readdlm() cannot process empty file

[innerlee]

• julia ver: 0.4.5
• os ver: win 10

reproduce:

julia> writedlm("emp.ty", [])
julia> readdlm("emp.ty")
ERROR: at row 0, column 0 : ArgumentError("number of rows in dims must be > 0, got 0")
 in error at error.jl:21

Is this behavior by design?

[ararslan]

It appears to be by design given that there's an explicit error for it. What behavior would you have expected? A 0x0 array?

[ViralBShah]

Cc @tanmaykm

[nalimilan]

The problem is that readdlm can't know the number of columns of the result. So if in general your code works with two-columns arrays, a 0x0 array will likely raise an error later anyways. By throwing the error immediately, readdlm allows you to create a matrix of the correct size.

We could provide a way to specify the number of columns, and not raise an error if it's set. Currently AFAIK dims cannot be used for that.

[innerlee]

I think 0x0 array is okay, because

• I do not agree that a 0x0 array is useless, or equals an error.
• Columns can vary, like rows. I mean, one might read, append columns to the data, and rewrite the file.
• When accessing data, one checks the bounds, like in for i=1:length(sth). So, 0x0 will be safe in this case.
• Its easier to check size than catch exceptions.

[alexmorley]

0x0 array was definitely was definitely the behaviour that I expected, else what it returns in this case could be specified with a keyword (like :throwifempty or something?)

[bjarthur]

a 0x0 array is also what i expected

[vtjnash]

It seems like empty files might indicate errors, but no less so than arbitrarily truncated files (which we cannot detect). Do you want to make a PR to return a 0x0 Any array?