0x0d,0x0a,0x1a,0x0a], ".png")
 # have one of two possible file extensions
 add_format(format"NRRD", "NRRD", [".nrrd",".nhdr"])
 
-# A format whose magic bytes might not be at the beginning of the file,
-# necessitating a custom function `detecthdf5` to find them
-add_format(format"HDF5", detecthdf5, [".h5", ".hdf5"])
+# A format whose magic bytes more complicated, necessitating a custom function
+# `detectwav` to find them. The function should assume that the stream is
+# positioned at the beginning of the file being detected, and the query
+# infrastructure will handle seeking to the correct position afterwards.
+add_format(format"WAV", detectwav, ".wav")
 
 # A fictitious format that, unfortunately, provides no magic
 # bytes. Here we have to place our faith in the file extension.
@@ -103,7 +105,6 @@ using FileIO
 # See important note about scope below
 function load(f::File{format"PNG"})
     open(f) do s
-        skipmagic(s)  # skip over the magic bytes
         # You can just call the method below...
         ret = load(s)
         # ...or implement everything here instead
@@ -112,7 +113,7 @@ end
 
 # You can support streams and add keywords:
 function load(s::Stream{format"PNG"}; keywords...)
-    # s is already positioned after the magic bytes
+    skipmagic(s)  # skip over the magic bytes
     # Do the stuff to read a PNG file
     chunklength = read(s, UInt32)
     ...
@@ -130,8 +131,8 @@ end
 Note that these are `load` and `save`, **not** `FileIO.load` and `FileIO.save`.
 Because a given format might have multiple packages that are capable of reading it,
 FileIO will dispatch to these using module-scoping, e.g., `SomePkg.load(args...)`.
-Consequently, **packages should define "private" `load` and `save` methods, and 
-not extend (import) FileIO's**. 
+Consequently, **packages should define "private" `load` and `save` methods, and
+not extend (import) FileIO's**.
 
 `load(::File)` and `save(::File)` should close any streams
 they open.  (If you use the `do` syntax, this happens for you

src/query.jl

@@ -72,6 +72,7 @@ For example:
     add_format(format"PNG", (UInt8[0x4d,0x4d,0x00,0x2b], UInt8[0x49,0x49,0x2a,0x00]), [".tiff", ".tif"])
     add_format(format"PNG", [0x89,0x50,0x4e,0x47,0x0d,0x0a,0x1a,0x0a], ".png")
     add_format(format"NRRD", "NRRD", [".nrrd",".nhdr"])
+    add_format(format"WAV", detectwav, [".wav", ".WAV"])
 
 Note that extensions, magic numbers, and format-identifiers are case-sensitive.
 """

src/registry.jl

@@ -8,7 +8,6 @@ add_format(format"GZIP", [0x1f, 0x8b], ".gz", [:Libz])
 
 # test for RD?2 magic sequence at the beginning of R data input stream
 function detect_rdata(io)
-    seekstart(io)
     read(io, UInt8) == UInt8('R') &&
     read(io, UInt8) == UInt8('D') &&
     read(io, UInt8) in (UInt8('A'), UInt8('B'), UInt8('X')) &&
@@ -19,10 +18,8 @@ end
 add_format(format"RData", detect_rdata, [".rda", ".RData", ".rdata"], [:RData, LOAD])
 
 function detect_rdata_single(io)
-    seekstart(io)
     res = read(io, UInt8) in (UInt8('A'), UInt8('B'), UInt8('X')) &&
         (c = read(io, UInt8); c == UInt8('\n') || (c == UInt8('\r') && read(io, UInt8) == UInt8('\n')))
-    seekstart(io)
     return res
 end
 
@@ -145,10 +142,9 @@ add_format(format"GSLIB", (), [".gslib",".sgems"], [:GslibIO])
 
 ### Audio formats
 function detectwav(io)
-    seekstart(io)
     magic = read!(io, Vector{UInt8}(4))
     magic == b"RIFF" || return false
-    seek(io, 8)
+    skip(io, 4)
     submagic = read!(io, Vector{UInt8}(4))
 
     submagic == b"WAVE"
@@ -200,10 +196,9 @@ skipmagic(io, ::typeof(detect_noometiff)) = seek(io, 4)
 
 # AVI is a subtype of RIFF, as is WAV
 function detectavi(io)
-    seekstart(io)
     magic = read!(io, Vector{UInt8}(4))
     magic == b"RIFF" || return false
-    seek(io, 8)
+    skip(io, 4)
     submagic = read!(io, Vector{UInt8}(4))
 
     submagic == b"AVI "
@@ -212,6 +207,8 @@ add_format(format"AVI", detectavi, ".avi", [:ImageMagick])
 
 # HDF5: the complication is that the magic bytes may start at
 # 0, 512, 1024, 2048, or any multiple of 2 thereafter
+# this detection function assumes that the stream start and end match the
+# file start and end, which is true if it's just a file on disk
 h5magic = (0x89,0x48,0x44,0x46,0x0d,0x0a,0x1a,0x0a)
 function detecthdf5(io)
     position(io) == 0 || return false
@@ -234,6 +231,8 @@ function detecthdf5(io)
 end
 add_format(format"HDF5", detecthdf5, [".h5", ".hdf5"], [:HDF5])
 
+# the STL detection functions assumes that the stream start and end match the
+# file start and end, which is true if it's just a file on disk
 function detect_stlascii(io)
     try
         position(io) != 0 && (seekstart(io); return false)

test/query.jl

@@ -336,13 +336,32 @@ end
     q = query(joinpath(file_dir, "minimal_ascii.rds"))
     @test typeof(q) == File{format"RDataSingle"}
     open(q) do io
-        @test position(io) == 0
         @test FileIO.detect_rdata_single(io)
-        # need to seek to beginning of file where data structure starts
-        @test position(io)  == 0
     end
 end
 @testset "Format with function for magic bytes" begin
     add_format(format"FUNCTION_FOR_MAGIC_BYTES", x -> 0x00, ".wav", [:WAV])
     del_format(format"FUNCTION_FOR_MAGIC_BYTES")
 end
+
+function detect_position_test(io)
+    return read(io, 3) == b"DET"
+end
+
+@testset "Detection function called with properly-positioned stream" begin
+    add_format(format"DET", detect_position_test, ".det")
+    # we need extra junk to work around issue #176
+    junk = rand(UInt8, 35)
+    io = IOBuffer()
+    write(io, "DET")
+    write(io, junk)
+    seek(io, 0)
+    @test query(io) isa Formatted{format"DET"}
+    @test position(io) == 0
+    write(io, "junkDET")
+    write(io, junk)
+    seek(io, 4)
+    @test query(io) isa Formatted{format"DET"}
+    @test position(io) == 4
+    del_format(format"DET")
+end