Added support for read_xport

[cipherz]

• C_interface.jl:
  Changed return type for readstat_get_row_count to Int (from UInt), as it seems that not all format contains nrows as metainformation and -1 is returned in those instances (xport is one of such formats).
  Added readstat_parse for type::Val{:xport}, calls readstat_parse_xport.

• ReadStat.jl:
  Added constant READSTAT_NALLOC_ROWS: number of rows to allocate at the same time (see handle_variable! and handle_value!)
  Added read_xport and exported it.
  Modified read_data_file, handle_variable! and handle_value! to support formats without nrows metainformation.

• Test
  Added test cases for xpt and types.xpt.

[davidanthoff]

Cool, thanks!

I was not aware that readstat_get_row_count can return -1, that seems important to handle, thanks for spotting that! I'm wondering, though, whether we should handle that case differently. I think in that case I would just allocate vectors with 0 elements, and then push! into those. I think the algorithm in push! is better at growing arrays than using a fixed increment like READSTAT_NALLOC_ROWS.

I'll try to fix the CI failures on master, and then you can merge master back into this branch and we'll hopefully get CI tests passing.

[davidanthoff]

Alrigh, #51 fixed the CI builds on master. So if you merge master into your branch, you should have a baseline that passes CI builds.

[cipherz]

I did not realize that push! would grow as needed, just started out with Julia recently. I am also wondering if it is worth splitting up the code a bit more and formats with nrows in the header, like the ones originally implemented, would retain the original implementation, the fastest, and the auto-growing (via push!) would only be used for formats without row count information. Being a bit of a novice, I am tempted to take the approach to making it type based used in other parts of the code, but I don't know if there is a better way. I will merge the new master into my branch and make some proposed changes to address the points above.

…

On Sun, Aug 25, 2019 at 1:24 AM David Anthoff ***@***.***> wrote: Alrigh, #51 <#51> fixed the CI builds on master. So if you merge master into your branch, you should have a baseline that passes CI builds. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#50?email_source=notifications&email_token=AAOXKUPHVEK6JGYV6POGS6LQGIJP7A5CNFSM4IMSSSF2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD5CMSEA#issuecomment-524601616>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAOXKUJJIN6Z5BGSHDDPGETQGIJP7ANCNFSM4IMSSSFQ> .

[davidanthoff]

Yes, I think having a different code path for the case with known size and the case without would be good, we don't want to use push! if we know the size beforehand!

[vjd]

Just following up here to see if there are plans to support xpt files?

[davidanthoff]

I believe this is fully covered by #75, so closing this here.