I've made a few more updates after seeing the test failures. The XPT reader code has been updated to reflect the changes to the XPT writer.

I had to make a couple of changes to the tests for what looked like false positives, and I think they make sense:

• One of the SAS format roundtrip tests has been changed to use $CHAR3 instead of $CHAR3.. The trailing period is dropped when roundtripping XPT files because of the way the format is reconstructed so $CHAR3. is read back in as $CHAR3 and fails the test. It represents the same format so it made to sense to me to update the test.
• It now checks if label_set has been set before running the roundtrip test (as was already the case for format and display_width). It was throwing some strange errors in the display_width tests that looked like it was checking unallocated memory like:

Test "Display width" failed: Column label sets
 * Format: xpt5 (0x200000)
 * Expected: "¤�~y²U" (length=6)
 * Received: "�~y²U" (length=5)
 * Column: 1

I've updated the format parsing code to use a Ragel parser and added an xport_format_t struct to store the parser output, and added a few more tests to check it's throwing errors correctly.

I've also had a go at the fuzzer setup but a bit less sure of what I'm doing there!

It looks like Appveyor is failing:

https://ci.appveyor.com/project/evanmiller/readstat/builds/41638815

You may need to modify the Visual Studio project file to ensure all the new files are being built. See

https://github.com/WizardMac/ReadStat/blob/master/VS17/ReadStat.vcxproj

Awesome, thanks 🙂

I've made the appropriate fixes to the VS17 project file, appveyor is looking happier now.

Thanks, much better now!

@evanmiller - any idea when these changes might be able to be merged into a new release of ReadStat? These changes would open up some important doors for haven, which will have a significant impact in some pharmaceutical industry projects.