Feature tracker

[kbarbary]

This is an issue to track some obvious features that should be added. My guess is that they will be added gradually over time as needed, but in a backwards-compatible manner.

General

• Iterator protocol for FITS type: enabling for hdu in FITS(...)
• Improve appearance of show() for FITS and HDU types. [Improve show(); reorganize files #38]

Headers

• Read a single keyword value without the comment, like read_key(hdu, key)[1].

Images

• Write to a subset of the image: write(hdu, data; start=(100, 200))
• Read to already-allocated array: read!(hdu, data), read!(hdu, 1:20, 1:30, data)
• Read/write tile-compressed images.

Tables

• Read and write variable length table columns [Read/write variable-length table columns #36]
• Method to get column names in table (DataFrames extends Base.names but that has a different definition; maybe use colnames) [Method to get column names in a FITS table #93]
• Method to get number of rows in table (extend Base.length)
• Relax write method to write(::FITS, ::Dict{Any, Vector}) (raise runtime error if any dictionary key is not ASCII)
• Read ranges of table rows: read(hdu, "COLNAME", 1:10)
• Read arbitrary table rows: read(hdu, "COLNAME", [1, 3, 6, 10])
• Read to already-allocated arrays: read!(hdu, "COLNAME", data)
• Append rows to a table
• Insert columns into a table
• Read/write bit columns (into BitArrays?)
• Iterate over table rows?
• Read multiple columns simultaneously (requires low-level tinkering, byteswapping)

[jishnub]

read! has been implemented now that #122 is merged

[mileslucas]

Possible feature (I'm not sure if functionality exists in cfitsio)
write!(hdu, data)
write!(fits, i, data)
to overwrite the data in a given HDU. Currently the only option is to append a new HDU.

[jishnub]

Currently we check the strides of the array to determine if the memory layout is contiguous. This is too conservative, but I don't know of a workaround right now other than using a Union of compatible types. For example, this means that currently writing out something like a Transpose is not supported even if the memory layout of its parent is contiguous and cfitsio supports this.

Keeping julia#10889 in mind, we may eventually use a trait-based check for contiguity instead of a strides-based approach as in the case now.